From c4a049c034ea323a5e20ab02aa6ccc9a10e75813 Mon Sep 17 00:00:00 2001 From: Tyler Nadolski <122555266+nadolskit@users.noreply.github.com> Date: Fri, 27 Sep 2024 09:09:08 -0700 Subject: [PATCH] lint --- .../test_pdf_reader_match_doc_details.yaml | 11664 ++++++++-------- tests/test_paperqa.py | 3 +- 2 files changed, 5842 insertions(+), 5825 deletions(-) diff --git a/tests/cassettes/test_pdf_reader_match_doc_details.yaml b/tests/cassettes/test_pdf_reader_match_doc_details.yaml index 4fc1bcf4..0d2bf73f 100644 --- a/tests/cassettes/test_pdf_reader_match_doc_details.yaml +++ b/tests/cassettes/test_pdf_reader_match_doc_details.yaml @@ -1,5826 +1,5842 @@ interactions: -- request: - body: '{"messages": [{"role": "user", "content": "Extract the title, authors, - and doi as a JSON from this MLA citation. If any field can not be found, return - it as null. Use title, authors, and doi as keys, author''s value should be a - list of authors. Wellawatte et al, A Perspective on Explanations of Molecular - Prediction Models, XAI Review, 2023\n\nCitation JSON:"}], "model": "gpt-4o-2024-08-06", - "stream": false, "temperature": 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '429' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAAwAAAP//dFHBbtpAEL3zFaM5Q2UMhIRbmkZtD1HTSlUOcYSX9WA2Xe9sd4dAi/j3 - aG0H0kMvc3hv3tN7M4cBAJoKF4B6o0Q33o6ub+bfvjzQTn2f3Gx2Uv/+8ffrx9tPq2r/+ecLDpOC - V8+k5U31QXPjLYlh19E6kBJKruN5Pp/OxvPsqiUarsgmWe1lNOVRnuXTUXY5yi564YaNpogLeBwA - ABzamSK6iva4gGz4hjQUo6oJF6clAAxsE4IqRhNFOcHhmdTshFybuizL58iucIfCARQoRiwVuIAC - r+GeQvSkxbwQsIPbvbfKqdQuAq/hji3prVUB7gNVRicC7lKxWOCw81Nb2XCIyfGxwAeyVu2UCAEJ - KFvgU79XsUk7bmtt4Y6FK8vyfeJA621Utt/o8ePpBJZrH3gVe/6Er40zcbMMpCK7VDcKe2zZ4wDg - qT319p/roQ/ceFkK/yKXDC8nnR2ef3smJ1c9KSzKnvFxnv9PtaxIlLHx3b+wC2hcfXbITinbmhj/ - RKFmuTaupuCD6d639stZNsvyyYVeVTg4Dl4BAAD//wMAjAmQ4scCAAA= - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c999d5d1867b2-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:49 GMT - Server: - - cloudflare - Set-Cookie: - - __cf_bm=gK44ohFwDd5WKiBoRjoA0YGKDmimgFNiQ8ukJRMnG.A-1727451709-1.0.1.1-V.Zf3Hhy8YwzkEJok22089NmBFWmCEmiPmP.GiqkV157_lXG1woQd1UeERTqCaqseJgfOhGmnEFH1L3FUqTMNA; - path=/; expires=Fri, 27-Sep-24 16:11:49 GMT; domain=.api.openai.com; HttpOnly; - Secure; SameSite=None - - _cfuvid=DoEGhZd5DcgECkMyAWnjKWwxp_c7OpMX6A9fxXzv9sM-1727451709777-0.0.1.1-604800000; - path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '506' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29999904' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 0s - x-request-id: - - req_74f9b5e3a661e39a92ed20c8b465b356 - status: - code: 200 - message: OK -- request: - body: null - headers: {} - method: GET - uri: https://api.crossref.org/works?mailto=example@papercrow.ai&query.title=A+Perspective+on+Explanations+of+Molecular+Prediction+Models&rows=1&query.author=Wellawatte+et+al&select=DOI,author,container-title,title - response: - body: - string: '{"status":"ok","message-type":"work-list","message-version":"1.0.0","message":{"facets":{},"total-results":17288,"items":[{"DOI":"10.26434\/chemrxiv-2022-qfv02","author":[{"given":"Geemi - P.","family":"Wellawatte","sequence":"first","affiliation":[{"name":"University - of Rochester"}]},{"given":"Heta A.","family":"Gandhi","sequence":"additional","affiliation":[{"name":"University - of Rochester"}]},{"given":"Aditi","family":"Seshadri","sequence":"additional","affiliation":[{"name":"University - of Rochester"}]},{"ORCID":"http:\/\/orcid.org\/0000-0002-6647-3965","authenticated-orcid":false,"given":"Andrew - D.","family":"White","sequence":"additional","affiliation":[{"name":"University - of Rochester"}]}],"title":["A Perspective on Explanations of Molecular Prediction - Models"]}],"items-per-page":1,"query":{"start-index":0,"search-terms":null}}}' - headers: - Access-Control-Allow-Headers: - - X-Requested-With, Accept, Accept-Encoding, Accept-Charset, Accept-Language, - Accept-Ranges, Cache-Control - Access-Control-Allow-Origin: - - '*' - Access-Control-Expose-Headers: - - Link - Connection: - - close - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:51 GMT - Server: - - Jetty(9.4.40.v20210413) - Transfer-Encoding: - - chunked - permissions-policy: - - interest-cohort=() - x-api-pool: - - plus - x-rate-limit-interval: - - 1s - x-rate-limit-limit: - - '150' - x-ratelimit-interval: - - 1s - x-ratelimit-limit: - - '150' - status: - code: 200 - message: OK -- request: - body: '{"input": ["A Perspective on Explanations of Molecular\nPrediction Models\nGeemi - P. Wellawatte,\u2020\nHeta A. Gandhi,\u2021\nAditi Seshadri,\u2021 and\nAndrew\nD. - White\u2217,\u2021\n\u2020Department of Chemistry, University of Rochester, - Rochester, NY, 14627\n\u2021Department of Chemical Engineering, University of - Rochester, Rochester, NY, 14627\n\u00b6Vial Health Technology, Inc., San Francisco, - CA 94111\nE-mail: andrew.white@rochester.edu\nAbstract\nChemists can be skeptical - in using deep learning (DL) in decision making, due to\nthe lack of interpretability - in \u201cblack-box\u201d models. Explainable artificial intelligence\n(XAI) - is a branch of AI which addresses this drawback by providing tools to interpret\nDL - models and their predictions. We review the principles of XAI in the domain - of\nchemistry and emerging methods for creating and evaluating explanations. - Then we\nfocus on methods developed by our group and their applications in predicting - solubil-\nity, blood-brain barrier permeability, and the scent of molecules. - We show that XAI\nmethods like chemical counterfactuals and descriptor explanations - can explain DL pre-\ndictions while giving insight into structure-property relationships. - Finally, we discuss\nhow a two-step process of developing a black-box model - and explaining predictions can\nuncover structure-property relationships.\n1\nIntroduction\nDeep - learning (DL) is advancing the boundaries of computational chemistry because - it can\naccurately model non-linear structure-function relationships.1\u20133 - Applications of DL can be\nfound in a broad spectrum spanning from quantum computing4,5 - to drug discovery6\u201310 to\nmaterials design.11,12 According to Kre 13, DL - models can contribute to scientific discovery\nin three \u201cdimensions\u201d - - 1) as a \u2018computational microscope\u2019 to gain insight which are not\nattainable - through experiments 2) as a \u2018resource of inspiration\u2019 to motivate - scientific thinking\n3) as an \u2018agent of understanding\u2019 to uncover - new observations. However, the rationale of\na DL prediction is not always apparent - due to the model architecture consisting a large\nparameter count.14,15 DL models - are thus often termed\u201cblack box\u201d models. We can only\nreason about - the input and output of an DL model, not the underlying cause that leads to\na - specific prediction.\nIt is routine in chemistry now for DL to exceed human - level performance \u2014 humans are\nnot good at predicting solubility from - structure for example161 \u2014 and so understanding how\na model makes predictions - can guide hypotheses. This is in contrast to a topic like finding\na stop sign - in an image, where there is little new to be learned about visual perception\nby - explaining a DL model. However, the black box nature of DL has its own limitations.\nUsers - are more likely to trust and use predictions from a model if they can understand - why\nthe prediction was made.17 Explaining predictions can help developers of - DL models ensure\nthe model is not learning spurious correlations.18,19 Two - infamous examples are, 1)neural\nnetworks that learned to recognize horses by - looking for a photogr", "o infamous examples are, 1)neural\nnetworks that learned - to recognize horses by looking for a photographer\u2019s watermark20 and,\n2) - neural networks that predicted a COVID-19 diagnoses by looking at the font choice\non - medical images.21 As a result, there is an emerging regulatory framework for - when any\ncomputer algorithms impact humans.22\u201324 Although we know of no - examples yet in chemistry,\none can assume the use of AI in predicting toxicity, - carcinogenicity, and environmental\npersistence will require rationale for the - predictions due to regulatory consequences.\n1there does happen to be one human - solubility savant, participant 11, who matched machine performance\n2\nEXplainable - Artificial Intelligence (XAI) is a field of growing importance that aims to\nprovide - model interpretations of DL predictions Three terms highly associated with XAI - are,\ninterpretability, justifications and explainability. Miller 25 defines - that interpretability of a\nmodel refers to the degree of human understandability - intrinsic within the model. Murdoch\net al. 26 clarify that interpretability - can be perceived as \u201cknowledge\u201d which provide insight\nto a particular - problem.\nJustifications are quantitative metrics tell the users \u201cwhy the\nmodel - should be trusted,\u201d like test error.27 Justifications are evidence which - defend why a\nprediction is trustworthy.25 An \u201cexplanation\u201d is a description - on why a certain prediction was\nmade.9,28 Interpretability and explanation - are often used interchangeably. Arrieta et al. 14\ndistinguish that interpretability - is a passive characteristic of a model, whereas explainability\nis an active - characteristic which is used to clarify the internal decision-making process.\nNamely, - an explanation is extra information that gives the context and a cause for one - or\nmore predictions.29 We adopt the same nomenclature in this perspective.\nAccuracy - and interpretability are two attractive characteristics of DL models. However,\nDL - models are often highly accurate and less interpretable.28,30 XAI provides a - way to avoid\nthat trade-off in chemical property prediction. XAI can be viewed - as a two-step process.\nFirst, we develop an accurate but uninterpretable DL - model. Next, we add explanations to\npredictions. Ideally, if the DL model has - correctly learned the input-output relations, then\nthe explanations should - give insight into the underlying mechanism.\nIn the remainder of this article, - we review recent approaches for XAI of chemical property\nprediction while drawing - specific examples from our recent XAI work.9,10,31 We show how\nin various systems - these methods yield explanations that are consistent with known and\nmechanisms - in structure-property relationships.\n3\nTheory\nIn this work, we aim to assemble - a common taxonomy for the landscape of XAI while\nproviding our perspectives. - We utilized the vocabulary proposed by Das and Rad 32 to classify\nXAI. According - to their classification, interpretations can be categorized as global or local\ninterpretations - on the basis of \u201cwhat ", "cation, interpretations can be categorized as - global or local\ninterpretations on the basis of \u201cwhat is being explained?\u201d. - For example, counterfactuals are\nlocal interpretations, as these can explain - only a given instance. The second classification is\nbased on the relation between - the model and the interpretation \u2013 is interpretability post-hoc\n(extrinsic) - or intrinsic to the model?.32,33 An intrinsic XAI method is part of the model\nand - is self-explanatory32 These are also referred to as white-box models to contrast - them\nwith non-interpretable black box models.28 An extrinsic method is one - that can be applied\npost-training to any model.33 Post-hoc methods found in - the literature focus on interpreting\nmodels through 1) training data34 and - feature attribution,35 2) surrogate models10 and, 3)\ncounterfactual9 or contrastive - explanations.36\nOften, what is a \u201cgood\u201d explanation and what are - the required components of an ex-\nplanation are debated.32,37,38 Palacio et - al. 29 state that the lack of a standard framework\nhas caused the inability - to evaluate the interpretability of a model. In physical sciences,\nwe may instead - consider if the explanations somehow reflect and expand our understanding\nof - physical phenomena. For example, Oviedo et al. 39 propose that a model explanation\ncan - be evaluated by considering its agreement with physical observations, which - they term\n\u201ccorrectness.\u201d For example, if an explanation suggests - that polarity affects solubility of a\nmolecule, and the experimental evidence - strengthen the hypothesis, then the explanation\nis assumed \u201ccorrect\u201d. - In instances where such mechanistic knowledge is sparse, expert bi-\nases and - subjectivity can be used to measure the correctness.40 Other similar metrics - of\ncorrectness such as \u201cexplanation satisfaction scale\u201d can be found - in the literature.41,42 In a\nrecent study, Humer et al. 43 introduced CIME - an interactive web-based tool that allows the\nusers to inspect model explanations. - The aim of this study is to bridge the gap between\nanalysis of XAI methods. - Based on the above discussion, we identify that an agreed upon\n4\nevaluation - metric is necessary in XAI. We suggest the following attributes can be used - to\nevaluate explanations. However, the relative importance of each attribute - may depend on\nthe application - actionability may not be as important as faithfulness - when evaluating the\ninterpretability of a static physics based model. Therefore, - one can select relative importance\nof each attribute based on the application.\n\u2022 - Actionable. Is it clear how we could change the input features to modify the - output?\n\u2022 Complete. Does the explanation completely account for the prediction? - Did features\nnot included in the explanation really contribute zero effect - to the prediction?44\n\u2022 Correct. Does the explanation agree with hypothesized - or known underlying physical\nmechanism?39\n\u2022 Domain Applicable. Does the - explanation use language and concepts of domain ex-\nperts?\n\u2022 Fidelity/Faithful. - Does the explanation ag", "lanation use language and concepts of domain ex-\nperts?\n\u2022 - Fidelity/Faithful. Does the explanation agree with the black box model?\n\u2022 - Robust. Does the explanation change significantly with small changes to the - model or\ninstance being explained?\n\u2022 Sparse/Succinct. Is the explanation - succinct?\nWe present an example evaluation of the SHAP explanation method based - on the above\nattributes.44 Shapley values were proposed as a local explanation - method based on feature\nattribution, as they offer a complete explanation - - each feature is assigned a fraction of\nthe prediction value.44,45 Completeness - is a clearly measurable and well-defined metric, but\nyields explanations with - many components. Yet Shapley values are not actionable nor sparse.\nThey are - non-sparse as every feature has a non-zero attribution and not-actionable because\nthey - do not provide a set of features which changes the outcome.46 Ribeiro et al. - 35 proposed\na surrogate model method that aims to provide sparse/succinct explanations - that have high\n5\nfidelity to the original model. In Wellawatte et al. 9 we - argue that counterfactuals are \u201cbet-\nter\u201d explanations because they - are actionable and sparse. We highlight that, evaluation of\nexplanations is - a difficult task because explanations are fundamentally for and by humans.\nTherefore, - these evaluations are subjective, as they depend on \u201ccomplex human factors - and\napplication scenarios.\u201d37\nSelf-explaining models\nA self-explanatory - model is one that is intrinsically interpretable to an expert.47 Two com-\nmon - examples found in the literature are linear regression models and decision trees - (DT).\nIntrinsic models can be found in other XAI applications acting as surrogate - models (proxy\nmodels) due to their transparent nature.48,49 A linear model - is described by the equation\n1 where, W\u2019s are the weight parameters and - x\u2019s are the input features associated with the\nprediction \u02c6y. Therefore, - we observe that the weights can be used to derive a complete expla-\nnation - of the model - trained weights quantify the importance of each feature.47 DT - models\nare another type of self-explaining models which have been used in classification - and high-\nthroughput screening tasks. Gajewicz et al. 50 used DT models to - classify nanomaterials\nthat identify structural features responsible for surface - activity. In another study by Han\net al. 51, a DT model was developed to filter - compounds by their bioactivity based on the\nchemical fingerprints.\n\u02c6y - = \u03a3iWixi\n(1)\nRegularization techniques such as EXPO52 and RRR53 are designed - to enhance the black-\nbox model interpretability.54 Although one can argue - that \u201csimplicity\u201d of models are posi-\ntively correlated with interpretability, - this is based on how the interpretability is evaluated.\nFor example, Lipton - 55 argue that, from the notion of \u201csimulatability\u201d (the degree to - which a\nhuman can predict the outcome based on inputs), self-explanatory linear - models, rule-based\n6\nsystems, and DT\u2019s can be claimed uninterpretable. - A human can pred", "atory linear models, rule-based\n6\nsystems, and DT\u2019s - can be claimed uninterpretable. A human can predict the outcome given\nthe inputs - only if the input features are interpretable. Therefore, a linear model which - takes\nin non-descriptive inputs may not be as transparent. On the other hand, - a linear model\nis not innately accurate as they fail to capture non-linear - relationships in data, limiting is\napplicability. Similarly, a DT is a rule-based - model and lacks physics informed knowledge.\nTherefore, an existing drawback - is the trade-off between the degree of understandability and\nthe accuracy of - a model. For example, an intrinsic model (linear regression or decision trees)\ncan - be described through the trainable parameters, but it may fail to \u201ccorrectly\u201d - capture\nnon-linear relations in the data.\nAttribution methods\nFeature attribution - methods explain black box predictions by assigning each input feature\na numerical - value, which indicates its importance or contribution to the prediction. Feature\nattributions - provide local explanations, but can be averaged or combined to explain multi-\nple - instances. Atom-based numerical assignments are commonly referred to as heatmaps.56\nSheridan - 57 describes an atom-wise attribution method for interpreting QSAR models. Re-\ncently, - Rasmussen et al. 58 showed that Crippen logP models serve as a benchmark for\nheatmap - approaches. Other most widely used feature attribution approaches in the litera-\nture - are gradient based methods,59,60 Shapley Additive exPlanations (SHAP),44 and - layer-\nwise relevance prorogation.61\nGradient based approaches are based on - the hypothesis that gradients for neural net-\nworks are analogous to coefficients - for regression models.62 Class activation maps (CAM),63\ngradCAM,64 smoothGrad,,65 - and integrated gradients62 are examples of this method. The\nmain idea behind - feature attributions with gradients can be represented with equation 2.\n\u2206\u02c6f(\u20d7x)\n\u2206xi\n\u2248\u2202\u02c6f(\u20d7x)\n\u2202xi\n(2)\n7\nwhere - \u02c6f(x) is the black-box model and\n\u2206\u02c6f(\u20d7x)\n\u2206xi\nare - used as our attributions. The left-\nhand side of equation 2 says that we attribute - each input feature xi by how much one unit\nchange in it would affect the output - of \u02c6f(x). If \u02c6f(x) is a linear surrogate model, then this\nmethod - reconciles with LIME.35 In DL models, \u2207xf(x), suffers from the shattered - gradients\nproblem.62 This means directly computing the quantity leads to numeric - problems. The\ndifferent gradient based approaches are mostly distinguishable - based on how the gradient is\napproximated.\nGradient based explanations have - been widely used to interpret chemistry predictions.60,66\u201370\nMcCloskey - et al. 60 used graph convolutional networks (GCNs) to predict protein-ligand\nbinding - and explained the binding logic for these predictions using integrated gradients.\nPope - et al. 66 and Jim\u00b4enez-Luna et al. 67 show application of gradCAM and integrated - gradi-\nents to explain molecular property predictions from trained graph neural - networks (GNNs).\nSanchez-Lengeling et al. 68 present compr", "rty predictions - from trained graph neural networks (GNNs).\nSanchez-Lengeling et al. 68 present - comprehensive, open-source XAI benchmarks to explain\nGNNs and other graph based - models. They compare the performance of class activation\nmaps (CAM),63 gradCAM,64 - smoothGrad,,65 integrated gradients62 and attention mecha-\nnisms for explaining - outcomes of classification as well as regression tasks. They concluded\nthat - CAM and integrated gradients perform well for graph based models. Another attempt\nat - creating XAI benchmarks for graph models was made by Rao et al. 70. They compared\nthese - gradient based methods to find subgraph importance when predicting activity - cliffs\nand concluded that gradCAM and integrated gradients provided the most - interpretability\nfor GNNs.\nThe GNNExplainer69 is an approach for generating - explanations (local and\nglobal) for graph based models. This method focuses - on identifying which sub-graphs con-\ntribute most to the prediction by maximizing - mutual information between the prediction\nand distribution of all possible - sub-graphs. Ying et al. 69 show that GNNExplainer can be\nused to obtain model-agnostic - explanations. SubgraphX is a similar method that explains\nGNN predictions by - identifying important subgraphs.71\nAnother set of approaches like DeepLIFT72 - and Layerwise Relevance backPropagation73\n8\n(LRP) are based on backpropagation - of the prediction scores through each layer of the neu-\nral network. The specific - backpropagation logic across various activation functions differs\nin these - approaches, which means each layer must have its own implementation. Baldas-\nsarre - and Azizpour 74 showed application of LRP to explain aqueous solubility prediction - for\nmolecules.\nSHAP is a model-agnostic feature attribution method that is - inspired from the game\ntheory concept of Shapley values.44,46 SHAP has been - popularly used in explaining molecular\nprediction models.75\u201378 It\u2019s - an additive feature contribution approach, which assumes that\nan explanation - model is a linear combination of binary variables z. If the Shapley value\ni - \u03d5i(\u20d7x)zi(\u20d7x). Shapley values for\nfeatures are computed using - Equation 3.79,80\nfor the ith feature is \u03d5i, then the explanation is \u02c6f(\u20d7x) - = P\n\u03d5i(\u20d7x) = 1\nM\nM\nX \u02c6f (\u20d7z+i) \u2212\u02c6f (\u20d7z\u2212i)\n(3)\nHere - \u20d7z is a fabricated example created from the original \u20d7x and a random - perturbation \u20d7x\u2032.\n\u20d7z+i has the feature i from \u20d7x and \u20d7z\u2212i - has the ith feature from \u20d7x\u2032. Some care should be taken\nin constructing - \u20d7z when working with molecular descriptors to ensure that an impossible - \u20d7z is\nnot sampled (e.g., high count of acid groups but no hydrogen bond - donors). M is the sample\nsize of perturbations around \u20d7x. Shapley value - computation is expensive, hence M is chosen\naccordingly. Equation 3 is an approximation - and gives contributions with an expectation\ni=1 \u03d5i(\u20d7x) = \u02c6f(\u20d7x).\nVisualization - based feature attribution has also been used for molecular data. In com-\nterm - as \u03d50 + P\nputer science, saliency maps are a way to measure spatial feature - contribution.8", "com-\nterm as \u03d50 + P\nputer science, saliency maps are - a way to measure spatial feature contribution.81 Simply put,\nsaliency maps - draw a connection between the model\u2019s neural fingerprint components (trained\nweights) - and input features. Weber et al. 82 used saliency maps to build an explainable - GCN\narchitecture that gives subgraph importance for small molecule activity - prediction. On the\nother hand, similarity maps compare model predictions for - two or more molecules based on\ntheir chemical fingerprints.83 Similarity maps - provide atomic weights or predicted probabil-\n9\nity difference between the - molecules by removing one atom at a time. These weights can\nthen be used to - color the molecular graph and give a visual presentation. ChemInformatics\nModel - Explorer (CIME) is an interactive web based toolkit which allows visualization - and\ncomparison of different explanation methods for molecular property prediction - models.84\nSurrogate models\nOne approach to explain black box predictions is - to fit a self-explaining or interpretable\nmodel to the black box model, in - the vicinity of one or a few specific examples. These are\nknown as surrogate - models. Generally, one model per explanation is trained. However, if we\ncould - find one surrogate model that explained the whole DL model, then we would simply\nhave - a globally accurate interpretable model. This means that the black-box model - is no\nlonger needed.79 In the work by White 79, a weighted least squares linear - model is used as\nthe surrogate model. This model provides natural language - based descriptor explanations by\nreplacing input features with chemically interpretable - descriptors. This approach is similar\nto the concept-based explanations approach - used by McGrath et al. 85, where human under-\nstandable concepts were used - in place of input features in acquisition of chess knowledge in\nAlphaZero. - Any of the self-explaining models detailed in the Self-explaining models section\ncan - be used as a surrogate model.\nThe most commonly used surrogate model based - method is Locally Interpretable Model\nExplanations (LIME).35 LIME creates perturbations - around the example of interest and fits\nan interpretable model to these local - perturbations. Ribeiro et al. 35 mathematically define\nan explanation \u03be - for an example \u20d7x using Equation 4.\n\u03be(\u20d7x) = arg min\ng\u2208G - L(f, g, \u03c0x) + \u2126(g)\n(4)\nHere f is the black box model and g \u2208G - is the interpretable explanation model. G is\na class of potential interpretable - models (e.g.: linear models). \u03c0x is a similarity measure\n10\nbetween original - input \u20d7x and it\u2019s perturbed input \u20d7x\u2032. In context of molecular - data, this can\nbe a chemical similarity metric like Tanimoto86 similarity between - fingerprints. The goal for\nLIME is to minimize the loss, L, such that f is - closely approximated by g. \u2126is a parameter\nthat controls the complexity - (sparsity) of g. Ribeiro et al. 35 termed the agreement (how low\nthe loss is) - between f and g as the \u201cfidelity\u201d.\nGraphLIME87 and LIMEtree88 are - modifications to LIME as applica", ") between f and g as the \u201cfidelity\u201d.\nGraphLIME87 - and LIMEtree88 are modifications to LIME as applicable to graph neural\nnetworks - and regression trees, respectively. LIME has been used in chemistry previously,\nsuch - as Whitmore et al. 89 who used LIME to explain octane number predictions of - molecules\nfrom a random forest classifier. Mehdi and Tiwary 90 used LIME to - explain thermodynamic\ncontributions of features. Gandhi and White 10 use an - approach similar to GraphLIME,\nbut use chemistry specific fragmentation and - descriptors to explain molecular property pre-\ndiction.\nSome examples are - highlighted in the Applications section.\nIn recent work by\nMehdi and Tiwary - 90, a thermodynamic-based surrogate model approach was used to inter-\npret - black-box models. The authors define an \u201cinterpretation free energy\u201d - which can be\nachieved by minimizing the surrogate model\u2019s uncertainty - and maximizing simplicity.\nCounterfactual explanations\nCounterfactual explanations - can be found in many fields such as statistics, mathematics and\nphilosophy.91\u201394 - According to Woodward and Hitchcock 92, a counterfactual is an example\nwith - minimum deviation from the initial instance but with a contrasting outcome. - They\ncan be used to answer the question, \u201cwhich smallest change could - alter the outcome of an\ninstance of interest?\u201d While the difference between - the two instances is based on the exis-\ntence of similar worlds in philosophy,95 - a distance metric based on molecular similarity is\nemployed in XAI for chemistry. - For example, in the work by Wellawatte et al. 9 distance\nbetween two molecules - is defined as the Tanimoto distance96 between ECFP4 fingerprints.97\nAdditionally, - Mohapatra et al. 98 introduced a chemistry-informed graph representation for\ncomputing - macromolecular similarity. Contrastive explanations are peripheral to counterfac-\n11\ntual - explanations. Unlike the counterfactual approach, contrastive approach employ - a dual\noptimization method, which works by generating a similar and a dissimilar - (counterfactuals)\nexample. Contrastive explanations can interpret the model - by identifying contribution of\npresence and absence of subsets of features - towards a certain prediction.36,99\nA counterfactual x\u2032 of an instance - x is one with a dissimilar prediction \u02c6f(x) in classi-\nfication tasks. - As shown in equation 5, counterfactual generation can be thought of as a\nconstrained - optimization problem which minimizes the vector distance d(x, x\u2032) between - the\nfeatures.9,100\nminimize\nd(x, x\u2032)\nsuch that\n\u02c6f(x) \u0338= - \u02c6f(x\u2032)\n(5)\nFor regression tasks, equation 6 adapted from equation - 5 can be used. Here, a counter-\nfactual is one with a defined increase or decrease - in the prediction.\nminimize\nd(x, x\u2032)\nCounterfactuals explanations have - become a useful tool for XAI in chemistry, as they\nsuch that\n\f\f\f \u02c6f(x) - \u2212\u02c6f(x\u2032)\n\f\f\f \u2265\u2206\n(6)\nprovide intuitive understanding - of predictions and are able to uncover spurious relationships\nin training data.101 - Counterfactuals create local (instance-level), actionable explan", " relationships\nin - training data.101 Counterfactuals create local (instance-level), actionable - explanations.\nActionability of an explanation suggest which features can be - altered to change the outcome.\nFor example, changing a hydrophobic functional - group in a molecule to a hydrophilic group\nto increase solubility.\nCounterfactual - generation is a demanding task as it requires gradient optimization over\ndiscrete - features that represents a molecule. Recent work by Fu et al. 102 and Shen et - al. 103\npresent two techniques which allow continuous gradient-based optimization. - Although, these\nmethodologies are shown to circumvent the issue of discrete - molecular optimization, counter-\nfactual explanation based model interpretation - still remains unexplored compared to other\n12\npost-hoc methods.\nCF-GNNExplainer104 - is a counterfactual explanation generating method based on GN-\nNExplainer69 - for graph data. This method generate counterfactuals by perturbing the input\ndata - (removing edges in the graph), and keeping account of perturbations which lead - to\nchanges in the output.\nHowever, this method is only applicable to graph-based - models\nand can generate infeasible molecular structures. Another related work - by Numeroso and\nBacciu 105 focus on generating counterfactual explanations - for deep graph networks. Their\nmethod MEG (Molecular counterfactual Explanation - Generator) uses a reinforcement learn-\ning based generator to create molecular - counterfactuals (molecular graphs).\nWhile this\nmethod is able to generate - counterfactuals through a multi-objective reinforcement learner,\nthis is not - a universal approach and requires training the generator for each task.\nWork - by Wellawatte et al. 9 present a model agnostic counterfactual generator MMACE\n(Molecular - Model Agnostic Counterfactual Explanations) which does not require training\nor - computing gradients. This method firstly populates a local chemical space through - ran-\ndom string mutations of SELFIES106 molecular representations using the - STONED algo-\nrithm.107 Next, the labels (predictions) of the molecules in the - local space are generated\nusing the model that needs to be explained. Finally, - the counterfactuals are identified and\nsorted by their similarities \u2013 - Tanimoto distance96 between ECFP4 fingerprints.97 Unlike the\nCF-GNNExplainer104 - and MEG105 methods, the MMACE algorithm ensures that generated\nmolecules are - valid, owing to the surjective property of SELFIES. Additionally, the MMACE\nmethod - can be applied to both regression and classification models. However, like most - XAI\nmethods for molecular prediction, MMACE does not account for the chemical - stability of\npredicted counterfactuals.\nTo circumvent this drawback, Wellawatte - et al. 9 propose an-\nother approach, which identift counterfactuals through - a similarity search on the PubChem\ndatabase.108\n13\nSimilarity to adjacent - fields\nTangential examples to counterfactual explanations are adversarial training - and matched\nmolecular pairs. Adversarial perturbations are used d", "lanations - are adversarial training and matched\nmolecular pairs. Adversarial perturbations - are used during training to deceive the model\nto expose the vulnerabilities - of a model109,110 whereas counterfactuals are applied post-hoc.\nTherefore, - the main difference between adversarial and counterfactual examples are in the\napplication, - although both are derived from the same optimization problem.100 Grabocka\net - al. 111 have developed a method named Adversarial Training on EXplanations (ATEX)\nwhich - improves model robustness via exposure to adversarial examples. While there - are\nconceptual disparities, we note that the counterfactual and adversarial - explanations are\nequivalent mathematical objects.\nMatched molecular pairs - (MMPs) are pairs of molecules that differ structurally at only\none site by - a known transformation.112,113 MMPs are widely used in drug discovery and\nmedicinal - chemistry as these facilitate fast and easy understanding of structure-activity - re-\nlationships.114\u2013116 Counterfactuals and MMP examples intersect if - the structural change is\nassociated with a significant change in the properties. - In the case the associated changes in\nthe properties are non-significant, the - two molecules are known as bioisosteres.117,118 The con-\nnection between MMPs - and adversarial training examples has been explored by van Tilborg\net al. 119. - MMPs which belong to the counterfactual category are commonly used in outlier\nand - activity cliff detection.113 This approach is analogous to counterfactual explanations,\nas - the common objective is to uncover learned knowledge pertaining to structure-property\nrelationships.70\nApplications\nModel - interpretation is certainly not new and a common step in ML in chemistry, but - XAI for\nDL models is becoming more important60,66\u201369,73,88,104,105 Here - we illustrate some practical\nexamples drawn from our published work on how - model-agnostic XAI can be utilized to\n14\ninterpret black-box models and connect - the explanations to structure-property relationships.\nThe methods are \u201cMolecular - Model Agnostic Counterfactual Explanations\u201d (MMACE)9\nand \u201cExplaining - molecular properties with natural language\u201d.10 Then we demonstrate how\ncounterfactuals - and descriptor explanations can propose structure-property relationships in\nthe - domain of molecular scent.31\nBlood-brain barrier permeation prediction\nThe - passive diffusion of drugs from the blood stream to the brain is a critical - aspect in drug\ndevelopment and discovery.120 Small molecule blood-brain barrier - (BBB) permeation is a\nclassification problem routinely assessed with DL models.121,122 - To explain why DL models\nwork, we trained two models a random forest (RF) model123 - and a Gated Recurrent Unit\nRecurrent Neural Network (GRU-RNN). Then we explained - the RF model with generated\ncounterfactuals explanations using the MMACE9 and - the GRU-RNN with descriptor expla-\nnations.10 Both the models were trained - on the dataset developed by Martins et al. 124. The\nRF model was implemented - in Scikit-learn125 usi", " on the dataset developed by Martins et al. 124. The\nRF - model was implemented in Scikit-learn125 using Mordred molecular descriptors126 - as the\ninput features. The GRU-RNN model was implemented in Keras.127 See Wellawatte - et al. 9\nand Gandhi and White 10 for more details.\nAccording to the counterfactuals - of the instance molecule in figure 1, we observe that the\nmodifications to - the carboxylic acid group enable the negative example molecule to permeate\nthe - BBB. Experimental findings by Fischer et al. 120 show that the BBB permeation - of\nmolecules are governed by hydrophobic interactions and surface area. The - carboxylic group is\na hydrophilic functional group which hinders hydrophobic - interactions and addition of atoms\nenhances the surface area. This proves the - advantage of using counterfactual explanations,\nas they suggest actionable - modification to the molecule to make it cross the BBB.\nIn Figure 2 we show - descriptor explanations generated for Alprozolam, a molecule that\npermeates - the BBB, using the method described by Gandhi and White 10.\nWe see that\npredicted - permeability is positively correlated with the aromaticity of the molecule, - while\n15\nnegatively correlated with the number of hydrogen bonds donors and - acceptors. A similar\nstructure-property relationship for BBB permeability is - proposed in more mechanistic stud-\nies.128\u2013130 The substructure attributions - indicates a reduction in hydrogen bond donors and\nacceptors. These descriptor - explanations are quantitative and interpretable by chemists.\nFinally, we can - use a natural language model to summarize the findings into a written\nexplanation, - as shown in the printed text in Figure 2.\nFigure 1: Counterfactuals of a molecule - which cannot permeate the blood-brain barrier.\nSimilarity is the Tanimoto similarity - of ECFP4 fingerprints.131 Red indicates deletions and\ngreen indicates substitutions - and addition of atoms. Republished from Ref.9 with permission\nfrom the Royal - Society of Chemistry.\nSolubility prediction\nSmall molecule solubility prediction - is a classic cheminformatics regression challenge and is\nimportant for chemical - process design, drug design and crystallization.133\u2013136 In our previous\nworks,9,10 - we implemented and trained an RNN model in Keras to predict solubilities (log\nmolarity) - of small molecules.127 The AqSolDB curated database137 was used to train the\nRNN - model.\nIn this task, counterfactuals are based on equation 6. Figure 3 illustrates - the generated\nlocal chemical space and the top four counterfactuals. Based - on the counterfactuals, we ob-\nserve that the modifications to the ester group - and other heteroatoms play an important role\nin solubility. These findings - align with known experimental and basic chemical intuition.134\nFigure 4 shows - a quantitative measurement of how substructures are contributing to the pre-\n16\nFigure - 2: Descriptor explanations along with natural language explanation obtained - for BBB\npermeability of Alprozolam molecule. The green and red bars show descriptors - t", "tion obtained for BBB\npermeability of Alprozolam molecule. The green and - red bars show descriptors that influ-\nence predictions positively and negatively, - respectively. Dotted yellow lines show significance\nthreshold (\u03b1 = 0.05) - for the t-statistic. Molecular descriptors show molecule-level proper-\nties - that are important for the prediction. ECFP and MACCS descriptors indicate which\nsubstructures - influence model predictions. MACCS explanations lead to text explanations\nas - shown. Republished from Ref.10 with permission from authors. SMARTS annotations - for\nMACCS descriptors were created using SMARTSviewer (smartsview.zbh.uni-hamburg.de,\nCopyright: - ZBH, Center for Bioinformatics Hamburg) developed by Schomburg et al. 132.\n17\ndiction. - For example, we see that adding acidic and basic groups as well as hydrogen - bond\nacceptors, increases solubility. Substructure importance from ECFP97 and - MACCS138 de-\nscriptors indicate that adding heteroatoms increases solubility, - while adding rings structures\nmakes the molecule less soluble. Although these - are established hypotheses, it is interesting\nto see they can be derived purely - from the data via DL and XAI.\nFigure 3: Generated chemical space for solubility - prediction using the RNN model. The\nchemical space is a 2D projection of the - pairwise Tanimoto similarities of the local coun-\nterfactuals. Each data point - is colored by solubility. Top 4 counterfactuals are shown here.\nRepublished - from Ref.9 with permission from the Royal Society of Chemistry.\nGeneralizing - XAI \u2013 interpreting scent-structure relationships\nIn this example, we show - how non-local structure-property relationships can be learned with\nXAI across - multiple molecules. Molecular scent prediction is a multi-label classification - task\nbecause a molecule can be described by more than one scent. For example, - the molecule\njasmone can be described as having \u2018jasmine,\u2019 \u2018woody,\u2019 - \u2018floral,\u2019 and \u2019herbal\u2019 scents.139 The\nscent-structure relationship - is not very well understood,140 although some relationships are\nknown. For - example, molecules with an ester functional group are often associated with\n18\nFigure - 4: Descriptor explanations for solubility prediction model. The green and red - bars\nshow descriptors that influence predictions positively and negatively, - respectively. Dotted\nyellow lines show significance threshold (\u03b1 = 0.05) - for the t-statistic. The MACCS and\nECFP descriptors indicate which substructures - influence model predictions. MACCS sub-\nstructures may either be present in - the molecule as is or may represent a modification. ECFP\nfingerprints are substructures - in the molecule that affect the prediction. MACCS descriptor\nare used to obtain - text explanations as shown. Republished from Ref.10 with permission from\nauthors. - SMARTS annotations for MACCS descriptors were created using SMARTSviewer\n(smartsview.zbh.uni-hamburg.de, - Copyright: ZBH, Center for Bioinformatics Hamburg) de-\nveloped by Schomburg - et al. 132.\n19\nthe \u2018fruity\u2019 scent. There are some exceptions ", - "tics Hamburg) de-\nveloped by Schomburg et al. 132.\n19\nthe \u2018fruity\u2019 - scent. There are some exceptions though, like tert-amyl acetate which has a\n\u2018camphoraceous\u2019 - rather than \u2018fruity\u2019 scent.140,141\nIn Seshadri et al. 31, we trained - a GNN model to predict the scent of molecules and utilized\ncounterfactuals9 - and descriptor explanations10 to quantify scent-structure relationships. The\nMMACE - method was modified to account for the multi-label aspect of scent prediction. - This\nmodification defines molecules that differed from the instance molecule - by only the selected\nscent as counterfactuals. For instance, counterfactuals - of the jasmone molecule would be false\nfor the \u2018jasmine\u2019 scent but - would still be positive for \u2018woody,\u2019 \u2018floral\u2019 and \u2018herbal\u2019 - scents.\nFigure 5: Counterfactual for the 2,4 decadienal molecule.\nThe counterfactual - indicates\nstructural changes to ethyl benzoate that would result in the model - predicting the molecule\nto not contain the \u2018fruity\u2019 scent. The Tanimoto96 - similarity between the counterfactual and\n2,4 decadienal is also provided. - Republished with permission from authors.31\nThe molecule 2,4-decadienal, which - is known to have a \u2018fatty\u2019 scent, is analyzed in Fig-\nure 5.142,143 - The resulting counterfactual, which has a shorter carbon chain and no carbonyl\ngroups, - highlights the influence of these structural features on the \u2018fatty\u2019 - scent of 2,4 deca-\ndienal. To generalize to other molecules, Seshadri et al. - 31 applied the descriptor attribution\nmethod to obtain global explanations - for the scents. The global explanation for the \u2018fatty\u2019\nscent was - generated by gathering chemical spaces around many \u2018fatty\u2019 scented - molecules.\nThe resulting natural language explanation is: \u201cThe molecular - property \u201cfatty scent\u201d can\nbe explained by the presence of a heptanyl - fragment, two CH2 groups separated by four\n20\nbonds, and a C=O double bond, - as well as the lack of more than one or two O atoms.\u201d31\nThe importance - of a heptanyl fragment aligns with that reported in the literature, as \u2018fatty\u2019\nmolecules - often have a long carbon chain.144 Furthermore, the importance of a C=O dou-\nble - bond is supported by the findings reported by Licon et al. 145, where in addition - to a\n\u201clarger carbon-chain skeleton\u201d, they found that \u2018fatty\u2019 - molecules also had \u201caldehyde or acid\nfunctions\u201d.145 For the \u2018pineapple\u2019 - scent, the following natural language explanation was ob-\ntained: \u201cThe - molecular property \u201cpineapple scent\u201d can be explained by the presence - of ester,\nethyl/ether O group, alkene/ether O group, and C=O double bond, as - well as the absence of\nan Aromatic atom.\u201d31 Esters, such as ethyl 2-methylbutyrate, - are present in many pineap-\nple volatile compounds.146,147 The combination - of a C=O double bond with an ether could\nalso correspond to an ester group. - Additionally, aldehydes and ketones, which contain C=O\ndouble bonds, are also - common in pineapple volatile compounds.146,148\nDiscussion\nWe have shown two - post-hoc XAI applications based on molecular counterfactual expla-\nnation", - "scussion\nWe have shown two post-hoc XAI applications based on molecular counterfactual - expla-\nnations9 and descriptor explanations.10 These methods can be used to - explain black-box\nmodels whose input is a molecule. These two methods can be - applied for both classification\nand regression tasks. Note that the \u201ccorrectness\u201d - of the explanations strongly depends on\nthe accuracy of the black-box model.\nA - molecular counterfactual is one with a minimal distance from a base molecular, - but\nwith contrasting chemical properties. In the above examples, we used Tanimoto - similar-\nity96 of ECFP4 fingreprints97 as distance, although this should be - explored in the future.\nCounterfactual explanations are useful because they - are represented as chemical structures\n(familiar to domain experts), sparse, - and are actionable. A few other popular examples of\ncounterfactual on graph - methods are GNNExplainer, MEG and CF-GNNExplainer.69,104,105\nThe descriptor - explanation method developed by Gandhi and White 10 fits a self-explaining\n21\nsurrogate - model to explain the black-box model. This is similar to the GraphLIME87 method,\nalthough - we have the flexibility to use explanation features other than subgraphs. Futher-\nmore, - we show that natural language combined with chemical descriptor attributions - can\ncreate explanations useful for chemists, thus enhancing the accessibility - of DL in chemistry.\nLastly, we examined if XAI can be used beyond interpretation. - Work by Seshadri et al. 31 use\nMMACE and surrogate model explanations to analyze - the structure-property relationships\nof scent. They recovered known structure-property - relationships for molecular scent purely\nfrom explanations, demonstrating the - usefulness of a two step process: fit an accurate model\nand then explain it.\nChoosing - among the plethora of XAI methods described here is still an open question.\nIt - remains to be seen if there will ever be a consensus benchmark, since this field - sits on\nthe intersection of human-machine interaction, machine learning, and - philosophy (i.e., what\nconstitutes an explanation?). Our current advice is - to consider first the audience \u2013 domain\nexperts or ML experts or non-experts - \u2013 and what the explanations should accomplish. Are\nthey meant to inform - data selection or model building, how a prediction is used, or how the\nfeatures - can be changed to affect the outcome. The second consideration is what access - you\nhave to the underlying model. The ability to have model derivatives or - propagate gradients\nto the input to models informs the XAI method.\nConclusion - and outlook\nWe should seek to explain molecular property prediction models - because users are more\nlikely to trust explained predictions, and explanations - can help assess if the model is learning\nthe correct underlying chemical principles. - We also showed that black-box modeling first,\nfollowed by XAI, is a path to - structure-property relationships without needing to trade\nbetween accuracy - and interpretability. However, XAI in chemistry has some maj", "thout needing - to trade\nbetween accuracy and interpretability. However, XAI in chemistry has - some major open\nquestions, that are also related to the black-box nature of - the deep learning.\nSome are\n22\nhighlighted below:\n\u2022 Explanation representation: - How is an explanation presented \u2013 text, a molecule, attri-\nbutions, a - concept, etc?\n\u2022 Molecular distance: in XAI approaches such as counterfactual - generation, the \u201cdis-\ntance\u201d between two molecules is minimized. - Molecular distance is subjective. Possibil-\nities are distance based on molecular - properties, synthesis routes, and direct structure\ncomparisons.\n\u2022 Regulations: - As black-box models move from research to industry, healthcare, and\nenvironmental - settings, we expect XAI to become more important to explain decisions\nto chemists - or non-experts and possibly be legally required. Explanations may need\nto be - tuned for be for doctors instead of chemists or to satisfy a legal requirement.\n\u2022 - Chemical space: Chemical space is the set of molecules that are realizable; - \u201crealiz-\nable\u201d can be defined from purchasable to synthesizable to - satisfied valences. What is\nmost useful? Can an explanation consider nearby - impossible molecules? How can we\ngenerate local chemical spaces centered around - a specific molecule for finding counter-\nfactuals or other instance explanations? - Similarly, can \u201cactivity cliffs\u201d be connected\nto explanations and - the local chemical space.149\n\u2022 Evaluating XAI : there is a lack of a systematic - framework (quantitative or qualitative)\nto evaluate correctness and applicability - of an explanation. Can there be a universal\nframework, or should explanations - be chosen and evaluated based on the audience and\ndomain? For example, work - by Rasmussen et al. 58 attempts to focus on comparing\nfeature attribution XAI - methods via Crippen\u2019s logP scores.\n23\nAcknowledgements\nResearch reported - in this work was supported by the National Institute of General Medical\nSciences - of the National Institutes of Health under award number R35GM137966. This work\nwas - supported by the NSF under awards 1751471 and 1764415. We thank the Center for\nIntegrated - Research Computing at the University of Rochester for providing computational\nresources.\nReferences\n(1) - Choudhary, K.; DeCost, B.; Chen, C.; Jain, A.; Tavazza, F.; Cohn, R.; Park, - C. W.;\nChoudhary, A.; Agrawal, A.; Billinge, S. J.; Holm, E.; Ong, S. P.; Wolverton, - C.\nRecent advances and applications of deep learning methods in materials science. - npj\nComputational Materials 2022, 8.\n(2) Keith, J. A.; Vassilev-Galindo, V.; - Cheng, B.; Chmiela, S.; Gastegger, M.; M\u00a8uller, K.-\nR.; Tkatchenko, A. - Combining Machine Learning and Computational Chemistry for\nPredictive Insights - Into Chemical Systems. Chemical Reviews 2021, 121, 9816\u20139872,\nPMID: 34232033.\n(3) - Goh, G. B.; Hodas, N. O.; Vishnu, A. Deep learning for computational chemistry.\nJournal - of Computational Chemistry 2017, 38, 1291\u20131307.\n(4) Deringer, V. L.; Caro, - M. A.; Cs\u00b4anyi, G. Machine Learning Interatomic Potentials as\nEmergi", - "307.\n(4) Deringer, V. L.; Caro, M. A.; Cs\u00b4anyi, G. Machine Learning Interatomic - Potentials as\nEmerging Tools for Materials Science. Advanced Materials 2019, - 31, 1902765.\n(5) Faber, F. A.; Hutchison, L.; Huang, B.; Gilmer, J.; Schoenholz, - S. S.; Dahl, G. E.;\nVinyals, O.; Kearnes, S.; Riley, P. F.; von Lilienfeld, - O. A. Prediction Errors of Molec-\nular Machine Learning Models Lower than Hybrid - DFT Error. Journal of Chemical\nTheory and Computation 2017, 13, 5255\u20135264, - PMID: 28926232.\n24\n(6) Duch, W.; Swaminathan, K.; Meller, J. Artificial Intelligence - Approaches for Rational\nDrug Design and Discovery. Current Pharmaceutical Design - 2007, 13, 1497\u20131508.\n(7) Dara, S.; Dhamercherla, S.; Jadav, S. S.; Babu, - C. M.; Ahsan, M. J.; darasuresh, S. D.;\nDara, S. Machine Learning in Drug Discovery: - A Review. Artificial Intelligence Review\n123, 55, 1947\u20131999.\n(8) Gupta, - R.; Srivastava, D.; Sahu, M.; Tiwari, S.; Ambasta, R. K.; Kumar, P. Artifi-\ncial - intelligence to deep learning: machine intelligence approach for drug discovery.\nMolecular - diversity 2021, 25, 1315\u20131360.\n(9) Wellawatte, G. P.; Seshadri, A.; White, - A. D. Model agnostic generation of counter-\nfactual explanations for molecules. - Chemical Science 2022, 13, 3697\u20133705.\n(10) Gandhi, H. A.; White, A. D. - Explaining structure-activity relationships using locally\nfaithful surrogate - models. chemrxiv 2022,\n(11) Gormley, A. J.; Webb, M. A. Machine learning in - combinatorial polymer chemistry.\nNature Reviews Materials 2021,\n(12) Gomes, - C. P.; Fink, D.; Dover, R. B. V.; Gregoire, J. M. Computational sustainability\nmeets - materials science. Nature Reviews Materials 2021,\n(13) On scientific understanding - with artificial intelligence. Nature Reviews Physics 2022\n4:12 2022, 4, 761\u2013769.\n(14) - Arrieta, A. B.; D\u00b4\u0131az-Rodr\u00b4\u0131guez, N.; Ser, J. D.; Bennetot, - A.; Tabik, S.; Barbado, A.;\nGarcia, S.; Gil-Lopez, S.; Molina, D.; Benjamins, - R.; Chatila, R.; Herrera, F. Explain-\nable Artificial Intelligence (XAI): Concepts, - Taxonomies, Opportunities and Chal-\nlenges toward Responsible AI. Information - Fusion 2019, 58, 82\u2013115.\n(15) Murdoch, W. J.; Singh, C.; Kumbier, K.; - Abbasi-Asl, R.; Yu, B. Interpretable machine\nlearning: definitions, methods, - and applications. ArXiv 2019, abs/1901.04592.\n25\n(16) Boobier, S.; Osbourn, - A.; Mitchell, J. B. Can human experts predict solubility better\nthan computers? - Journal of cheminformatics 2017, 9, 1\u201314.\n(17) Lee, J. D.; See, K. A. - Trust in automation: Designing for appropriate reliance. Human\nFactors 2004, - 46, 50\u201380.\n(18) Bolukbasi, T.; Chang, K.-W.; Zou, J. Y.; Saligrama, V.; - Kalai, A. T. Man is to com-\nputer programmer as woman is to homemaker? debiasing - word embeddings. Advances\nin neural information processing systems 2016, 29.\n(19) - Buolamwini, J.; Gebru, T. Gender Shades: Intersectional Accuracy Disparities - in\nCommercial Gender Classification. Proceedings of the 1st Conference on Fairness,\nAccountability - and Transparency. 2018; pp 77\u201391.\n(20) Lapuschkin, S.; W\u00a8aldchen, - S.; Binder, A.; Montavon, "], "model": "text-embedding-3-small", "encoding_format": - "base64"}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '50126' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/embeddings - response: - body: - string: !!binary | - H4sIAAAAAAAAA1SZS/OCvLOg9+dTvPVunSq5KEn+O+53iFxEnBUoIiiiQALkfPkp/Z2aqdm4AEql - k+5++sl//9c///zbl211mf79zz//Pptx+vd/fa9di6n49z///O//+ueff/7579/n//dk1ZXV9dq8 - 6t/jv5vN61ot//7nH+7/Xvl/D/3nn38hmCWsOfEDMHYdi72ynkSsdq6i8bDkGhi3vkUzQ321a/TJ - TJSFUocVCY/tbGl2jMAxjempShtvlq9lB8Tt60x9WbonyxJNPoTPPY/9Q/zWWIiKg+Sm8YEg12Js - MS5lD8dBfYfTozKH+dI6OmSwUmjQHhVtdo7RA44FZ+N43RTlSD+ggVery6hVzsmwBttNCqs4jLHc - LCc2m4X/gedXreJTiAtt8s46gdO6qlSGhcLEcGAqyurHGReFrgM+8YwP5GPpQzb0tmFs464x+tgX - DuNrFwzkzJQYVrEf0wNUg4HxVmwCR4Ax2V0ezrC4/q6C56J40nDgXCCE3MUFr+LAcLJGwrBMZRuj - 4xNe6eH7+zzAdQyjYcCEi85SsoJKa1CiWjtshDVMVqCoJto185Xe0mufzO+mP6DXnBb0lNu2t4be - roHLc9wT7lFO7QyDKEaXVlZotanebJqznQ+u3sbFvquZgygP7xUZt6NJ8aWehyXpsSm1KGqx5qOc - scc1sZHskQmrYkbaOesOMrqO7oBDajYeJ+6RBLvNxcS5bz7AOr4vF+iMU45DNN+9uYHgAco+NwjD - TswYfDkf8CnbHQ08cB44y7ldQNyGFj0jrx245ia56C2gFhtvNHrLq1p9dOx5mcbH1Nbmi+x2gBUz - DoeuDVum3VuCBo/L8Vm/VqWYZMUKSae9sH3/dO1C7nUKE9XYUe08OwPfNixEioN9qvloB97r6XMA - ikuuNPjuL26yVIiQh0oabh8dmOX20COfTzmabcQ1GQMvqWGwuVWhcBq75D29jAvC7mOgt2p0h1lq - 2wf8+z5gG8l8id4C+rwOGs7u3YVxZU98KDUepq5dt9qqnFiNbpLrkUchMTAHl3VFFZN07FSLPXCC - 27porTOeWtW0AYsrhC5UursbCt42HhjRByi9ZYCo+9R35ee1E3uoWy9C5dOil4vpkQrGrOhp8Msn - /hzmQBZsm+aCx9pJm2QBYV42af4xWm3C9bQDnbOs1P9sD1p3U6IeNWZf4MM2fbTLEG5NqG7ihqpv - /qhxwvTu4SPi91h+3eWSa9V5ROdXo4ainfsa22xzFXG4Najfs0wTfEPJkaxuDWo/XhEQDrm+oiFz - uHD7fPOAna4AQqvGbrj6XgN4/hwW4M57DjWX4cHmvg9y2J+WGB+bfhxmlX9DoBePA452WjzMFBYr - 2guuQj1zajR2FtwMDpo5Y2V/18D8W997oxuEX+JhWIsmNZGQzgj7a4oH7uWed2A/AAsb+yNj65vz - O9je0BM7fIUZ96jHENaCc8apMOXa7DPRhtR9wvAlBlrJt1ECYQ3lClvyUWPCqmEV2PPoET7czcPs - aFIGcktTaEDLZZibsJCgFgu7cCVg1dZbvJOQ0n5e1LexzGaiiBD6yVsNd5F/Z6zezRWMuHyHjQlr - 5Zxp5watspvibH6+wJqJMAN4PFxDIJ5fg6hquxBBMlzDOFYPjM84FiNyFE0sx0niEdGzXWkyxQxf - j5w1iGfr/QE3L4X47AA8cGxNCQqneKL+uQxL0X96O+mbbzRlqu6RzY1+4KV8f7Cj7F7lvPetA6z8 - k4+x6lzZ/JEqAtW7Y2LvyRtM9M/vERpP36e3eqezKTry6z76OBtqXx5OS0IDXKDijleq+l7DZqDc - XNjQR4uP5aAkq9xeCcw+U0K9aDq2c332VJicNohsDkoH5qGNM3TYfhTqjevikQs3FTDIyx31hvTW - rsXlJMP3JpAwlm6EzbckgLv32FsEjtd7u97ieQdf0jUN+22agyUUcxc5lWzT6zd/GJmsDggS1+Ki - z/DAqXSUgF/PN3phxdiy+4FAsFPMAYdGcfA+7yjUIX3tJVwetj7jBsI18LURbWqhy5ONg5wI6MV8 - hLGveoyTr0mHeiTsqHHkS7YCxTVBVYOGGlrds2XO5nAPTPrA4V7Oh/U+P3L4OhUH6m+UtqRbKZDh - tz9815dp9NuP9uL2eaaZw3YDiefERR0Vc0Ji8B7miR96qOzUktpSazGO69YagrU/4BtAtdbb43UD - S3BzKD6dBm1B9+XxV1+83U0HfPe4peBmIpv61yobxGHpG/gxHhG+haEBBNHQdHhO7JAIcnzTlm// - QZ97scHBxjgMvAB1CenonWL5Zmpg6c/Uh9lF1vClZ4LXu5nXw8Woc1pE774cp/hpI9zWPD0rdaIt - 92nI4JB2CrVO+a1c7eLeo9xSFHzgjSYRn8DskXAoHYwD2wYzeWwqMDqpSK9ffpiiHXf46zfXea95 - wq4ROnhtS41GI0m87/vUKJk4FefxUA6C4kfpr/7QnIV9Mmcl3kCXuyy0XN83T9T3tgzN1EJ/8RGy - RYyRErxk6mJjD94vwD6ojDRKlVWaNJaDV4XUvrxiZzvF3gJK9wJdMNn4UF525SrLtxC6rpLhUr7b - nlA/DjYCoqpjb3KfybyEow1+/cH0Xptkqi+6iaR32WGbhX3JHmQtoFZ2NfWVpPJWRVpleOTeE03E - ogSLIu42aEeKnur2gddmOdrncPsyc6rybtrO8jV5oIxaFpX19dwuqVE3KNSzB7WOj3NLbuBSg+IE - dOrHVav98hUs5HLHpmnrw1/+cPtNjcOI94fFljcZnHF8wH4+PdtlE/k5yFO7ovhbD7iAxDJyNuE7 - zE7p6lFtCnZQ3UuE+iqxNOF6vctI9buZSN2u0mh0rAmcLqQkvNx13tInuQyv7VnD2nl+D4uxLWz4 - 3FZZuE7WWI76A84AVheI8X2jMF7WEl+SZbOhzgWpTPDcrobNfr8l4qHkypUrrwX8xVfepo9h/RjD - Do5ZlWI5en6SmdeJDq/72KaBQ7fDUrsjB0/jq6PuIjrewp/NHJpDlWD/zNR2+fLCHy+VCp5Bf3vv - QkiUa0wdFvPDxNpNDUPvAfD5ghq2qPeQwP7E4nDKz5uBDUvdoPOdr8g7I3LJt/0uBgcjfJE6lxzA - aXndQLovPeqvKW2nzIwfgGeGR71pF2pj/lLN3f7RzzT/MA+s8VzakmZMJU2CY+DNssdlcNrkAb2p - DmIsgH0IlXnWqV3JwBsrGX5A7uwfWO9er2HsVtTswbi41L21m5aVAixgGsotNj4Dx+YHmj7AupOS - euLZapfbuD9AOXE6eusDvRS+fPvrf9S9gZixW3nS4dgkCjX8URjGX78eC8Em6MhZ7cLRXQ/X4kap - G0SpRjjMfJQUfE2Wtnkm84//+/1Np94mW8GUVz0ECixwmC9FM8zTx6ngLcQ3HAzatVzFfS/AJX9e - 8GnlscfCAaggR+aLCMH1oTFdb2JU+Ucfx18eoRZHTXAu8ic9xS9SMtfnG5BK2kL2+FyD6Srec9gc - jx8sf0wxme01kaF1ut3DLXgOgM79pYHQtWwiPDeRNgYXaYWnc93g8nhSPPF2v7ig21S/fvkEdBcq - ITy/LZ7agfQZZsEdXKCKqkcDMtlMjI71iH588uNN8Re/gsdq+N68jhrrHu4BnuKDR3M1EMBUPnMT - WvexpKfWYB5reu8ATWe+Y/+znT1yHHodRuEpwraXXbzlDZ8jbLaLHn7qxErErjJ8GHZyRk9kNRkj - zP9AWD7OX37UNH7SQyhJn51Ilq2MhgUd4gyVQ9wT8ZRvyxlVcw1JvGmxfjWJR/VMaSCq2yUUFxp7 - C9m8VxAmCkcDU/eGpdlvO7iNggQbh6zxyNV3VqhHQk8PstIP8wM9e6A/9iNJv/MSvWejCn68L9/M - FqymnWfwLI4tPkm7U8kr3ZlAAnYjKURy0lhpJyPiue5Cjab3W9rx7xqW9HPBf/23zyITkinZUEd4 - jGyV5VOIjNvJJCs2nm2TPOYNyhUPUv01PdrlKr5z6fnEV1IrltFOuyWdwY9X9YscecJpLGxYRgql - 9ocNYOGqcw5PwE/pTbCkYUTVrgYcpR01hmX55f8Idh+Zw26dvMr+oX1W8OMrF8ikZGAkK7wqNKLe - 5BoJC9OKwHayVKq12wHUifA8wK2Rd9iwvAugh/GpQ24lTgg60DD6Vl87OLdcRy1eeSfLrYxlST/r - W6pyl4Gx15w18EiIEQo8CTxBue46mMzACp+R/gEfb3qq8MWRMNw/5cGbquMFwh6eXt95MRxWYSIj - CPhJwM4FNWC4rayA3cuYqcs7dblq+dRBU/JaGh4oHT6+oRRIXT8nqr425tc3PHJ4DOSElnRMQK2f - 4wd8yoONvdfbG+bTPVCREEkFlUvOb//qd0XkDTk6SjNMkta5wLyIT2wje8MW4eZVsD/HLVYTewJr - Vkg7GBnPN87N5+BNl2XM4PW1ilTZvGpAOV8MQY+VNzY4dPubj+HzGVypC+SwXELx4MItz3rqfn3B - 6GhrCljwWbH1m38s7mXCgd1WMskbMjBlLDnQX/sTvW73WTLr5+IB75dwG27NvVnOie0/pG0zPnA5 - uUY5L0jn0AmEKfUKoWDrVtQ2iNMkicpeZrHpOw/uv/xCw1WbyrUJbxCScvTwGRrmsGQciOEFCwgr - uDh6U/aOdERs+g5F1p7Y58sXwApsgwb6wQDs1C4yqrKHSYsg4v6n/q16Y2MzMDDguSrKQXvbPkOp - sSaNFWdZQoNVPbDrt0OybtUuhHR/9rByz7yE/8UrntrmG++Affk8hXsrW2lIZEPj5luUgiDy7liG - xZ3Nb6JekOpZN+qHJmFLXK4ces1ZEfKWBxn57g/IeU5EHXyuGTvK9gW094uI5W++caumbOA5kxdc - wc9Dm19uJMFN9jzRcJR6728+KU2SUGO7Vz3Ock4VWL35hhNTntr5nF128OMLPQ338q5l/YPLpDdO - HfydH77rb+kw57kDPq4yV86qpRyglw7xz4d5vKoHMbgfHly4306xtgQ3wMFd6lHyHlDmcad1jGHU - gBe1vj6McJty94tXCBtogREJZAPu03jA5r2DbN6cnw38xjeceTZorH6dfQg2cRmi7T4r59eU5/CT - RvdwY6iADXTvENi9rJlcx3vIFjiVM5CtdB++w0Rj031qU9T0qU/xsX0mi+LkB3iOekTP52Fov36s - h6Wa3UION6PW02A10c+fWd5H0/hvPklUOz7DhRfz8juPFz/+Dpdz0nrzzphNVBzVe/jhHblc4ee0 - QllwbRzhgveIyr9kyPU3TDgYLmwRVFVGzfH0weZzE3k/vwF0AnV8iu5RuWD0SpH2SWYitYGrMTkV - a8jMvMTJ7SUOv/4PteVjhdtAiNpZeb84CKsKksm1EsarL28FKJ10ms73jTe/mzre4xOxp03fsO9+ - 36vQ0osr9r7zGT+OhwodrhtMtX0naLOwlSU0+rgI9+T+aJfDzvUR2csOvsz2cfjWUxOe9bb+xqf1 - Zsn8uJDDd4PMXdmWnNQOHfjOPzjx8ewtXWW7EKVUx9bxsW/Jd55Az+0lo44TdIw1+08DG1yoIR/p - H8ZOfjpK3lkxqXpMe40o0qpCdb8j2Bo3ijYdSnlGp7UasRW/wv/hVSHIdOpr3ENbnTIZkfcOXvib - bx6vle8Obp25oHjua9BtC7eGmr9dv35UZPOu2TeQMXLGlmGmTACVV8MrVUJsWvCTLGgDZLiQ6k7t - i38s52yR/N+8QiPS7EBvBHMOmi3TCXHrSWN7MZPQl+/xLYRh++WfGkbDG2Ovqs1EuFI5Rr//c+Qe - DVvl9kigBTUhBKNul0w76BXy2ljG2pAOyeofhg7ulttEvYdSe7Mat5c/nxdU+KGtNqpt9OXfEK4p - HQb/fCco59ItVfMt9GhnyRw8KoVNMU5VMAVLqcONtbOwkgedt3K7jINfX4N//piLPpGKHPO9UkxW - oL3J5j5LjSeH9CYfd94KgyqFidwYBOw3Vcvg8VP8/CN16j4Dy8azMhDdhIbqMIwAmRPOhz9f4wdy - xJZVqlfIH1QrBLeuTX71Hnx9C1aUfTm8rNOzB9/1xKYcb72fz/zjjVB03EQ0trcQetFMaSVHLlg3 - 1uLDrdpo2NM0YfjjbaVrXRoYWcPGrx8D3/sh4Kx5+MZ7hN/5LKTyMdeEOLh8wKvqXzTCq1POcUJW - sN20hy8/QLAgfedCS7potAL+2r4GcUvAYWPP+OZUqCRUkHskcKiixiqnJS8m91haXr5Lw7DlywmM - 7xEOo7ontaz07bIhgQAOGs2osWGGxvZ13MDIeL2xv48KMJY9CfcKjVccRPrNW1fNUmGZSRw1HEVt - Bctx8x8PEyQKRFv9Q/uA57tY0T+fcHXkETaVFZPxzNRhLRnLYOCMPdaPS9hyzjHqIFROdfjL1zlS - jhXIgvORWhrCJeuqwId+LgsUS2LrLXDHbyDSMAvfH+eQiCdZqiEPOIa1+iSX/HvRavh24ZHevI+Y - rC7e1/B7nkC9L7+wPjnqP39MZbPbJqwhdvbjW+peac5m8hAusGk3ORG3B69d12LI4HO8cRhv+1Rb - 3nAi0qDpMzW/PmeGj+MHSjs5pSe609ji26cGelt9pLKw773lnbombIaQ4kDIL4AdxkmHx8nncPnk - n4x9nl0uKTc9omoiZMPifdoNuO16CafRuSinHHwKWE/lHHZyfPOms7Uh8HYjMznvylKbb4mxQUny - dGilXnUw7evXA3apOmLDoUP5qAX/Ar4+E8tfXqdX7HKAPY0B68+7O6xrZzTwvsYddTbbd/k+lPIK - W6N3qZK2xTDuojhGl36x8M8vjfK1fMBCF2Sq3LMhmS5pI6FvfQ3Z4fVIlswsOsCXnxgrQUfLCbGc - g9W2xvR3PjBZppNKSSHW2Li6IFlbsWzg11eEvSE2CXvI51Ga641G9nhFbd9hUf3xNdZrbV/SnM8P - KDVXn4b3gCtJKJ196LpaFnYxROAtdY8PfJeuQOagUT3xgIcebBZ5xV7jFeU8RMyEU2hNWElbqZ1+ - Pvbbnynmzwv72Kh3f/MK9TrQgEVyWh8ZCQFYrtHq/d4PBiZn4K8/BWyPyg14vMYN9tb3Vlu9aVLh - 4a6kOA05rXy5j0hAYadmOJC7IpkvZIgBwFef9OJ2Xy4f0e6AkeuUhnFQg89W9DbAq3YevlwroWWX - MF+hmRoo3M2p8T98VCD4+c7fnLda4qDDTW721Nw+gTZ+fRGMhcijNhFS769fG8U2DrmAdOU6P5ML - CF5q98eHjEy4k25eBkOEn4Y2zQdmw+nhpH/+hEtsvwPZS5hDiXfkZDrPsv+bx8PBfszJ4ia4Bx0/ - VtikVvbzcRCKyTyFWy5JGWNyA2G6LgWWd3OWzJvz1OyjvPNCvng+SsY/yvrnA8je5tevL4pthM76 - LpTUqG7n3LU/MNhcK1zxx8cwDqedC49TyIXSra1aYomDCQ+CdMbq8JZA+yn8EN6G+5W6RugBAc0X - AfrvxiLuNldb4eo7M/z1Dw07KyOjvONgQ7sWh33D2ONN3As8EVOkeNtzHjECVIHa5ksa3KXcY8lL - kuH+vgZYsziWjLlr91C1QxiC5zVqv3xeQNyVORGTaSj7Jowl6btfSbK+P7/zPBcduWEie3XzSCY3 - wR/Y5lOI5Qd+J29VDw5/vGY/J8JYrccjOksvk8rcRLS1huEDboeLj1MZfPkHD70E1s8BG8DZDyyN - rBXiEN6oXvmDtvJ69McLOMey1fKcifTf+lDZvjy1sV3OOewi26H6l8+o66MaKEll4kBS7fI5Omco - ffcPduLM18RdqPjwiCUhFEMsefMuKmLIgn6lDmQ0Ec4HdYVd5Dpkc23khBOzbQ4xr5pkft3rhDWX - tw66y5pi55ycEmr7WgV+PinsmwTMW/Vtw9LTLZxOjxj8znshf4qrEM17TSPOsfrAb32gTmf23vrz - T3u9wli/5EY5C4fa//EVvp76KSG5GxBAxGZH8W5wvGE9vTJoTWmH7d0pAKJNHQjsF2y+/Lt4bERw - hbdm4cP7bXdmf/mfnCDCMTaeA7vdUxe5gNr0/wAAAP//nFrLroJMEn6X2TKJXO3updCICEhzE3EH - iALCQUAaupN598k5/3ZW8wKVVNV3SVU+3Li9yy33nsE6EQbi2ejTLvji5pr5KDm1t9yYfv9DCxof - +LxoeL5gfn3oKgBCUBAvPd9c5WetXlCk354SMU7A93G2F6DARabkXD2KLVAND/3rLxXwn3//H4kC - 6X8nCnT/HPra/XDEWzTMDtSrjRFvO1ut/Ox2BjxJt4HqMG2mNbk2Doo+cUdImgt8tDo1RwEWzzTv - KZjYJ44rdFJanZo338KrfLfeUPNCTkh5GTG1yq4BoHvUxP0GS8HZIX/Dk8uuC+9wMSn5/pChLpSB - L+AiLvgI4QxftZERw6E3l5u3Ooc/zZJRPccnPGv0K0JyVhpiCbYer7f5nMOim116kfU05uFXMqBf - mTJ1s1qNqd7EDmqFWCf6Jqtgi7f7AqOWE7+6ZBZX1KPYI9v0TV/p0DhtxyhoYDU8XBqZI8f8oS8j - 3I11Rsw3c6etHoQcbpVyokbdhoVkDncV3mIUkiiLy0n+ON0ebkc1pQHFX7zenDWD4mP40MCaLq1i - rp4MwqS3yTHYEGazkfpAKUuX3pPHAlhJbiuKhTyl5EFvYC37YgVq5ivEdRADi6suPnyE/ZUe9+Tj - rlUxHiBZRGkBeQ8LHjRdjobqiSkOgw/49iisIL9lCcGTvWtXW3Mg3HyvJ88z7DHtL/sDbL61QcJo - 8V1Z8j89nIT2hxipr8bf3wQBDNIBU/Ic+bRGqpLAkSavb4asquA3I3NgpoOU2tvPw1VCeXxDVEgJ - vVrZUEi54eYIdM/a31vH1qW1lqrwZ3Nf1BwTLWZ2FJro9u4YcWIrwizfVAdph8ORXC/fuhDlc7Cg - 5nB50hOZrXZNtYevHVUR0UuTeUAsqZ4i1n0FGr1dl89yLTogsfcHWpzSA5B3EitRz0lBT3cLAyar - Lxv1/FLQyKGKu7h6LAOhr0riyi+MOQobE+UJ9GjUB5QznfUJ5LHeUHK8U7D9VE2DQN1gQuTqwhVd - N2wU+9qd3sz9250tr0mQQGLqq9exdWftZuRI3hKdOHNjgVVJhhRW+KlSc4Ac8FBuepTApFh2v/Xp - GewTEIu1Tx0lORdsk14exKFv03MKDmArZl1A9rgrqc4B5Nty2o2wJ3ZFr/6ccV5r6R7h98GgUXJO - CtYfPwxga6mI/cI3vt1M8oaAIofELy2OJQNoPjyVuUwNq/O59LyoKvSeOSOWHDCXX5dIRXClDrEG - +ep+1n3LUHrMJmJs3p6z3IkHBEzlQO3WfLli1/YW+tUTak/AAhKJTBX98ZUY27WQ4xwYsDD2IbU+ - dzdmhYdmuEvymvzta/MtKYCnXn2RckqLYlXcPUMMRmdyGrwbZvvw00BWsxfBSxwUMt37JQzLfUZO - aXUu+LO650g5j6GvMTUHSrqKAkxCP6TWRY3a1V1BBCUcVP5+kgb8D5/fQvYk0e7suyyTqwwukb8j - xuN+dqVHNg6wjR3F35AaxOtPGkWIHPemj7rDT7Fh6btANahtWnJ/nJj7Mhh4cTmiJklTzoLWWNDo - vXrqtV3trq9eG2C6XEtfQ9m73bAQB0B+jwKxvNQEmzDZDjIXIpFLk818ZT81g/aISupLa19IUnry - 4b7Ou6UOeBQrc5wYaLd7TSRmsjUxfo9TqPtuSILa0vg3LoIUaV7Mic5gV0iWdzSQpKoRjXRcu8ot - 8g8wFluf4sl+tj8XGPhIQBeRYLudirk3dzakj91ITifxVrD+nS3QELILiZJ5bJnd+BA+mHKjDqjC - dtZuTg7WD4mpN95azmDaDShfrdvvvE6x1DFdRo16mMkzUUOX309JhOzznNNiy+WWe3wc4V70nkuv - PoWJpYEeoZB/Nmqup3fLnpe4R30o3OmdhsM/eAFtBzeSPTVS8DbVFyiQkP76h4LZHz7s8pWQ0zh+ - XF47zELht9HosXJz/FGlxICmJ3DieEoHZI5DASpSjqiOGcaiL7whFOPzgRg/DE0rH13xb38kIuW1 - Vdgh76EElZUSWx+LrVTCHjlt86HOdZnbdUxVByW1zOix/MwFU4Djg9DHkOKT7E2rlM4vuFO8glTO - Ohfbho4y8JPLQo9mbBZcfF8GyL396CvtdXW/UWDP4E9vAtfHQDHVfQ4HwH7IQTscJkVWDRH+HDPR - lzuCsTLs8wHeEX+SXz8sPkg/9+DR+PKi/ulB2DsWPIJDSq/auZyUJzokiH3vX2Igc3SlBvIeDdUD - E3OHSaGUO7OBpG86X+1ib1qiLINIGUFM3Fbe3FW++z18jsxYtK5CYIC3VwDPQReSsDCEiZW7jwhb - R+x9mLomVgTwGtDrqF7oM1gGziun9Pa29LCI5YSvaT0MYYIuK4/J0YzfsXgEJxvpEnySIG7EeAmX - PIEHnR3pg7yjdvnj768/kEgRr654MdXX/k//S0MowFaYXYmktTd8UYd3oFwSsMDSHyH1h/QcS7/z - h5fhaVDvU+ympTY+8j7xpIoaroQnUT60Dto+yUZvZO7bWUyMCu3No07//IOJ36BBO3SNiFMJdUH/ - +uuPYUoCJ9q1Izr/XoSVKZOo+2oTdzpVAEOWer6wXQLM57g8wF//WJRWtWI2BFoOfzJvoVZHWvd7 - uRxW6HPLIUfVObqS3byDP3zScigHzGg9GQiHnv3XL9+shL1R/rNaNNcg5mKszxB24ihR42QfMR/u - 5xEO7udMD88ct+u79g6QfSryp9+crvMj/S8AAAD//6RbSbOyvBL+QS5kTrJkFpmCgAg7QERBRaYA - +fVfcd7t3d2lVZYHku5n6j7SN9Gu2LNVHfDfz7VCSs7EOINh6tAHsmJoNdHk7/UExjSxVOAX4D4z - Rw1HbG4vEBazXGCzHUew7PcBJbe4zIdXc9DGy1UPUEp8EWvYMYb1wzQSDAVLIfbtZxXrySc6UNIe - YENbJkDvP8eCh7MtYG8e7xHlXkoFvhoLMT7FTsOvSKj+6mvvR7vY6HpYYMb33QyU6eBs9CVLiPXB - QAxeF4rlygELXrF9xcbp9qCrctk6FLBkJGfBwoAH0RzAsT2XRFl4peD4sgjAVQKKD3C/RnPJMQFM - XjPGmgECQL+fewX+9IKsPo1ibS0tgGJ1d3HQiy3duN/XhFOfrcQzrmbBvFAgoevlEGC16a7781AL - yPnMEfP17YZFbYkEL8OtJq4s8pQI08UEX42H2L1NF7occjuGOx+T84WNI7L3L7Sb+key+6QNXHMn - sSQFi06K9JlGS5cyCTxUtoM1z3wOW3qXILz4CsQn6WwX25qsITin4ElOdnmhO//1cNNLOIO+o/Rn - cloL63Q8zaDhVm0790wNi9/wI9rOP1ydDj40c+NI9Ps9bH4a+x4Rf6qEeXmaGViz4zP4p698o3+A - kSkXFSo5F5NT2kzNfD0VI3yxbOT3y1mgK07THBb2Rybu+3IraKyXCXSFW+tv8dgP0yING+Tml4nt - S6sXc2iyDEg0L8T5U3sPS3i0OWCOUUq0e6hE5H7abEj1NscVHzLOcMKXEUp+25MH02zOXz1CprOj - +dCYTLQZVHfhM+ldH6zIpfwVDwfx732NQ7ZGJOveoZTMt3KGaWAVNPYUAS794uDEbMSBOo93CbvP - GmNDHDc6q+V4AIKpVMTAxges9gQkmNk3fWZwwtGlkgKIgrfhYBO9HWfz3KaE2hKt2AuY00AYp3eh - 9j3WxI90CCjtMxuGzy9H7IIuw1IY9gKnTxnh8n3hi7Gpngeg36o7Pi/o6KyeXriwtbbB395T1myQ - LVuYimNDghhTjXr3XwhfObGImwuvZkvWMITEGHL/yDShwzd8PkOGcOp+3m1Bb83TQq+D0+CzzrwL - Kj9WE5lzuRL/8PUc2lfGAuxQsogLwM+h0MwreDq6C3Ye57DZ9C/8QFNxfOJqz82Z+NiRoRj0H3+j - p2SYzKlSxQj6iOx87myI53rJkbIHLvlVBpzRVz1Un5MzHxPEOjN5NjLa9Ss5KYRG7x0/YAONDtvO - mjnkSzQVISorJDn9TG3ajl8JZV/1hLXHr4rG8Heu/vQO8cVbGq3yw6zgN09P2AHec6CHDZiQsYYE - +/huA0Y59ync+W+GmlQ7dHsoNXwkRUDOx0guyMyfTGk6bA3BDQPokiykA1NfrP62eE305+/EeZw7 - 7PKrTNcbc+nhv+cfiRfRR3XJ0QLSkYTm2wdDlF9HSPp+moHUD8Wa1GUlHdqLNsND0wzTszuk8DjO - 6zxN05uu7LePYezyFTEvmhYx8klOILBvybx6bl1sSG5DFHOMgeWtf2jLAUQJHFUpx84rbqItaOzx - r79n7qWHxfgwuh5MUZfNR3m8FFR1sw0IiBt8/rVldLjl4APN/HQkyuf5oNtfnlBTJiSJ0aoO803C - ADk6HYjhjwKY9EebwAfIG+z+isewFcPY//kngsnoaH96BP5ikmH7ovjO5k1MCe/908TBH19lmHkh - M7Gf/qzU/LBccdJBbk0UbOodol0QvnxEcP0i98FwCuYdCS4cB97BRrQZBUO24gXWdVBnHi0w+k3j - poPG5j5ElUo8rL8xdmHTrCl26uzU8Pmac9Di+wfxW/fprMxbDpDx0bTdj45gfsEqkJhrnO39d4tG - UYhymNxngK3dL09Dygho97/zRxDGZnk7ao/EVezmw1NZok38viRo2IIw81PYDOP12M2gsPIvtgbw - AUtrdR84HwQFh4mfFkteuBbMdN/DRmpyGrn/NBvKLKvMlDWo1qnxJqOP9/2QszB8tLWfuRA2TCFi - 1RePxZZHio/8QrzPbfBx6NL79geG7/iE45wVtH/5yp8f0Y4822wtv6jocLaEmXcOj6EnKVuB/feI - cYmjiGqmH8IRPkZsYsQNvbikJRSw9CEKb3rDailRgFD2A/hfPyBhDYBX29t8kO46/U2jpPO9ffOw - MsaOtijnVwrnc+Dhy1LR4vfntz6nkpAyas4Dx12cDh5LPSCP7E2G1QVTC6RMj3Fk7RsRWX2V4SjU - PgkF6UkX4AYdfC0J8ueBhnR6ZnINB3WB+FKHdFjQpVwguKkZPrWqD3iWXAPw8Mpqzt+Lrm2j4yxQ - HT6Rz/SxWKx/fiaznME/Zm88DKs354CR0XdeYyQ10x9/yW+3xoVJ+2aVl+cH8XybYDmUP83GWZUJ - 0FnJ/XHHR6JxQQKjJ4l9eg+fxZrnMwerVxzP3K4Hfm/luIAf8dS5TQZC6WGjOpIb7UT+9Mui9CCF - f/rFqyOLbtos6+hPT+vrbysmJa99pME4xEUsfSKa6psrvh/D0+/bnm+Gn2aFEkMYFVfSvQX0Fprq - n17EOEo0jex6C16ZzcbloWma8TobLfxcRX4+Hq9xs91/XQkPy8HE592fLfLJSkApMwnO9vyC10wz - QC8yFnObcXLzz/+Hau1h7fbgnF/s3gJoBzOZawxe0are1QCoevZX/ybll0iUwIJ7mah6Hzn8X56I - FqbDKheeouFZgwqcF93H+MXqESs//AqeTkUzH33Tj+jxoM7o3TvOjAZjiIa/5xkLt8f5rje6ZN9h - hCo1/vnB5QdJDt+n22UWD7fS4Y3cqeB+39gT3ZPDP1W9Bn946f1s0qysmcWw/eh4XiuPRIvxGkZg - 1r2DrXiuKNW/twCWfgcJ5pVWW6PBdKHqhwR7Ms7ATzYbGYEfE+E93wGU9d0QMlnEE+u7bzwHsFtQ - prsevpPtDLjCRTOUbCzMY2DWTqu8DAalZX7EZ/z50q1HuQpvtPRxBY5hw7bBwsCIPCUsB0ZPl5Kc - Y5h8fhMxnolN+4LJXShg4UMevjkXnf1uelg+ort///PvB63pkcukJt7zv6jrfbVFYWTPxC5rHDGn - 5Z5DKH1CYv7h27Fs8398pr3L3zDs+gsF75ODTSnPC8K0Xg9rMemI66ais4xawCA4EG5mb3BqJqYM - RzS9jCfBZ/hxNi5bfQSlNsR//NDjfUNuWZ8tdnC/FuvtxdWQbq8HVvlLBdbHp4yhlEgA2683LZap - fXJoTceAPNT1Gq2LlqXwPAURyY+dCOal9XS4OID49K2BZl3aPofhIRSISk/csP3sSQJ7nkzs6FFr - 9NB7NRhntcXmzfpGS6LaHNz9yMy+60jbQhNxoGCcyP8w1yRajOxg/8sr3MCUHT54TSns785pz2vP - lGmP2wFt19OduC4gBek6/gX2vA/b19kd6D4eRSRM7uQsLb1GeWmU4KUUUmztfpwz7yoDNZiEPrvn - gfyf/jos0CTJnpezIK4lUB6WHKfnjYLdPyTST9cFEsvoVRDwPufweis/OEtSu1kfhpNIf/xmfA/7 - BsuhCMWCi8e5Kd0YbFQaY4kPin7mz/oY0RPORgGXMiCBcO+LblxRD+qZOeBTFAkDua/LiID7Ij6K - 5nzPO7cYPmpBmg+ghNFsxvcQfM6dhU/Wk9PGn1O/AKzN3wz2/GK8XN0AelJX+JsTTtFkX9MKYjF5 - +OJwkhvOjK8haK5xMHP2RR7Y7idZkLF+CXETR3emcm5lZHR+6SPVZofFpfcE3pxBJs6nchoOi7oO - 79eG+uKulze1bCHMytedePoWaEu+GjWc2zSdj8FVor8g/kC4oI9B5PqkRqt2siQYm+OX4OCa07/8 - Ap6f3vaHJ3THhwTiyo3x1R8FuqjxpsKtgm+sLaAHy3NTwr/6Idhq62YTUBtIp9j8YD2L7GLLf+oH - 7XkA9sFxa94bkWS453k4P4+stgrJZII9/yLu7kdYBjUzkIjn4Pt98KPtnnoJHKMaEBmDV/FPz74v - DMCK59YR9e7PEHGvx5s4xatuxl+R7BOukmK/rCRnuuirDvb5kI/I9gOzKPspfBez5n86jnWWQVBs - VMKvQrxEHIZVvdsB+Jtn2GI2DRMWnzliP3qKT9m+IVzZ4gv+1ae/5wHELKcXTI9sjl2We2ur9GRG - uOc3/nL9CsP4gFSGwbFlyP3zYrXfnp/CTnz7/loOSsMYGWeDI7qFZJ/30PUWWR24ldHTF27uAt4B - 1Kq/eRJWindMCdMaHdJc/uoX8W3Rxtqyxz/9P/MR54Pdn7TS1Yi8mex6dwmFY8J/2i0iVvKTmj99 - B9J4CnB0e3Dav/cvfr8flq9foSFUhT3kOO5G7KdmDIwfWBZ0f1KJleiiF2zVbCGqDcnzxQTUgAJt - SWElGiPGtdQ1K1OGM8oLQ8PZk3fA+IrCBO153T8+2axVVsHffEkeD2eHyQvdgg+lQDOyVZ2OVCUx - cNxHjj3VZhv619+B0i/YUQMm2tTomEJ99liC59gYaDVMrkRODCX3eUTR0lRPCHDlx7NgOgpY/vxx - +tNVn40f7e5PhwoUwsvAHtnOlKot68Pwd3nPxx2v/92PhfkLUbI6ACv0r6k0nW3ir+7P1sgx1m24 - z198qWMedJX6PoBCngfYNc9uQ/NwEMCffjLM1Io2Hqg+ODf7huvffdvvpoN5uNB5+wQEUFkoa8hw - 38K/lMNz9wNVB6Hbe/Oe9ztE1VofbmZ3/Tf/ooxRzVB/i++5+/pdscryo5YCdhqx+YPPYT26mQ8V - Ss258Y3MWeJrb0vH0LPI3zxtnP2uhOIKOuLs+TXxYOrCfX5F9Olha7QLSh069xMgp08tU3ZpXznc - /Qoxfj0oVqbMR2h45oX4zc8e6PP3kyHpHjE2NO/WbKiTl7/z9NGuDxb/lH9gzmisx/NPQtc6yA/w - ZT0JxoetKra//OWP/+xZj5tF6WkKc9UO5+NhO0SLI9YMiiXa+JX8lLXl7MkuMoom8znhcWiGbBYW - IAHBx1qn685o5bEMubY7+HHG1Q1lk5X5Nx9Idv4eTF3V0T6f3POsT/R1e6RDwdQqokf1Q9vmh3wQ - hdTliZXflOZvHgKHFFDsnR9zQ010eIF//u8Oq2IZstD/01czF9STxvVX9gXfHOP7myAplKFOKUjm - +V7Ni/o0IiZ2mRYG8yHCJzx+hnW7OLooZWZMZJyOzrr7CXgz6jO+PalazNd58WGjvpKZ71ze2fXn - DNjiM/tInRy6nWNqoQTID2zMyxfQfT4K4KM84WwVm2hmCjaFz5Jqc73nzys0vpwkB9jzB5P2A/3y - ywcelfaLH3fC/81zNyiiw3mWGuvX/P0etPIEYJNUuPibZ8L6Kack2PqjQ9sV+0DUUINPF59p/vge - Vt5Nnuu55sB+/hwyLyuZt3ZWnTUMrBnu8y68520De7mX9d98bBaq5xl0p+Wegv59T2fuLtnOqgt4 - gf/HRgH3vzcK1GP5JB7lkbZq12yGlzT/+GKXb3Tz8xoCtdkqYp6vCuD4pzvCvEV0RtpqRKzhWj5C - FH7J43b+AvrW0AytXz+Qc/JDdL26uQCO4nYmpyQVNXqeziVASXgk2uX9BotCnwESRqYj2s+/Ofxd - aSxA6P2CzYr/Rj1p3BbakhfjxPs4w1Ytdw4qU3ciYWPKA+cb7xTuf3+u08UG/GT8Oug0X5nIuUmb - KV3XFgrn5U5y+JWK1bXlBMHufcOy8ZucdVZKH5b9eiXR/fJr1sPo9lBhVYyVdWa11Q/fPmD0SvSX - 53NqNudRq4h79s28WOwNLOf1mSMo1jF55LoF+Ew95AD/kIMzNtcBixO+g5NuZcS8TTeNiqVjwxkP - IfFzI20W/se0UAtfE97fX2O8Iv+AsHpBYrG2TZe54VwURqCeEb7/nPU7iRzMq7rD4VRx2mhY+47w - OznNxyubNZtUej2YdDubt7m5OrxypQmSjdojt7/zF0vHAte3Gfos37qUf6diAKpYYMjJ44xh2ygZ - 4SqGFHujpABGzz49ej+YBedf8QfW/X1QhwDvi2zeggmKCoTAGBPiOP4BUP/xeMG7x9bkPP5eDj8Z - zw7R4L6SLPk2BeuHk4/evC8Qg0aaM/1wu0jevBrY0VY7YqGdLdB00tw/xv5XI9fZ2tDZ6TC+eZzR - zPVrFcBfvZpX5d381RM8XFbRF8ef6vAjahIUuk1KFCXznYFzzQ6+8zdPchv2xWLcwxJwZfYjXuy7 - zrJdIhc6y20gZ8bkhpHlVgm6JRywumQ3p5v9UoYsPYB5yW+oGBLu3cGDTEJyRlAolvf94qMuH1Xs - xp4ScbEbpEh3hJTE8ufQjBKOYuR0qCHmvRGd6ZQGEnKjwMXJJ3wWDJEXE/2u0ZG4Y38CnKp/THhT - 24WUzAkWW7/ebZhZaYhTezAaarOfBFSZu5CKHUk0FWq8QWvfzi7Zom8o/SghUvDhQaxchHQ4NOUG - lS7zSSKRL6Xt1SjhzbtxxBzEiM7FYxv/6gMHhLci3lN4E6RZr81H4SM3nP7ya+CaPwe7iRgCShpV - Rx3kA6zntyjiQRdxCKdC69/fy9rQiA8DdF+5n8/ESw34z12b0Q0cCmL2jePw5FvWkC91FiffznaY - 6DoLMFCDgJyzlx4x0Dxz4NLnNg69qwNYwZlkmK2gw84xUAZGOTsvNFgwwtmOL9sY/iqQKBzCBhdU - A2/9rj0wJfuH3UfPDpu0Kj669iXrX7NLOzBDVOjwg24+Uc+66hBy2RLIGv6I1cas/50v6vhyws5v - 4YZFIWwJ9dunxDKvRQ7PZg4ErrSi+XiJPhrfHhsf1ZBtiXaYVYePgzqBLGYscsuFzdn0l/mCW9AB - gh8/GUxW2STIm6kxX3KjAV3fvhjwjE3V51O3bhaMjAqGVQ2JUhTjMFqq6CNMbjO2U88sZqFdXkAs - h8pnp3lwmKjvK/jMGxdrgAUN5SWOQ02u+eT8e6xg5Y5xCFcMTjj9ij7lWe25oS3owfzNIsmZlldd - odFXVazb+RusgxkFiMexTTTTHkD7Y6UPbB86moF10RpaPZwWnsTjg+jPzStWfhViKJy3Ozb6Nqfb - jVNqSK70N7NPKjgE3lwd3lpymLfvrS/Go6IuKKqumy/y7NisB8UK0XhtGP/dpIE2jfVJQveV+ZGg - dnTKijjggGqa3IyEz7shTyKbyJjGjZhmLtOl6PoPbLMxJ49DqUZLudQMvKaPC9aW9Fss/NSl8FGH - Oj6Haawt0VUOUG3Nm4+G668Z+alOQafd3+Qx3B7OYvJYB4/H1cSazt0HltMsG1KahiRgbs+GdW5j - AMnzMc1bKDVgcVW9RZsmBTgVd0Mv3JYRvYeamY/vxR2W4fZ+oUoFgc9M1eoQHYQqWtGHEN0NToAV - ccqA39W6znBIxGJ2xbhDA87u2PczBmxi1C0w7/2RnNYjAd9fsVSwfp6f/pqpx2I9afcY2m6u7v19 - AnwAjwdw5z0X49v5BLiBlSEylNNGrA2fAXnxLUSxC284RWAaPq+sy+Gdxy62SGxR+g38FEZrcSDq - 92YX7ORnNVyfwvivPpjow/Xoj39vX6HVpmKUXtD6dQN+gKcSsWdWjFHpy72fus/TwFJr4uCTtzxy - Q5IKZjV7dqLPctme8DkD20e8C98aXvzVEE8ae5qojnTmeiLJju9LbSQlJN/Vw/qSbc7mKyFE388s - YfXCPyLu2j9KeIelga9iPoPFSosUlGAOiTUdamfto6MLTXk7YBP1FmC6CcYAvmyPVIf7/j/y7pNB - B3kKcUW5qdiUr/cB/edACE7nWuMOS/z5wxty1ur3sOPtAuXbUBAnTImzaTQJkSDUIwmu7Jn+45/9 - foj1FX0wNzypYAu3t0+D2qTzUHspxDf0xSeZo8NsSUSCObxKOAtPfTEFVCwhQapJZIvl6aRlrQXG - M79gJTkdolV4YheYx8adD2HXOqu1eS+QV6+O+PPr4Wzv+BdDdJgmkia3c8SUI6si2Xh55JR/rWKp - 81BG2el+JRV5N82aHK81nAKgE4znCSwO2kzkLNdh1ydCM4ckqRFIFuQfecWKliftZLgIjewzn9CM - Zvq6HRC+uzWWzeMVbMfmEYBVxzpRvFhxuCoHPcwzyyCP+BFQ7p48XSiHoYCd92MEcwyICy9p+iGB - n2O6lCMrozDF5R+fFowkriasN7Mi2ngWHGrEnQ7X/NViY0xfxSInMoNUeY4wXsxTw9mTA+EfnzlA - 2aL13uIe2j9rwvcw7R2iEiWG8zZVBPup4bDYUEL0zJ8uSeFbB+xNU3LwfcTbfDSR41B2u0CYLaxC - jNvoFrwxlyPU2UDH+mhZxVIW7wDOYxj5oo1yh8bHtIZLgKF/3gJmmPZ6ht4vfxFLe16L9cioAtRK - 7PpIOc3FWrK3HOVepZDTVzoMa83oNnzyEY/xJTK17d07L2C12UaMJNGGiV+FBIoBKvznzUXR7Ces - DXe95x+Z21rM/a2rQf9ov8SG3K+hXmHoUL1sEvGHJC4Ij6VKSlTQYqvpj8366rcOAGF5+uU1rIv1 - jGcV3ofG8jnBCMD+WYbNewrJnZ0e2gy7KoEP9VOSXQ/Q7TK4pvT9jBLROuTt/Xgr4fiyWJLteowx - TksAP2P49X+nlUQLn7YMZJNLhGV/yIalXDoGplmn4QevWMXaZL0AT0wLsULVpti8Thcg/fLCvBC+ - izYPySXMQHUn9iteAfmVD/Ovn/zxGsrRoh/FA9QXzcGOOPyGmSulEH4ziHH5rvmBviy4wfPpjojJ - iB9t0Xouh/E7M3Gai+W+8edxQL64F3KT3z86x0HPwEf8kYmH7z9tLthYgNkqdliPHKvhVzq58PcE - MtGontI/vIfTk1PmVhnuDe1StwP6ATbYHcE6rP632wAVNpbYCFgNZWLxJUlFoJFHI/XDTCrJ/sOX - Ge56kr0GUwz47yZiPQvjggYX0AKnOzZYD56XgfqL1cOBH1/E4HzdWcR6MEHEThIx7aEftujeWZAp - WBl7Rc4Nw8GTWli6UMPBpa0dLubLBfzp1fvXlCLaBFRGObxJRD2QlVKJ5TeoiRJDdvzVtojaCWy3 - 85lYquMO3O10SeDMwxtRi8Ec1ihLXbj7C2wz+4aEUNcfdHZ6PC+GlNDBX6wOPIcvnaWtnofFUlcf - cnRdibu6oFifXu7DPvd1EmdKP6xc2sew5VOVKEf6cSgvpT70701MzrqaFSvNUAvUZqlwya1RwSh5 - FKLuTA3sMebY/NOr+doyJDCPV7pNY6XDHgUYB8+n1yzGaQnRPXYUcjaS00AlNrNR4QU8PiOYRguT - ZRXY+50E9/AQrS2vJfBtlT98XiSHzjGQWzieZxN7X1MqFqJUI/jzO9jTOUCfuV6Lw+E84rPq8tH6 - 6U0Zvq3qNwtmyUXTvfmpgOqdhqOzMYLlfHJ6INvS8qdnm2302R7qs/bBtq6cm7V9Pm0kTAATRxzO - DQ0utIVQfMXYzDvB+ccnV9CUvnBV6mG7XDgdzmffwvY1PlJSoqCTuk9uEk/4GA0v/KzqH55aTWc4 - ozwMKkT08MX+zb41NHouJtKbsN71Pi369r3qyHeryb9c3j2YPWRV8HChIrGyJna2p5zn0I0Kjsie - fnKWQ5i/4KYfeKJKWKDdAQoh3M+DXEDLAYIMGMPW3zcYffAb/vAHRCanENn4eQ43NqKP2mte+cvO - h0tWZhIUg2OBnfN3/Ie/YLOvOb4b329BP6fnC/35T01/R2ArW8+CwyhXuKw0zZmGRVThX71fXHkD - a/hhLBh5kYP942vV5j/8khJ4IjsegOlohDPY+cjnTGcdlsVQX3DXiz6TKX2zii1Q4f3tdXv9FtFi - 47MAk/VHCRY+9bAW5tcHrtRH+Izdc7TeZm+D7nzAfjyIKVjf4Lqgw5SE+NSzj4apfCVB3/F0x+dP - u9AdHznYxo/nfEgrc9+Au0GJd84C3vlQG0XHkiD9sgJWQJuAdalgB/MsIvj0iVyH9YTl9cenxO2K - tNmmBAigfj/uRFWPC11Cwczh+pRGbPCtC5hZEDdIcXHGpg+ZojvAJURhy93n44t1mlXTzxB83scT - Np401Whuvnzk3KOHvzA3ZRiYI5Igpc99I7NXKXOLox5y3Wny6WF+aZvrpyEM1EwivuP1Gv1wlxGh - 2+3jo4zqGmdOSwueWfnG2D6zlF4ykYGR8rXwH/4uqneHcPenWNlhfueD6h/+ZcLVAFwyO4FE9V4j - /pl/FhsaVevf9wPzyNIxF6UYviyXwarzONONMHEMy2ye/eNwbZsVJxcZNXE/YDk3o2G5PzIdytL0 - wdren/Sei5b02LLQp4yUU/KieQ0P0wPOyz2sim3vBxBI8wH/+etP93YryL/yZGbHM3XW+P7qoWsO - Dtbd4AsWoFYqfA7Vmci8RjVyc54hCO4wwtGt+A7rhJQZ3uowxYpnrGDVdOWA5DAQsCerJBofyccH - Az+/iK0rv4ZOhcYgsfxVWJPjXtse50aSJEx+RO+bRduI8FlQ7B5u83G4PbTtW0sqbPnm3/1TdoV3 - E4iM8plfop2BJfocOiBiieKzcH0DqsgXG+bed/DF7NUWS+ipMkzcMsPXAnHaekubFkYmo2Df6EJt - tNtWh0fmGWCzs7SCuVIcA9cRdPxQUupsA2tBlCgMIicN13Rl+ESG8bswscb5P42ksWlLJXfP5mWx - wmg7d+ILfq9dhItBb+nWfm4z9Mz4NnOh2f4HAAD//6RdS7OyMBL9QS4uL0my5P0mQUHUHSgqICKv - APn1U/jNcnazvGUVdUm6T59zOnR6HuIaw9yWkwk+d7XPvrl0A5u/iOFBP+jCPf1iuOH9xC00A7N+ - UNeff0aDal9vfF3hUBGWA7Fhw/XLCegWNOTuQ9Sp9HJ6HW8evBb8jern3O75X33jDc8kZvYJ8xlX - Fw62kI+IdoS7fDocdgYcmmLBoIw1f5scFsE0KK5U78p3PIY4m6HTjhMxx/srp6t4VUDaxT3RcM9V - I7vy9Y8fElWjAmBBbAnw5rQ2/emV4ecP/vRD3KIxp2bSWtCfsTotYfLSB3m/GOiblRJVjNiIF0FX - POTzTb3NwLEr8Z4tDogvf4Ro0j7yxxs3z+BcNwlmgKn+tB2uhtrXV6nxaak/j7wJYWTVAQlcsYkZ - 2FcNcurIIec3DnW+UN0nVAl8UM3Gii78VY8j7NfHl+LX81iNLr+k8FSsCw1FdfynZ6T5VPDTLB7L - eG5HmMIu6sk/P3HzY1tYPRNK1A2vWfIXPZFykxPqmFbY/+M7Wz0jm/+iL0G+m371lQSqU/z0vAUi - axQI7lMunx/9d4VWxbtE+8V3R5wOju4TUj+VWn+tnKhF7m0QaOGHnr8EvDHA4fTisCSZPtviN4Xc - 1rK5VXsDiPUpLEDoT/K0c6pY5+3LRQKb30FczkqreTpOBfj5QbGAXZ3+/PEN36hffT2w+ZcFRPJ1 - IV7aTv5kvtcErmP+ps6t4PXl9EIW2PgMNZX3BMZ6lSSopA030UcP8vEqpzdY9OcdNUvtE2/5o8HH - 42zhaRc4Oc9HtfWP7zgbv+VP71aC13DmqbPxj2V42pJcFO+YKFF0AFs9uMBPrgmUTC+aj3lerqjL - AoPE5r7U2eZvwOB4IjSK/2wgTi5nAH7ss2nTt4BVzSmARJYdjGJU6uOVZAZs3cXEzXkUfRqgugHC - Lg+JbnCVz/ZWuwP2FOdESy3cr3j2HYi/5EIxlzbVcrkJGjwXESPk8X2yxfXjGpYZvlC9EF79vLB1 - QOn8CSaGh0f8z/8cpM8ffq0s0Wfjb9nBj6se8bwrRn0UEuEJH8/IoEnsO/0CcR1ASpICV135zv/1 - A/j9qaD4kJwY96Qclrd8IsGXWmAJ9w8P/OI7UgxeX0vq1T9+T/3LXYmHB1xKuMXnpod5tlQ0NGTn - 3NsYvp/nasOvJyTFwSa+arN4MpOugeuQ7jAaIIzXg9KkPz9i4jc/Y8VqtoOe/zBpkISvfL0P/gXe - VvdODsil/jB9Ou3nZxNlm8cxLMXlCSM+yUkWvMpqCXJhgOqNehOVQ+b/W09Ldr7kYWOOsSFdss3H - Vsnmx+vTba8M4OsWh6nKUqdnmMstmArSQKOPZVbcj28pB3wg9u3e5eOraTFsV8Ojyol3wRzmWQ03 - /52ErnTw+eHxOqL25Ye//eyH00tsYCFIF5o9rhGY7ij20OaPbvXOrgaM79HeVm4evcYTZIslzxlK - 671DyOuhMVZ3qgYzJ3SI7+OCjfk5Sn7+E31sfHvNtGzz5xJ92vyxeH6axQ3I4/VA43vJ4knavT3Q - SXo6CbxH2fw3EA9Mpx3DkL7baiqpVqOosBGe+dGL36Zo/OvP4Pl+3OXzdeEMeOvYacO3p87yfS2A - e+O9SRCsn3z4G4gDnDpficvqEoyVoUzoer3tyeXhDDpNhXf3y0dK9tDNuenTKfATXTPi/736fNmT - iwDXIdn98s1n5V1KoXvJZAwvRa+z21GW4cycO8krOeknPPsezKFuEes8ivpwcNMARIv2oJ7bqL0Q - VHIBNn5CA+025GtzAfOPf+M2Wg/xErnb2edU2PDSqqolYsttmzFSUNXcf/zlhzf6FMz0kolaPg/W - aYZ/yqWYWJa2VRc7QwP4D46IuptsfwkrbgdDO9sR76fv6rsogM1/IsHmL1Td05PB5h9Rxz2L+S// - ocDYMn2u5TcuzfeaoqczrEQ51IrO73euApyWTpMQ+3q1zJLXydfESnEXvLR+2pmCAMv++If5Xvrq - mx/2BOnydGlS7yR/Dj2Ng7fFEoh5ELyqe58IhJG/o5PwJTcw5ufdTk75P0o05bT/9UcKGFBzT7ym - Un0Rz0qLBnewiDoGas9v+hRs/YBJ5F3FFwT7YoDwb9Kok56/+bD1F8DPf3P5rGazyr5HQIK4wjO/ - S3SW114Lf/3EckcPjKU2X8DCidmm172eu3v3J2SD5lFXeOhgeh46B/3w0QFKUAmAjztwBjDHlkAw - mFn3wohFiU0yzmzYwjfWEfJtJBJTeWM2L8Wf928/tvWPl7UPU/mJzRNx9jlXsaNTGsg+C4dJrFEV - L6f3U0LD6eJNsPYmsF6/QQA3v31auoPSC7511KDgRTnR5KL1mTuqN5TCmBB/FyJ/Ft6dA3160Kmy - +XkDloACt/4ejZBT5//iH/l9Se1xsvzpnn4DsKS4JWaaVtWAsGPBWnwNeLkfcD9veIo2f3paSqGp - +C6OGrDpXyxtfvUcnxoJoDT6o4b9+ovXnVu28Mff8V259tv7yKiNjIaevHzu1yQQFXi/vmpCaIPj - Vf2YNcA1uG0qj/pTXOAV/vx550kLxmyfD376hLjxoOX0m0sF+PVH8V3ZVzN6CAVknwlTrJ7bmJUs - K2E3nI9kq6/5HNHXE+RCWxGy+Vv8Vs/h7Z2teOvP6syJwQ2e7oNHDkpCfOGuji3Ye2VEtatK+wEM - YJXHD67Jr5/HzORpAbQ/e1s/OuybHUc1IJetT/OG5/v1rr47SI5vg9hBec43/I2QcMu/mLMfBltx - yjvQ+3rjJKzKoxrl8SH9PycKxP99ouDyWHIaBP0Ys/0p8+CBuSbxTumosyJWVxgPnUy9wGn0lZOE - JzjQ8ESsYU+rFXGljMR9J1GzRqnPfOXbwTBSavp4BaCap/RigG8qn6bZBS1b3kCI4MQlFfUre9HH - N9hFkKXlGaNEWP0pUx4WyJy0Jo6i2owpeXMD6j6RyS09DD0rMnrZDzI30sNBLPNVkXMOfnfbN8R8 - xPJlQu0AYjON8AI1TV/jy+yAEYctTQ3T8Oeu/oNAF8eAHLv4AWbZqiZk0vFJC9FzmHB26wat5mFP - VHBc4ik19yuMGIQ0MsYX+60PwKLhU7epv4Bdg2eKVGU7Y/mNRMDkt2tBv9UIiVFt+sJ1QEdYRbZD - DRnosZAJjYcUha0TPGqveHadOkPOZRcTq5dZtbp6D0FVvlJqqCyK1xB+W+Se3IxaeqHk2+TDJ3hP - jytRz0myTSlWjygfz8IkGKahr0/iCiByhhJXra/3fHa1MtReuj11hyVgTI/2F+hCw6dW3zsxB6td - AteOulTBZRBP8XE10P0ay8QaDzbgMk7ZoWWRduSOWNfP/g07cD2nmAa8ePCFPf+xQFVZHiWy3VSz - eL/J0HD6/bQXzoHPj+MyI9ZfGM0L55Rzo5fv0HSHb3Kpm9Zn+YcO8BQPJXk8dkk8ofXRArTb1fQX - fzNs1AZt8UP8YpsNo7zPKRiOSknJXy2xdZZyB7pd3dHLFNi5+NeXJdr7nE41RYrAInl9C37v54bV - s5/8p1DDj62VNMjFHswOUDqg89GOpp8SxHPUSRK8p+KAQRJ3YOZtKkG9Tg6TXMAgH21nKaFldUdi - EynKubPT7FBEjQfRHqrGuC4nBrT7vqZ20Ko6f5PqCF1dL6dkdom/JA5nocr7isQO2pcv+nJVov54 - eNPjo21ybhbvJXy/bjd6etp2vj3Pgg9V1InqOKkuviGT5cbMfHqvVzX/tz5lgAg9Iemlr4u5k+Eh - 5zLiTLocs/moeGjyn2eanaZ3vIQlHGAZKCXNdXeb6hTpN/Qi0YecpsTTOVmTJJgqxZ3qCUlifuyb - 514lSCZ6m2E2W6dhgL/43Tl/bjynqpqifeVNVIkvuS7sjIOEgkDuSZiFls/dP04Hywc8EAUvYs88 - brxByivhJN6OaS4misPJx11eEPVvz4HJSrsEovGukxM3S/k65coTSfw2Rbp8jj3TxGuJpLD5EFdo - xHjlaNjB4Q0PUytdvZwz/v5aGJa3lSjK6dsvCfyT4f0qTMS4ZXt/ya44k9kcRyTI22c/f8/bN/GX - /DGtZqvGfNrgCX4GsZ/mKuniafQPNRqI+sDgDI45l+52CniCRJq4FDUVj6dZgMuovahx2HVg3gtw - glHuzdQm0pyvC1YwkiJFwGifr3mL6FqgmvPHiRtSJ15fTXqBqXK7E6I6Sr/cnTgCpRR+p92AGp9m - maGhPV9DEpdvQ19WLsYogWyegL6vATezboK7dKvoH/FVDQ41ShR88IVanzLPZ4HsMayf60qug+fG - PBDfGNLbW54SzXf6GY4fDHOa+dM0IMv/vR98fRyL2rlBfHYK6xLpnlJT10eqvp6Ggwy237EktXy1 - mhbL4CfL78R7VIRNiOtkWHGopoTGT33ZJ0kJuQT/UVXoR7CG8NX+3ofeAvlSsVj2JrT3kpI6xfWt - U0t2PPjD63gVIFv8vwqCa1Cl07v4K/I5PsslCqLDh274zWhjDQLsTlVHtVoY+2+0za52xhgQF7yZ - z5zwBRHqNX5a+gfV6bG+DzAMmxNeu9vAqBO3GspdBdN79LHANF34COZc1BIf2WnO1ba6A30yMnq4 - yZw+P4w4hUzTORJGih4v7eeSoPo5r+SSBFosbPgJ9lxL6GPbz1XqBQz/lAbip1Q82fThVQ/tNCmh - lyKS/OU78fAfnv6ex/8Ns4PeVAmJceARWMljvaBBFkbcxwkGS+A/PfBU//Z4GYQqXw1ZDGCKIaD3 - vS7r9HCbLZS7GsZ8PXVgKY6BA+3RNWl+kaeYftl0gUVafwl57JJ8fmVKim7m6pLffq2y/U3gnusI - /vs7jT07xWUN70ZeTABbNhAUORZQxlsPqgLd0vmL5MigrohC1T7SmQjWekZRhnWCwyYAArlyKbrU - vURVwfDymeTuDjbL+0jO72+Xj0AcAygJGaPY+fvGrHEkAyW7w4EWjnCLl7uTH8Ffuttvt7B883U8 - igEqq1NIruG3YMIDkRpy50Alx+hmAzHmjBZu+EM1Hz/74e+qRuhzKfVtqvSzX06XXoP2NUxpFO+V - WBRV6YKc714hufQ8AtHsDgFaZrjNMOmrWNRtpURfFB1pdLXtnlOJH0EFy2fqkZeWi/PR8dApdWaa - pMjq55K1HrotSkWd6jP47FHOE3rt9SfxDcrALLjAAc89fyZuq8KcCauaoWZ9/W0zY29gGkBZo7NY - GkRTI1cXs7vAwe0UBbF/6/Xmog6eUm/GfUklsKjdYwWnj32n2kMtwXTUfQca3oRpAjxcsd27LpA9 - mSv1jTOoFrWdA7TVB3p47SlbXb2C0BkkbULfqwH4Sc0baN3uObUE0dbZIas55Jh/B+KjB67m6Sy2 - 8MefdMO4grmQzh765cutP54Ae+m7QN74E26cwe7n7qNj+NrXCTU2vBNCoa+hQkSD4k+zAOZ+3Bmu - lyOgIRf68ap/nwM6v++UePuh9OeTeMmkX/wY72X02UsOj3KsSRrR9o9dT6sqMWS9Wu40ZPzKZvHR - OlCVD59JUq6K3iW3bgenh/iHwWi2bNHgFcIyfOr0erzSH58q0Ybn9D59ejAE9CND7RBcyS++mSUr - Dlqv641qfBTnbNKuASyFhBBbjU/5mJdiBNfrfNvqxyuem+d3Bg/URuTgHCp/MU4LB0375lJ7xu/+ - 3/OOZ4sjFjjdq42vTuhxmlfqFNHFFzzytFApke8EkC3E799+3S/ncGJuPoJWrI47hC46m+oNv8dV - PkSgTYOeWudnyOg4FztoOplAgxup4xWt5xZW3F9NvQo4+uINGoa5VSvUdXKRLeO4rNBXvIX4W/xv - 6+fBxyeKtxOfJlhhJytQf50ILT7Nwqgw6Cma7vlCTVYe/F5yU0UeGmlH8fHtVNxJjC7w9DHv9Dyb - Uj+XY2dA3d/vqP6rX/TgleBxCnfUdHe8z/74/AJTrOFpucRzNQ/GXYKcui74FCWmP4PMvEFw7V/U - 3NlBvNWrAGQuZ1I1HwbGYlEs4JtqIXHQxfIFA3kFxKLlk9DUKWAvGJXIMdGBJMvJ8LlFvw4gFCWN - +CMQK8b+3AnuYB8Qw/aMeHQubQF/+f8PL4qLF0B7ahuq7AzFF8ApVODGd4m5s4d4acL5ifxDeyQa - zdtqLualhOf6pVF/BOdqjrHbgYswhZi/LmI+L6HW/OPL9o49e9Y4swW6TgupYa2DP91Sw4KvfZNQ - IqcLW/g8neFWfwkuL952UO9xgcK7iYk/Jom+/PL1+00qeir2T33+C00P+oqz0PzzcPM1qsYAHszb - nXiBY+mrYkUKMvhdiyX0OeRi++VLaM+OQ+9mx7Zb2JAENz5LT1Hy1mcPlwU62GeRGlBBFcuScAdu - mbn803uiZqAIoD9woNZFLLd4KBTwp1ZfEixR2C/Hb+HIsXbG0z6slGrdozkDD9RtE9uuXs98uXrK - 2EsionljnvfG5aHB0PdseqafF1j1BSew0D8fap7KwWen8z0DZ+Evwu037PuZ15EBrwKvk0sh5r1I - A81CQ/RSiS/ILF4V+6+Gy3y903tmXP2xM3oNcPrxSPT7TqkEaVILGKPPh6par1cU5aBB6znBRONV - R18j+k0hdaIjvUjXLl4HM3+CMKxPJKD7fT5mD6OAh3kSiduvM1h8veDgD4+UeP/M2VIVLYDZ2Sf6 - uJSAPW5rB0p8m/HwuvcxjSdcA1WOP0RtDo4/fw9pAO3F4kmwqGW8CrhqQDMFKXVvO69iAN47cCnD - jga2GYKtHmKQPT9wOmE0s8WjvAZDUdbwEtoHMCXlVwPed2UkuI0yG3j7I0Nr1Ah1Ro7qc9kXAnS7 - psPin+aDxYJ8CpeV76dRCV/6csKyB/2THhI9IVw+iso6o00fEUPTQD/a1nSEn52sbHjxBmNiukdo - FKZLDUeJ49FtDA9u+UY3fpS3GZgaeFtQRGOUX/SJo2EL4lkXqHFbl3z2jqcIKp+hImGnOLkwkiBD - E3+IiOMKx00vD5kc75ycXDa8ZOk9vsG+IQnxpcOnohyeJ5Q2XEZ143b01zMpW/QKF4tc+T8GPp4w - T6jdGS7NdmUEhrJPBeCeBkzUXrTzBaD+CH768brMbT/75Tbl9DSvRPOENl/4e6fBfeQO00iiYzUf - fSGDq8mNBF/81l+bdxbAH/8rzoz2K8frA/zp102PshlWQgpuy180iVQI/RnR9YZcpEhEMRMzFo2s - PKLzo30T8ldfWOcIDoa//bdPY+7/+CQQDl1NwomwalnnZ4byhcuodtrL/trdggFu9YIEWz1eBU9a - ocl7y8QjSdUFHFxaUAVDQItzEGz7uzpQfV99Yiqu14t47DEA9moT87qc40XeOnxS8Uq2b8Acxqrz - AwJhvwdEe+M3m8VraaGZfUuiGz2KZ304yeh4qU+ThIRbJfwd+Aa+p/sVy4/EqP7h6QvcJxp6ml+J - Rfa5oNdhEIheBZ+emlOggbbzMmJY3z+dHtJVg646P4huhBYT7m+gQCmsP+R2dP6q5TuhHTzXlUbV - Sxz13GN3x3tkvt94q5dsuNZHRTrv4ZEkcuXrPEBVBCsxXzH/3W51yuCAQRVMAQ3U6cq2z0YgfN/L - Af9xf3c2tK5Uww3/fno6X72zOf3q5SR2baOzlR5ntPvrMNHxNwHicEMYFvr7M/HKjP3ly0FN3vQo - xZeMbbdcXSGU892eKOPerJbsFDdo0wfUcga7Wo799yLvReNOQtbYvZhen+2P3xBvJ3/1UrvXDYp7 - /zqJh4Kyefv/ZJljJl4JX/ur7CcOeh6fIykuvqMzopoK9E9qSJNudphILUNBVtaoVKljK+aZ4WFI - mGARZ/vqlZv3zxn+9nPzL/zV2YPtWqJnSozeUX3uqJklcAZZm+ASjf2Gd4L8869M7hXq7Lb/SiCU - BEq1nbBW/UsXgp8+J274LcD8TiUJEv/KiPd2Sc75N8tBr3f7Jt7pLeUDbykJCn3HJvF47frRpNcW - BnH0pbrYhPH6cPUZ8on0nWYluFWbfzTJT1wj6pbB0ovOyW7gvKwOln3X83l09j0YyO+YONKiVdw3 - ft0AtGBEoxIe/ZXcqQeHw30i4bed9Vm7QAf27+OH6O1q++J8HbebST+faS9QIxekBR6BEq/cdDtq - r1x8Dt8G+vFrIKq97/q19uUMtH+KMIE1pmzLzwv68dEbNBL/9/c2U9AjSvy33SLVOA5YzQ3n5CLp - 1xQ63D5BSU08mq46/dyuHFSUZaX4g6/9agZLgfZ+TGhYsXe+xukgwdFfMnJ73fuchZWcwpNGVMyb - 9AtGa689//l71vHYsJl+9wIU3nVMLuJqVeLfHgfQPX9agoXsEncDKBsQPdWQmDEM+nnRDwOCtowo - 8eup74a/+wXYV5JSXQoXfchA0yAvOVWkKD+Cvu7B6EDbFHLiIHLVeeU5yf/4ZxiRafsyrZLg/XIK - CT5yDKym7jRQJY1KQrOL4zVYpRVW93ZPHzdb7YXa5BIErKih5FBa//Vj6tf4oNrpmPaDl7Ql3BfR - Ac+4HHLaOLMBEqXpMKry0F93xkGGmz4jigtaMOd2uaIfH0qX79Pf+KYMw8PTpPEnP4Ifvsm9cvHp - 44minHFpmsDi78JNEJiSPv4Ns4ei7A0nH0q+z3m4K6Cn7Fq8Kt93tTzFpwEjphBy2vhxQx5yBs8i - QJP4p/Vs/oZVho6GveJqJft8WOzgKbfD5idMF6VfW0NyIKdHRxqL51Kfjq6K4cHe9Cwn5foSBfoT - 8on8xXvhPPjLPbAM+GbbF0LlXc3HmM8iwCHVo142vvwVvlQHbv4v0VfhxkTJ/0YyvlOfmrtHXY2v - P3OAwuc+EvVo/fI5FaC23cppq7LEGNNgA9+5VRB/0//M9osBvDBlFJcfwZ9bXX6C/X28Uye+LdUv - f+H1bnnEOu69nKKjNsDFsmeqV7rcz+QiDGDzh0i41zOftU+jhX/tLSOHY2GyRcJpAceoF6l7vvT6 - 8pUmAW7+GzEnMFazrM0ypHdnokezfeVzIXcZaHvTJm6o5Tnr0PkIP2P2wMvFeMcLVFPv53eT4Lbd - UsbJhgP3r29Io53x1Md6DTOw8b3p/XD6nLofdYWCXFDqvS9ytfUrjiLrbhbxL33lL24TeOAsFQs1 - YzhUM+LnJ7pNTUL1seXB3N2zBJ73uyOxn/srE9QYTDCoVw+Llv4Gwz///LreiO4GZcXqMdhB9gXH - 6R1VUcyVY2nBMvgjxCGhlYtyy2S4+f2bHlF64ecXurvySjxutGLGN+YRrpcIUF+fRbAEbiLBGL0/ - 5Dq0f4wWoT7ATvliGmbGVafb+shhqL9I8IdNfXk8KAf8d6NOzLitPmWO3EFhX2rU3PTE4u/fMlpW - saeW6xX6Itf6TSa3fUJM7vgCm59t/PwXqs2Hcz8+XsxAtXfhiFNjuvkT5xX+pUpAjz5WKmGLZ7RL - OZ2eUzPO129wXyF+DphcG1j6Y6PCCzTtwiU//cqOz1JG1nKW8E7IpJiCXStD7mvnWFo7xBgREwNm - bnyglj32/YBJyEG0gzUxuVmKV0f8rD89T4tzfo6Z8uka0B+FK3EvONPnV+Yk4PrKGYb+cMj7SXpG - P70/sXEp2Wg8ugFcQlklzjX75hueSHAg+mMCW/6vinVRgNPxI8HCOlfbeklw2G+3wnZJnc+Cyzz0 - TU8v4pmroA9/5JL98mkC99cCVhILEaz48kL0VcH+OuVOCSs/6ElmedN2wiVuYENd/adn2dLYcgEf - CWeS2DD24Bs50QQ3PU4sYcr0dcEOBt9OMKi+/U6boJSgp4ITdUxF75fNrwDj3nCxRnhDF27HwxGu - xjkgus5P8Wpa4AJBhk+b//T1mRRb/973h98968xn++vXbP52FK9HXfdAn1AWDlt/7ls6uw7IVG6n - fZjF1WAd9hIE0wsRN13s/uutzhFGTz2k1t0Kc3E+rTsIbl1FrMvDZf18fa/w7bQRcWLOZXx5/DzB - bBf5hL5tpK9/Z9bCrR82rY+2ielbaWaYYgUTZ3xX7Lc/AJ1ta0poc+5XZ88U6J7fLd5dfMfnxGtp - oD9ZbqfTxoenv0QR0Kmtd+Sw+cNzbj6eMr1hleBEy/1N/yU/fUWUyOkBo4STIdIHf6r0Rvf5X779 - /NHHxtfnjHMgRLp7pqpwK6u+iNUZ7k5JSVw7jPoJW8cash2bpr/D9RqP34nfQR6SmPr6Uc3FG0oG - iDrLmd6ZS6r1dtMcpIs0mMT3E8UbvkU/v4NaZZLkjAvGAmx+FjEn/67TMHOLn/9DlJVy8fBQniV6 - KFJKHvPrE8/i4+nA2+BZuK+tUWfK+5wgVbqE0/6CZf8b01sAzdWlU8fQqV8rKaphVrjjNBL1k7Ox - XVK06pcHcTzsVcuuGySIXXYjP31K/5ZnjVTlSYkZVVG+7LNjjWyTy+kVniowrLHSAUnaZqah+u1P - 4HyJYILSmuAxWGIq8BD/+CPRnqTz2eYH/AcAAP//pF3Ltqqwsv0gG/JO0uT9liCgYg9QERCRV4B8 - /R2sfZrntk7TsddGCZVZc84KVXCvn83i6P5i5u9+LupJmZek1LUleSwNrN/Nb152P3e9JIuP9nyO - s08qDeNWiBV0QZgQ/Dow8fh4qw0aPuEXnyPzVG8ZN7sg4OrbXJkmT//qB3/+Djl9FH9Y3vXqI/5D - nEC4b0pNZ5ibwA+ZHKufx1njtNW8wtcFH2ZGoUu84LetQvx42lhGowX4ablC0G6uiTGGJmB3vxa2 - pbUG4pE5xP/q37te/tMzw9rMUQsKc+SIjXUxX6+tOf7xyb//T/lZPfvweXhaM7fXK4fe6Hqofmyb - 5Jqo0w024AD+PtvdQc7Z/OOW0MUv85+/NKHkKIFBzjxids5RG/NTx/wvJwqE/36iQBi2/QzzeNRo - qi4d5MqoC5hwONBNIKkt2D9ZIDdaDGB+PdceJRF7wErWQG00mzVCC2sa+5mtdlhSMgUQRttElONw - opzs3X3JpkseHONvpvEH7u3Cdy8kxBDDtV4swYfwe9X54HDXtZx5MJcInupIJeqx5bwxtr0roKH9 - wfmhu3mbPGsqHMRCJomyELoc7CUEDgMLbOpXTePe2+kKp0pC2Gj3riuIbXvYLCFLzi84xSONb08k - nlL/xE3RllOdV120LY1FTlhQNd77uBG8ws0NuOSnxbO33K+Qsy4V0ff1WOqVVmD/HGzFzFLyVaoZ - ma9bvDts55w7nR86OHGjih+vRM3Z2xy2KJyEI7lqr4qulnLc4Pdq8kTzsy+l09GWYA1OZ3x9DX09 - q05tQm4+VcSF1bcepVaS0akOVfLgbIFOnLLqiPpjg09yoWrMgt8SWpdVIedUar01p3YPPZ+yxC+m - 98DYGcchvL4lrL/WvYuPfAsg6J5jAH5HW+MUAGSY+dIJ277z8TamL3WEFsOfWS5wPb5Wn1fohZjD - GH1Gb8szpUFzNWoEz84AthcFT3j42ATrddLRrXDrAsoeHuYjf80G5u/5npcXJAqLJsB5sRcgCw4b - to6XOKckSjoke6eBBL17HwgwfAbOQ1pi3IZ6zHFezaEhamWst9cw57Z1leD1d3Xnpa3yfMW1epV4 - zBtE6fkiZ/fPiLHzMvhompPTGys3UOvsK/FWAQ8LFcQdAV8C8QOdz9eKcxsoxLpPiiFSasZ/tYJU - f+kZO9ffKd8ufZygrwPvJMQnWvfVw6ugZlaUaB+NA6vxs6/o4+cRdkJwihmoRjqaev9Fzh+niNf7 - rezQ7Wt6AZKOAtjSuM7QjWwdtpmAybnnWelRPHYWeXS66/HB6rsQXNSRGBr3yinIgh76bs1h2VH9 - mD9wPxdM132u+/N2G2g7zFd4rKRlv38Y7/tDhpfv6zizMQ3pUidTBsxrz5GAKGK9bXjveug8s1nk - R1ZbwKvzkdpdQvw8y2LcMkJoIph4HVEZ7VFzF+tdov37sH93bcAPl0VADnMosDzc83jdXhKHxCY8 - kaKDPN3SZS0RZnudqE/14XGCHqlIPyIyk7rxalJl8gJfA4ywfEiTnO2GooRhur+zoq+axlW2MiLi - NBqOuf6Qk1kfezj5Kpyp/i7zsY+8HrHO5431g+/ETABMCZqo5HAod/HAKslmI3X6Tlghc5GvzuO1 - wTrv6AyUuzWw2qlLYCUp4bSJ+UyX41CZ6KoWbnCkske3JTD2ObK6jYPbeKbU0nQdvXU1CdChWgdS - q88EtLV+nxfu5dOlarwS/KbpSgxI3XxNn9IT8pbbEf3tvIcFFnwL8GyI2OtcMpBUGiNwSssXsYpD - E4/lR+zgxTm0RN7jaxqKRYIFifC8khnmi2o/N1g9hA3fEWiHNfluGeIS2w2QP6Q585y8J7TjN8WO - oRqUd9olg/oms8Tp4aleu6Go4LvjMY5X+szHQqNPhK2TiOUxBANlG2aElnrZiOEaC1iXd3eQmm9z - I+mBJvHaTmqPQHT8BuJlO8VUPcQHuAjnx8wXBz0nJiqviDWdNzZw9BqofmdUoBwUTApzfGu9KQ82 - /NGgmJdSMWJ+DdsSxDdPJcapc4aNHt8SaD51hG3YTdpydtYnov7czECdAsr7of2ESpGPM4m/mbck - PevC1a4MkrjoWW9ro0jozt9toqCyHbb6ss5/eI6vcI7qLY2H9G//YyN6/epRKHMGiOOhx46omwNd - PZVDsvoOyAU/VI1tmOIJ2ar64tP39ANMtJkHhBP1TWyGGPWk52kCLxhrWG0Lli66okOkj2+CX46Y - aSs7wFmSFNoSLfoQbyzO9RM68VXEXlWQmAZ14kJxSQfymPGaE31mGSh1v5bIBQPi9fWGKZSHIiZ+ - HfD1JBceBNHIKtiaPx7dbrVTwUVWK3JazynYpgaVcKGKTK4Z0w/bWzr3iPvUBsayJ3lNm3sN1JL5 - SrynKdecFeQ2XAr3iO0DTfLv2w83dNfTHhdtrgwca8o2orpqEZk8jJx1Po2OQIS+Aft5VgOn3JUA - qas2Y8tjakqcdknhdC5V/Fo9Gi+ebgpofM8Q21T47K8qPlq4x/9ffvXYYd1GFPUUYn95LcMmTZWM - /p6H3LefmDKvu4weZXEmuRl+61m8TTO8BpcjOZ18Z2DIuMjoYCIfvww10/ilSyFqlojFbngP8jX5 - SimUnu2JPFGvgU1TzyEiZaHj13Mahu2znmcIT+cvyU43E9BPq3Zo5AMFm/a4guX3PRVQz12GeFpQ - anwmcC443OUzecyLOnAvSgtELuM7uH/vCWDaiI8Q7QWLJAf/l2+ydw9gxDcNeQnz6G23WqnQNWYN - fCq9qqaemhZw//udQb7zVdgiCQH/GuD4Gzy1+SpoCXLeZREg++LUrHpcQnRxYDszpvkcJnzJrnA+ - cCXZ4w+s53gM4HTdXsGosG9vjfmwggw5lyTZFkcjX6WfEcqjkKisUXr/8P9Y8gPxQtainBXELshc - 3p5RcRCG8bNPEdHE95l4+GANf/sF/fEDDM8Nna6PI4QhTQ1spKcfoDB3XBR2XYXD1cqHfT8c4KN8 - noPX+97E5Koc9rm5DUPUekHeFAZ2iWwpp8SPAJ/Ty9yncJtcEwfI2eoFn7UQza9QxUnj/bxFDWsd - Jn1qB0x5OHor31QS7NhnHhxMo883gYQ2fNNmI1krs3SBaqZD4axZAUXZW1tq7IwgFQdrFi78Rxu/ - HW2h/hFu5NkPLliHZGzRv+elvmvAWq6uwyVbdKJsoI1/nnBcINMrGilq0oG1YZJCxE54IMVbW/Lt - b/18vwWBVEQsoOeqLMEXQh7vvwdMe36Ev4sqBgMj8d4yv2wZeqb9wMm2/LQRvLoAhZegmpvCOnpL - UP4yOA9ZiZ2n33rLpJUyms6VSlQxDwA75e4CfAEqQRML3bAsPc7A6yeeiKujWVtTZ+kQWiyfaIJK - KImQKYNo5JUdH81hUUolgFAmHdGNuKupV7AQIIH+Zu6Yed5mHyAUsYVFcvqJ6sBfFc4FNxr8iKfc - v/XSHIcKzklZkhM113wKtrKAjuAxxD9/Q4111C1EBQkxuSa/OqcH7TmLY5/fCY6/kjbOA1vCIK/Q - zD50BzDD55VAYfl8sGvZl3zpoBXBPT8RM/nVMXXxCULXTos9H690OxSPDebXZSSYygP46pV8RSvZ - 41FcK7BkJTZh8JY1fP3c+/2dIikCl+/jiK/q4TxwafcJoCqblwDeNj7erpd7BXMtGfGDftWaqMIS - gZ0vEBnWOVgk3Q7gar5k4oXJdRhvRy+BGsYh1rYE18uP3Bf4K6450bDI5atw8ReoJiCcx0Dn4zEl - kw8M4RFjl7EDwFtl5EOuDDvi8+8boMNyzeDmlTW5KZ9tWO9jv79zmwGy/z3l+14aIWWO5bwxTUDH - iyyWMJ3K30z3fL4U7RSK0f2ukKs9ngFvwA8HRV37ENW4z95a8sII+bSuA+Hm+flv7EsBWJd1w4pn - 52Dp5lcI37y84ZO8hZRXjA8DecOqicoEpbY90viJuIL5kuztXCgzB98U+H4DgncqmR4LzJ8LH1Q4 - E3WIL2D9PpoS7vpg1w+9ttKpZ6SAlzSi3Qrs/d4oyeDOJ0jgO2K+oA1IQL6gAFsBN8SUt38mVOzj - SPDINsOyXWUdFnaKSZJJi7eeD/0u0e9f4twQM0w3aDfS8XHv/+HPP73yx7fpxsqU2/k7HMNEIcnh - eIz3fFkBYpUHrMAjG6+SIJrwlZYROdevVluvD/6A/KCX53qyZEBxoSYi8xA+5F5VDOillyRL8/dQ - zlXQQkCk488EqqxfcPJ23vXsLecrFNnBnhekT97yeOUFuIvP0/x+5YU3S5/HDN4wxn/xnS96zS3S - eXVN7EuzQnnWmltAfvYbJ/e2o58eyxJ6GWZEXHoOapo9ww6uC1XmxjSf9SBc7zoUfqTB6pmagLMP - 8ADShvOD885/5zmV5b/4wr4z28Mg8cUGUvFnkWCaopg5uR8T7nwNnw7VeZiyT9pDgQcWscxiq+n3 - KTAge55trB/LcNjCGw3RI2D2OdBg8lZ7H3O0NOwDy5XvD3TXw7Dfbum82O9cW5yFuvDS2c4/ftJZ - 2y+CviY9iWbzxBvhFProI2wpOR21CswomBb4Lt0Nn9y9K6qGcI/2eA7ojmfU0LcAAly3+A9Ppj8+ - 79dPiDWuf8a0SJ4BPK5mTHT1/Pb+6b2d3+HLHz8XT9sBrdNNJqfoLuQLny0u/Ms3O18elm8HWtj6 - fkr8uLAHNqdyj/7w2rNz4q0xQyRY5z3F3naXKad/PBMEzwP6p6/GVfYWuG6cSRxD/dBF/sYyFJbv - BwfUqIc5uhouTLTSnvk3q+eseNogCgpRwMbFU2OmmNcWwV8/YTM9bt7EtGdZ2uORaNYLDbNuMQcY - zYKJsZbc6Vpctgoe+WmYmd2/mQWHHf/46zx/YnEgod8EsG7TawDNUfGWrLR0aF3oRrDGeh7949dA - ZL/Y7MMebKsSmvBEHB7buKP1egnxAbbMKwloviGwdNzQQmhrLvay+kuXprpkCD4qjG2LjNrS3S9P - 2D2bjPiNwdOVU0QTtq8+xZhTaTzp7nGWvuV+ovHEJDUbbF2BXmkVzWIPp5pmAmeDHW/nsG+NP/+k - gbu/FPzTDx8tZKCpOGtArdejJnJFDyjfe3gF2n0dFmr+oHQiHk/86/GdL3NAUun0EzMceJGoLUX7 - CSXOSb+BGOCk/ghbJoDyUAczA9w3Xe7ZkMIonBt8ouY53/0Z4S8/Buzw+wHqSb4OK0kLg/qGyngx - mzVE+vFIgvXZtR5lRK6CwzikRC6VT740xTrDXe/gQGzvYBletIN88HSxsl9+afh9/jxyS4ynLqv/ - +Cq05/iCPTJ13ng33i36vNgIB4XZA2qmJwl6wJfxo/SqgWTCwYatLnEz4n699kuVJkF8+q7ncj2A - eg0qzECUhyFJ5amhywuzCYLbxSPPR7Zpy0sCT3iRwHFu4/OZssLjbEK5ssZACobZW99Gt0j3k2Fh - LWXkgXVUKfrzM7CRpatGz+SqwofThLjwRExXI/2OMOWwFdQrfcaNrvgQkJU7Yne/3tLNt0hqjz6H - C7T84ia8zhDu/g45wVsf94yQ6nDPL9iQbwvdtt/UwIVqMrbydu+Zen+nf/4c1sD7GW/2gTnAwWIe - RL6ZucddijqBY70p2K+E7i+/RbCfqgKrlfnVyP58gFr4G8b8NavpfjIQzopZYq3PbW/c+RkKKBWx - Ol9Q/PnLby9BXwIJcTNYnG7mAPYSjTyDV+3xjtNH0j98+MY4H1OdmDDNzB/WJfuVj4vUyfDeqqd/ - z3/b/UTAErMlp1soxDRYdXd/Y8uaD6ivAbmHZQ+Pw+eEvXTG8fQ2ygWtglkT7IiZtwZ94qKHVXvE - L/wl/30lqYNveMY4PVAmp8zNH4F8kGKCgyXTlh85L4ityi9RKxF53XaVzX98/Wq+3ZgZqZVAIPJf - jN1syMf1cReg7765GY3A9dZnkY7AdoqMXFxjoRP4RQfogUAOBLE1asblrhw63z8Aq3U2xxsXLza6 - /N76P37EXYohAcJZscjtnVBtXnPUgE0MG6Jcg1O+kHFR0XPtIuKlM4nHxXq00GkmhpgI17SxLrGN - 3odrEsC724FFUl4NjC8GwobGHfM1ehcBvA9Rv59wXfPJ8LYAirfXJ5De/ddbh0IQ/vIzMYnVg1EE - lPvTh1j1YQv+9BJKK2qRU/Ne8zmAsQ+uF8ElmfVr69GuPRfcvdnHge/c41WUXQj3yVPkZaiSt319 - cYN2xl5IsAVaPTaFOMLANhH29/jeft19kf78xfPuz5H+1LTQ4M0Um0F2pt12Aou082Vc5JyS08A0 - VcBs05vc/uJ95xNI949ncgqJmq+/azSD3W/Gdu8a+Z9ehvqmskTHij0sqQ9SsPO5QOIYK9/xqITw - ++mwXh5eGquSVw9aHiREU5JrTntbliD/STyc1IkN1sCqrpCVgU70a6PEzL5/gPy7BzjhYuU/fFoF - NJxJKVBv5mLBhuRrFMEaXD5gegTliC6gcbDJv7WaIcUtlR4PNSTOn3+FCc+J6qrMM6LmGq80fcvo - HrUnrJwYZqDt2ebgdvspRHl2rbYw/qmHpxenYFWQHjXtq3afybUoJBTfTE4u97KB+/XxibP3qUkH - 1YW1HdDgqa+19qe3JdYOWuzdcVlTx+lDIOttTHC3SvtUqKyBuz4nVtIMw7TrcdgIx4iodw9p6w54 - UOTH07xuwMx5T02ff3yBXJzmWC+JMNnQFKI7cbrw4v3WyOXADNecuFFgUka8u+o//wrJWwi2Lvst - cPzYEDvTgdW2mwEySDtWxadLq8ZcAONAaiIjInqeKznvPBsTNvJ4I45wageiJekVrNXniY2nWOV0 - 19d/enGG16OSz7cM+f/qH6aQ1mC637oO/tVvLOfo5VQI/AXab3LFf/7S8llOGQjPWzVz5NCDObj/ - MpixjhN85mGNqek+JHiTfY9kUdBS2h9hKfz5s7f13AyTHuuy5H1G8C+eyF4/Qv7av2Zi8ltN0sfT - BbvfQ5ziNtXT9yMwQPkFSSDlOTvM9WnKYAPUG/b5Nw/6zL83cAgrQPSEdJRUmb3B5Hqwsc1JIxiT - 01cFf3pA03rZ+727qIHt4xMTNWgLQNW8stF9CHsckMqli38+SCBB3oG46vkS/6t/sHnf4p2vDaT0 - nqFo6SmPlbhiBlKIQofKyY6xhuvvQG637Qlg60/k5DGCN5VZ3wDKqx8SdGVd1+R7LSDLwu+8Xiao - DX7/OQCV6hTfikfszQoAKhTS5EdO0jGlO3+OkH8InkSVq9mjJ7lP4EWJnAC8Mgsw7u99RefVNnHI - sfkwTo42g/vL9bEvy3a97f48PJbsgB3+auZ/+QNtS2sFa3NtY/pLuBQk3U3b/Q/F2/FVhoY7RSQr - ogulT3LYexjcZOwgYNb8b9NLsPv3ePfrNSKtq45KlYmx7PIN2NzsIUMVcZe9j4cU917s+cDUfI4E - mqSC8UfOm/Snf4rdf6LZVb/Cp3U94z0e4u3zO2Vgr+fh0y1M8/G1fW14tcbvLMqeAfhHZfp/enNe - Zsh42yPNCwhsIcC4W7Oa8rXZQChPXcCgixOvyg9AeDL0167PAdjK5pGAaDmr2D+CpV5Wnk/hn/8a - yJPobUNb98CrBw77q/wGm6QOAnQ28Ub840ehmxxCCYzvEeKgH3pKSdYdYPBoVJy08rumB+5nA/99 - 5/HpXAk53cDY/+lbbAT5cZjkmO/gFS5uwNVBHa92tPeovluEWPE0/Of3Dnn0CJYsNjz69sMFbTx+ - EWvM9JrxDSeD2Zs4xK8EG6wuKFPo+StLbFUYd/9KGVGO+jGASkIo3f0A4FhBTk6lpw5j4dxcYFv3 - C/ZsKMf0MCcc3HzwxD6j/uLVpYkJJ9S//vQlWGi96EjbTiSg+luOucG5lOhXJDkprsd33Ct3x4fX - ufkSDWVvj54uow29/phifOeGYTxNVwmI96rH/r3k663uQQmextnHD9VMtZXEAoTv23PvIfhO6uUd - WD5aGv4xU82Z//JLArnIMna/afA2c3iHcHQ5gFVofevVE/gNSkVtY0VbPsP651fMDx/NzOH4irnt - N7XghW4nYhw6lTKc+yrFvd5ILPZL410fQShMXoO1ySppbzzWCBEZFzOM7He8ek5awA9HYqxW4sNb - hs/tCgNbR8GiUglM/uhGUBeTG5bB1Qeb1xkybBL39k/PjyKX2uBl6BGWWe9Tz5luhEDts8u8GfdZ - o0na6rDG5YUY/aUaaEu3Hg7z+AiQiVJvephFBtNz5s/Mno9pZ5kpFK7nhjh5ecr3+ukTbTpWyGmo - hWE52EIIb8W1w+rviWI6XAQBdOR4IQo8XuKZixcX7fVSkipaXCdSu6nolE88PqnfX777MRX4qx/7 - nogB5SRRhfVWMcTrXFyvPyX9n3oUiP/9RMHUOhYJmJVo6wodGTZGq8/82jDxeNLPEjrWuUNcRmHo - gDe/gnkzvuf+NhjaYoxqi5qYx0QP6wks5cSa0GKsmhhhc6BzwDML4L1cnkffYeqlJeUGJVSoM/s6 - +XQbq0MIheR5CtZlNgATvIUN3G+lhc+SzXgDlV4FmJ3kjLPlUXnUcjQVwsOXn6lsGTn7i1wIU2Hr - CV6sTSPbg3Whsgw1lmtPoRu0fwlsTaSS6GJ/6wWeggpql9LDdr688sXktx6B2IQE49+D8mVVm8iY - tRmrYuXkPGvpLizRtyOn3iJ0mU1aAu6yWcTPNnVYFx9cARvzJbEt1sq5IgQ+5H8gwcZJ9j2GpoUg - ibwxk0ddxZSN3iIDnDyXiFomx3i8iVoBfhdGwrfHmwerdv/Mf+tBvHn8DaPnFDrSxm4hUZmuw8J+ - lBI9P7qBlYl9a1vfsSrabuKPyK1uDhwllydkfvVCrOhSegz7USoE7YkNZPa4d+FvoAml1kywXvM/ - uuTuKEB1a1z8UjEZaH7eEcxNV/z3/QvDLVd0OohnjJNSBSx3qgpET+sQHBcr8ngZlQnc42Nm16es - MQwLNhiT7k7OwtUFzBeEDPocvzHBpKgBG9GuRyU3+wQ/cO2Rp251aDwvK1HfL58yYBltWB+CS4AU - VdYYFGgmWmUmx7ktH701yKwUQqp9Zv4iRmCMgyMUnoJqzoe2K3OurbALzW0IiHv8iQMtplGGKult - cgPPyps+/rmDxqGUsXlhyDBp988IF5wa5FH8SEw9LEgwo4qGdfU6guV1IAn01+FNNEYY4/X9c1L4 - Fy+eb2sDc24/FRqokhD9euG1TcH2CI8zysmju7T19jj6CUzfnRF8PyeiLaoKenDXWxNfhvIIWEbZ - MrTvBxLGNM65KDxXqNfbG3kdltJbwcvj4MeIG6zq097lMc8K8fq1a1LoDxSP3LK2aPy5L4J5mau3 - 6FD14DXtc2o7sObb02WeEPfCiwSm6Ncb4zMpmsNZngV0hvWad1GIaowJPoWRq/HzKbwiIW82cknG - OmYuedPAZ9mnAbg8X8P2beQQHp1rhRXrvcbbFNstMh9yQpLKkGuOPJcOnej1uE/elT3GyPJFqllr - w2aomjmdHoEMbwS2+PROHjnbzNsIxYdmEeXVlRp9H40Uyon4xHfiHMBSm3cOvh45h7ViRTX9MPoT - ATnVsd/4Z42tzPKKsOj1GJf0AjiNe5ZwC3QPp7mjDWxweLdoSBaHGJV5HpjQMTto8omOsfww4/l2 - vG9QgScWa6lJ6qWPVhlowNGD4R04Gr1USoqwdXgS5TJH+Va8pwI4+vkSbAu50mUhhgRP/DISu6w3 - b1XLt4r8Qv6S15E6HvOJHAHa3pxhUwrO+cA1gwxoCbmAUa8+WCX1XSIwS7cZNqEWk+Y6tfAIVBn7 - 9y+7d+2TegjLZMKFSfa5p0o6Q63Iv0Q2gtvA+K0EwY7XQf5KTzEbjNIGJVV0iTyNXr3pbcfBWJYQ - fqkeirfRSisYb5dolpZHpY1JKJpQiTxMHNBcc+Zp9DNIn6NGNMUNYjrODIcenPXC2F/u+SolaQLq - qtewO5WDRmdplKRaE1psbQMeuFOePaHWYY1oBpLp0hZsCuknq8nf89q/v4TFCiPsTZ8g34h0zZDP - GcHuCF+GwS1rF8FkyWY+tH41MyhqALOkJP/2+0LN1kR/6x3JqqqxVWWa8HkPf+SxEgasHVs2aOBn - 99/vXapuE9B7bAIckmdC93hioNW9Q3wqp/ewPKuTCvPzUgb7q0PDdkjuGRSVJ4vVqu08PnLGGf7F - z1Wri5jaCdWhoftn/NrjZ9Ehw6B77x/xQzUcjZK7nElwkI7EwqmjzUiZVXA6jcaOXy9tC69hBHPj - qBOZN0VKoiLUofw3FcZRWm8L3NJHezwS455rNX0U4gIXnBkkMOajRkMEG9go7g2787fNqfAqGRTz - qk3SDjs5Szk1QswhaPEJJiElC9uP8DmgN5HZVKXcLfR1+DHODXaS0wNsbJ9H0MoiA6uvScvZHQ/g - kzZ3ogQq8Ja8iEd0ogZHlNa4DXwZMhFSIGZnkZFr0H+kNYI3sKj4PO2KQhnWEgn6ScfPJ/x5jOeG - PUrfvUHcPX+vn6ddgFvHKPhmPEFOj/ekQx1vPObDYr615cdGOhK+nkG8r97U46C4ATSNrt3zheZt - MAttNCSbg1/+IsZrcPg1qFJPNr4NWj8sS1S30P6sCnl9RylfdLSmsP69X/j1XfyB9TexFPb7J5fy - yOWkyFkf6ZCpsI0uAWDv81nez4wa5FG3rcYdg6yD5Ra+yJm9U0CPcAuh79HnzL11cdjuniIhn7MC - crorbs5ePJlDSVc8SFgGw14hZgI0AvY2g6Uy4u381hb0PpYYn19d6dGeHlWYdu83fm2Xrqbf4iqg - iDXeRAuffU19PhjhdY5tosNPUvfQGjfYdVwUVAYz5Muc9lf4OTIFTt3Z11ZkZwyMe1STouOP2nSj - cYBUf7TIRTQ/2uK3G0Rj8/oRm/Kztmnctdq75rr7fi3BEm+PFt7SC5mXwhXo+hFbHy2TcMdZmhk1 - tTKRAYadsoG4uba39g+qorv0tPFJvN0BZWDawhNtbeKV1kQJBYUKS9ED88o173iNx12S3OJyhju+ - rYbg2HBsHj+CG77T5kQsiz++ifHdW8BCmiyFR6me5/1+vO02tql0LicPB4evO6z+tpbIu4CCqFfu - sNMeQYfr6zoHx/bz0uivt3UIT14VvM+aHi9qUj1hKTMuSSnNwZIcTj1qo+VJwkGqPL43qwLwJoqJ - eXbe9XDO4xKJxu9OXi/FzxnrK2fQclVIrJ/w8PrGGxrYgS2euR3f1578GOnFyjM+B2quzR9JjCC5 - lxqO3q+97z04CwDahMU6zyUeeZ8FDjmD4ONo1G3AnF58D1vI6/hUt6bHMp9Uhlpx3yuwxM75C10C - WNHhMZdGwNd9W1kuPH3res78HmrroUE23PGUeAOJ6rGvhwiaiS1hlftE+dZLNQOdQfKJEh4jbZuV - +wZxL72IjR9+viLrGkpLlVhErmQmpxd8leE2Ni+y/35KR3mY4fP7ZognHeFAHmKiw2M4n4m2Ajam - OJgFeBEcRIrbPtXEEBQbrpX2nWtZo952F2MZktP4JWf2YXrcbZxTyJJur/JWikbz0HChdqk8Yugn - sZ4dwfUhfl19YlqDTCcbshK8eNmTmHam5KODhgMcFMbGSl6CgZrDaMIGFCa+Vl5CuzsmJlSRGOOH - eBMBfc+xDaTn+Am+ndtqvbmlEhyv/hOnurDU1M/GDA6SIxI/LUWw8GAyoSfqLbGFNhjm6NB3cK57 - BSva46jNxpokMCwShuSdJMebvL4X5AKOJb7R9XTZ8Qtkg/QgqpDP3nZ9SwK8NWlCfKyK3gbt91Wa - UMnOyyCpGtGWNoL85/UNxP4ZUvasZxBe1DMIKNezw5oVlQ9f5vlMnLjCNXUD5gruEXsm+fKMvTXv - sgjwN78gMWzqgcJ4SOAsF/ye75R6O5CjDU7U4vDJTQtvK2nAAHqiQ1AmFaUr+7pH4A+/sttgeMuR - j1SE3K4g13vB1LQ9Dyl0wXcMaB+HOcvfsxJ+vzaDg1vSePPThQW8vjYmOID5Q/d8FsCrZDbEGkUz - 3k6jP4PpPWXEObynfCWPqYWfcYjmRok1yubfx/zHx8gdksZbrg67IGI2KgnJkwH0ONoqSEK+/cuf - YO0q9G898DP2XwOjBfWMUPOQgvJZKDVFq+qjg/XTyVWee21lSAAh9jMPO81o1fPTOUp/+EvSWYn2 - 9eFdZHzdiHji+TNsF+31hCwobuQm3Vrw736U5Vdjf6m5fBTZ+SpyDqmwW6r9QKeHKUOjehTYVZ5N - TnAwS7CX3X5GMNU83iiFBq49UIid+CNdqi+XQGzB57w+3w1ov7U3/uE3xlok1B+XBzO4gUokylZi - bTMDGsHIXVfsaj7IW0PzfJiViReMf/G681spmg4Aa8c7C6h2sUv4cK4y/ssfq/F5bODeB0eivWYm - pqwtz4hNyyv29TeuqWLHElD92SKBs51j7pCcM5gNwgPHn67K1/FX66j6pNFeiSuHf3x4dq7neW0N - fti0hswQdeA7A+XF5tQW0hGE9VwTPxESOi7KekWk7H7kNkGcL1ZykeHvnZ+IIjfSsHHpWEDV1s5Y - l+jgzddtctGbaChYr+ypZgVOzOCZDv6+3m69sB+nghmNfKyFT7fm+XjZ0FH/jcQKtVvOxbkh//Fz - /GQmSOcp0lUU1mM900KsvFWw6hlmSUWC4x9f6G7Ah/CUDcEm3HKwdTPd4E+UdexCxsqX1FhcVGS3 - DFs3XY75U5s8gfDN/vB2G0ah3xgUv/KQyDH/qddZNmfwtfMbdiDRtUWKTxHY+QnGhc4OC3yyCdpq - YgT3Y2fX/OXtc3DH3/nyhSdvx3MXLJF5CNjuC2LyopcrWNjIxvqut1hJ/VXSn16uDRbGW6i/Cyg+ - ohCf7Dai1CiFFtaa1OJAYJWYSvfehOQeSiRW4a0e9fsrAH/8A7efo7fn0wo6+V0iTnqp4j/8gmHn - hlj1L+w+6LKFECrcfe5Zv4rX72+pkLd3/XXI6azRNXZ6uPtJWAu6gDK7fwR2voT9Xa8uJ+2SgKt9 - N4mZvVBNAw2kMLiWDjZF66ptxutTQveRvPFp0k3AZ8a5g0n3fOA/PcErmToi8aFY+PwNP954OboB - tCNentlk1HI2omUP3cf1TYz1WXpb7GYb7LZhwdbjzVO6+0PgVHYWyc/SCDYtEzY4HD4t+VvP9UVk - G90e6gdbnUnBenqrI1K31p1rQwyGvdJ5ANmLFbB3/UiUWtyrgSI52jNTjUW+uPargmgpXzMnRfWw - qtDcoFG9ipl7lC3Y9lMu0O0EBms2ZwyLaD0CCOqrgS22awdyHG0ZWgEusbfsXbh/x1cF/q53fEtn - b3O17YpY0oMZPgYrn27llwPV/TIFq/u08+1bayOC98AK1olVtE1PJBne/LUg+vX+i9fj8vGR2GTc - vH7UoF6ct9Kjoy6w8++SevGaVh2Es/7bT8/8ymEknOLDdYsWjFl0GniV+6jIErY7lgV4qdniWj9h - v59wuehnj/LNrbfhid9GIjOnSzwSzvGhxqivoN7z1+zDA4QRfn4xziLf40yImb/PRItjSMc//rav - J/Glh67R5+VQ/OV7bJ4UpJHp6TbwXBKP6Epc0+Z1IFdYX7lvcJQfZs7c3w9JRI9Lgn3uvgJKzIcK - +blxcWxzRj3++X+RuNywbN4Nb0wORg/L+K3i4L530T/Y+r4fHxB7c3fLF9mWMrDn+0AQ2nlYJ0tV - 4a5fiKIMF8DXnmTCy+dlBMSwmbjPvmMAH4UsEMUHR7CaSV1COLo29otm9JazUCSgSSuDOL/8S1fy - +LRw8vWS4KhNAH12Pw50/jrNrFYNMf3Tb8x0fBBbMy2PiqdNhp5otvP05D9g5QclQQG2M3yzFt1b - p3ThIP0xIfGun4yu85ktkX+7rWTHH48yvymQFFxc8UMID/VGpGcK+VtQYHeYlXq7e44kvYmCgvoQ - ScO680/U+OOPFPW7oj/1xPaScBUhMU8GiZfpNm0QTc0Dp823icmjWBd0oslx5ieIY06I+wjmHTCI - /VztenWyhIED1ZK5bt5fbTjekx7dtuEVsEb69SjHnRnp0DMBOZfxnJMZvAvkF+oXG6L58X5DN5dw - 9bMpkN6OAZjHtTjA5VoO2GFkDSyF93v++bP//JNtPOMCbk7/DL7GM48HXD5meEscD9s6rMBcXscK - vpaPQp5g/oCVWqsNVovwRHWKJf98/HsPkiPnYTMCLt3q2e9AqE+InITwMKxM/A6Qa5B+PgqSXs8s - dHXYfh5XLH85QhdtU64w088tcZ7bvV7++CrCphNIm3fX/l3vyS8Ix4ob5Ez7ECEwQdURr7roMXMT - tSfa8XH3B0yNphfrCslw3AIGM6eYin2R/MvX8q5fiPBsQhjK9Z0Y570Na/qVRtjfYjz3Ob5RJncb - Af5d/9Scmfh3M1wT7p0vcFQmx3whnBNAdfBzonE9W6+EJBnspeKOH3elz6cdj+AsvR+zSJwnWPn0 - AuHJ0XAwdFKZ04msKTAOET/T3a9b56YU0K81NRLosu2tweHdANFfD1gtlrHeovBeQk6Ybex7zzZf - IexnuP87cYcex/x7jt0//MJWWb217dXlCZTe+ZOYnVEBqi1zCMalCog1y+96+xEhAUt2SLFy8Nt8 - vZWEAc+Egdgngr3v91MPHk4ik92f0FZcyBVEzUsicjQK2tzJVQ9bLyPEF/Cp5oPvWYane5TODKf3 - YLsaaQvLXlpIEMZA25+/D8gYgBncqzP4p6/2/Y1tKb1RKg2pBHc9j7WdLy9SfSzhWTDFnd943j89 - v/NrfD8tZszl223+q2cEixM5gIV3w4SH5/r7z/1ew2jny78I63ER5RMI7i00eRPMyr5fFrs09L/9 - j8/L/KHUcjwZlL2wEPs01F6nrv4C8RYGGOMfosPh1DTg8ry+ZkbsHkMvLn4L1zRw5q1qO21y8nOG - fKkExLPlo9b8LBkic/sF+J8/A9RchrFim3j3l4eVh4uNpjYL5x8Pzzn5EjVCe72IPMCvqsldv+hQ - 5bZ5Brt/vpW3nwl/wcPDMq0sb0l10AJ0XlVidx6v/cufTaOn2M4draaf6tKDnd9heddLf/sVtiYR - yclNoUeHRklQC1mdON+PrzE7P4dn1edJdtp7Li3Pfe7dju/OQ17iEYFXLxVZU+NgwNBbrsXbRqPb - e/jEjANd+CCsYBqc1llSss2b//zl66E6E81gi3y7Hc+bEJz5mBhqvmjL1/fcf+sfeb1Jp+PyCZAv - Py8Bgmmtkcl7BbDwTRwwKf54f/wX7Hpl/kJnydedvyCm6hJ8CZsDmBX4CIEelRTr1KJ0vbh+BLl3 - kv2rj62CNczQ13tx/qbcj1K7/QnSJap18udfroj9pvDm04LgwrbAP//Ne25cwART9G99QNdkn6DN - +0Fbuuuhg8e6j4gFnTBfzUunA/uZR8QNhK1u9bZkoL6GIVFnMfe26twnEoLmGki7HiH8kusguFZO - sDVFDQgOWgn++YWkn7SBZrRaIFPlXrCwaQW2i3Z7QseeVIzzuxWva58F8I9PW0VW1Dtf1eGej4hh - mU+PPqW0g6d7mBKPMVOPZm96gME1loP2IB7jZcfPf/oYz2du50PeCOR7m2DvaL29+XTY+cUSRPPG - fbaYbEtZwVQVOvy6W896oYMdwfvzXmP/fFy8dbJcFWiBXWD91HKARi1W4Rq/Y2K0oq7NL8m1Ib/G - WrAUbkqXOa0SdFFjQHS0qN6q27SAvi6pc8sUtjf/1UfGZtxIVqp9PUlJeEVznQlEfVYsoH9898+f - NLVvUC8IvDrIMVGOrd9ZyRsB/jqo+sWBBHu9a8FjE4Gdr+M/vFqCbVD/4hPLzNmqN3HRW1jdbxN2 - w62sh/jrNbDjrUdwcDit5nJDs9G3eh2IPp3Tf/UN+Fdfscs68tbRNP7lT+zZ8sujz+8Ugg+jKVhv - l/nv+2T4mKGIs/GO6SafrAbGaTTMK/32+XxY6IieH9Mg/s4fiJ+XEOH2uWHL/HD1zCyZvE+5UXDw - FK50188qvM7K8R9+L9m59CFbHDT8p6c4MwAR1E88JAZnALoel8mHCnu9E/cZKto/f8l5sV+87894 - C7etRe0abcQvGt+bQCmnaK+f4YTUb28xsnyD/GdesGuF07CETtBD5coZwfbOs2Hr5KpDbMyWWFsY - ySNrZJgwuVQFVtLhl893TxFgKTrgzz+LF/1+86Guxv0ssX6Vk8YAV6CmUTaLTcfm9IKI+xevf/xE - 6yEnFbBLvCLYVGAO/Hlyr1Bj5Be5vE4+4P7i159hh4PuytfU/nwSeEsbgZgH8ZX/BnCWUBxzBxLE - /rFenckORH4tfSzv+2sR77oMSi0W52NJWUDoyZVgwqjLfGw/R42+p2cD/ocTBdL/c6LA0sOZ1z9K - zAJiV3Crv0rQnWPN25x1zdB2HnOiP+9+vrqh+kRs1n32OXttvMbs1qN6qTiiPa5iTjOrlqF1Ky3y - goUQr10ZJzBSEjbgPsprP5MjJjApuyvW+SzxaMh5DOy5Yz1v99Ly2IyVI3CCekA0WLLalksnHeZu - OOOHIgnezGf3J1TlC0ceI2SH5TCeF8AdJh07RT3V632eN7jNFcEKgTJgcbo/qeOikafyonSSBrCB - +R46ODdeMB/7qVTh/HgHRMu/Xszjm9fDmv1Y2Jg07BE9in1wuPENUW3U0S0UfhmEZ+GFXVWa8/nI - ZhVsv21GTkTQ9/USUxDuGeqiBs3A8Ck14dk5P8ktmcV61cTnAZrBr8GOpD7Bxlz4Cg77mbjAyplh - TrgfB4Xf80SCTHrXy/lrNejthg8Sm7zkrZCrBGBz1SU4rmxG2SNTX9H5WAQkP156j/5+hSnt64kV - yRHyBcz3K1qFy5H4X1fNOeUYPSH7/j6wW+muxtcXNpVib+mJUy/bsPl5uaDvzDQ4u7fywPWLFaHr - 9ZPjR3q/5cyVagJa9LjEyvVNvZWr9QBC183natY+8cpPIIGVTApiey72WD2EDGRPXEhOGXgBjqtB - huy10An+3dacdm4uQOYTK8QPoUgHcvYEmEroRNwHz2tLqQ8mTFT8w0YaI7BWoZKiD1XV+VBZJp1c - sF4hf0558ozUyGOfUm0ieTIrrBwZdeAtqFbw92A2UpgViQerGUfQ6kZBNNq8wbo6aQKvbhORPD38 - 8rlhggy8qWjhIq0Nj+HEjQGZLSjk8SNhzofPWYU3psTYC24D2FatVZEvGk4AjfOgbcY9SCG5sz1R - 76WlLQzRRvQaIjyLtwGB5YXkHqnyjcNOYjQe512sCJ4N+U4uCnMYaPaSOSTGgkZuB+XnUf/uBfDw - eRkzi+yakg8ufUieIiBX+fLzaPiWNshALBB/Y1eN+lMSguv1m2Ofw048WVLjQrYdZoLBIag38WO3 - Uh/K/0falTQtCyvrH8RCJkmyZAaZgoCIO0BEUUSmAPn1p3i/e3dnd5ZWWQqh08/QoXskp+twByti - jBAVXxHjIow9ymNVy1HIXAi2PvIz5g/PUwn1IT9j2RLCeIT++Y38X1ZibUB8s1yG3wsdrMtAQsNO - qJAIZxlVC58HYOCRu2K9hhCWioLlJJ/pEk1PHcnPSxdIcXgbhHo0VXD/vTHZOxc1i2E2I5qrNMeX - S+wOfKRlb/DwuBeu8hoM4y91AjTfmwA717vXcOL9IEG5vR3JySE9WH+HuEbfq/nEJsCdu27+h0HG - A/5w8qH6wDf6SYL+ckn+4leb//LbXdFkonDik+7Xy8DP5vHETmo75jLm5SFd9FqiM6xXCKoj8jBP - LGZe9+8vjNxLUDB+EJ+iU+MKSpJvsLzmKdYffTgIOg4q6eyY/bw6vzWecT+9YPpKUuJ1I4oJjc46 - zCpbJPhbhPHiH50WHgI5wdq98QoexUaLxgtaccrgvFj1LurR5ekLQUVnvuCuj16F97mrcfxe1ma5 - GfcKVolqYO0XAHcNhS1Ax4xJcNJXiruNv81Eh62dcFDzr2HPryYwAl3DqvU4F8LHK0pojBtDzLej - FKMf1hW63fvD3ALcaYRK1JOGoQHz8Rn02mIdWQbszxOfXArddTxPJoQl75OT9voMa27qHYSdkBLv - 0VzdjYTlCG51shH3fGXc8dWKKjrVzgGr+TrEK2uFCwR5l5D9eTWceh9q0AHokWA55GB5st8WBr8D - Inp98eNlLSUPzJ84IvpX7+gYKVIAIw74M+2cH5jpR4Kwys4ODvj+o1EdB+U+tzklxtOd4nk2Pi0K - T+uV3J6Bo7Hq6ucSu1kp1qGxAFbQexXG0fQiZ9uyKXdYqze05orBpn8utelwN2qEH5cnvjub1SxZ - cMikzAsjkktKPSzSoLco+CGEy0cjaFSlOQM5MTLJKf2Ow8pcBhve7t2BKIF2apbwdYRguegj2eNJ - m/C+fJ8wCYnyuL2alS1+LISG6QRHUKfuFjDMGxqP052cD6za9L9h7dGOn9i+vI8DPRQPG15T40Ww - TaqYswIuRL6zv1PVy7LLJt4zkc6HKiDqrBnF1ugnEWSx7xD5s1hAOEo2C6Xbk/u3/qzK7kf4KDVI - AIVymPPq1kH+bLxxIcwVpaGTQKSrnIpVz5ndzZiuFfLfqbtPEXi5q56cWviaU4jNevm586WdM6jx - 6jSvsqQ1gljYHTrYUCPJkr806umbA1eNfokTbKErZN1rQenB7nGyDm3M5yZlUTt2A86GTXVZJpNH - 5DevAzGVm1SM3/KRwXQKdRw+WBPw0+Ung/Tg9KSsHgZYQG6HqFkTFWdXNfybmhAe9++T6Pd7xLwr - MwFktawgRTVdXT4L7xswtn7Ht+vLpc5nlZDIPVJiKGw1kAN/muHliQVito8UbJ1Yp6icrz6WH18K - KO97CwwDrsP3y+3jbkPBBH/xgJPHHQ0rziQZrU4ZkuxZFvFUdLc3/MunDJ3TYs631wx9BxGsGS0H - lqNAI7gG5BLAyjjQUjpcOnhrA0zi/vGIf6mtByio+++eb+7x3iO4hPllTMkVsbxGr6fXG5UH50ZO - aX4cZpVbXjC8vKIAvc83rbuEPEQGejs7P4iGvetmCrhhOgRLNNRgoyjb0I4vAfmgm7sk3jOFcp4v - +DRfF3cMH58AtmM/EPOjHJqteJEQOmEZBev14BRs9xJFqL+nbj66tNTW7XmQYXW/FlgLzltB03wW - wc4PMKa14bKhoMkwdrc+EIIEuysNpRmk3/xLTMoRuoaem0C88XUQ6rrWrOP5Y0rvbSTEpB2kS8R8 - csCJoUmc5vUsNgsUKRpYwyWR0oSDsFpPCFm7HInmy4G2jF08ouF5uxH/JgQDH3UTBD2PGmJIN1sb - hcuuyKPvGmzAc+NF/zYlXMXrAWPlrA4kfG4LPCOOx/6tL4b5cIcjlB+PvSv7PW8W93Dd0LyqfoCY - hzJwBXz1SHoU0cwO5FNsyyFM0MpKJjEhk7rzfHdzuJYJ/8c3taFMvxvI8+2LfV5M46V6Vab0WG+A - GN3nM6yN7EPpedU3YipULnY+noo7HwqOl9Ml3jqxS+EfPxF5bxwW6wgZ+G2MnjifSi1IXZxGeBIs - nuh8qoIxZ28mfH+MIKBw6tzVaswEntRnRczjuuz42UHIlEeOnIezXmzWV43+8T+8qT6Y5L0HiHIS - BaJ/6HvgPKJ2EJasT+QUXuONVeYSoEWcsDkdibsIei/D98cKZkZdRHcaUczABA7VzO//N1wYmkuZ - dPCD386v1nqf+rLjQ8DG20xnv3Ed2Cgywfc1P7ir/4ohOFkfMl8vB0QXyIEZvha/wXn7okNbiUYN - V239Eudhg2L1j+sC9vWcl7/967wmG/7xCyf1unjn1wksr1k60/j0pMurFWWIAF+QP/618m3GwCYe - nwHlzBJwj8iG8N6t2qxXt7HY4IMNIboX8wz6bQCbKzMeyN5qSgK3sxoe334eBJdWC7b7eh/mCuQh - fNTJMxAs6wQ29K16yG35j2gGtcG6PQUV1g/2SyJTkdw/PgpWBpc4CLob3bIvbQGT9Qs+sfXbpeEx - e4P0ePYDWnyHePnTZwc1zokp1uMwdm5sIv5svff7+cVL8LMcwFQvn/inp1ZwSp97YOuGKJCCg6YR - 4h9sKNll8S+/U/kcqOCBoEKCgonoRLvw//A8g7bfTDImPQzex/HvejQ6vq0N+jLnk0pW4DBeQp4B - dmZ1xNKzoFgbLZoh3x4UYpSTRrsdj0BS9ymu/Lc9sH/x+675E9Z2PrT6/CRDqbD9XR9s2pyJgghZ - gaFYv9u2tr0+px4kK1mIc2vrhuoSShG7ia+5Ebgs3hDDVnCbpoZUCVPT7XvnNmgejRyfkFs3Y9be - A0lIH3bApPjjbsy02DD66WLw6nnZ5bZjyUPVM5dgmUcJ0BY+GCk2xxv2sVa5WzRsM/QNUO18TG3W - Jl9e0M6MLkBkOLtrvtIXNJ3s8cc/Yxo6JQMP3g1iA9ka/fs9+G2sPlh3/B/Yux3AlzyV80d7dHs3 - szmB4vyqAuCbVsPmq5eC+xUJwdE8K/HGTKIDV72l2GDvViFkZ535x59+83XRtuIDVfD4ZBve8WMf - K3Ee0eQ5z0Da+c2WfcEbxLfcJXpHebD47/UFPVw+iEdV3Z1+4VWEutYF5Ay+eKDw5PfwgWyOJPtZ - NuotTxmGj44lkTvqBXU+RxGmRq1gfbg8Ckqxl6Koi8fgWCk6Xe/WjYWZFShYXUOnIbUp9lCIPA0n - qOm0NbSZFi7frSKBilyX8x/Yg9X9UuDqVvYxdYLERHLSxDPbn2e3b7KIAYDeN4yLZh5G4XfIJB+a - AVHNYHK3tQpbaIwL87efC3phaAZ3PyRAOFw1ynGzBLkt+xFH50nct9PRgzSDH3KTae8Ol/QWwiVn - Ttj3txvY9A+nA/FO9orAQQHcwjxtdE7FA3Y3760tL6DoxwwsN6K6ox4L8tlUAWpij9hGEBQ8NjIH - dKtc4Ug4mNrG4z6FD5XIgURyTVvUWtGR+2CyQNrxbdn1ElTYdCLmLV3o+pd//vyI4nBxNKo6C48O - PfsLtvbnu8Iff9n32yy1r3jY0rtXweAzijidTVXbVlKKYGIPh2C8K3OzWV8ngvPgEOxKQj0sBzdX - ofxiZCL/HFnj+MbzwN00ztiRFLkRbO0N4SQJJcEG8TS+Gd8yegwhDljU2NrKS70MM2eQgpUbAkq1 - 5lHDP7/BCu+axv3Chwjb02nClyy+0xVaJIK2Yq8kZcqpWAfHGf/pQVwsSrOsE1/DyCxMrEeJ0NA2 - 5ljoeIeYGC6K4hmyfgoR0PDuP321ed2eHfrDb/+o9MV6STYR/lRLxCrwhmK734wElPr4xvrOP8df - qnroQ2UVR2V9K7iF+TnwOHQmSVeTLfrDUymBYAwwYL96BxZDSyEsy/KLzS+ata17Oi9o6k915rBz - bdbayEMYakDC3gvhYuko58FoZi5YlQ57efb67NFDAewsFPfLHu/8G+56hci1cW+mg6d4yOxk+Z// - tSHwHWHSLyIpJKFu6Lt+16j7hXkg7PyP6Ixqou7ePGagPnpt35/dP39NkVK3WPQ4yqHVfZ74sbz5 - Ya3TqoU89hAOuFsNlsGrQ3CMdQGXS2w30wXdVWCVbTZzgxVoVH2cRIB+zg/LZ/83UOu7lkhURgdf - L4c7EGYzWZBSrPf5zx9cf/HYwX6bq2CZCr4Z//jsx+/u2DgVJ8r7waYDKb8qQQcvdbO615GHd1Bx - xH5Nj4HF8iRD1T2/A3bfX8I+vx5NjfTBsmGzdKtwk0M7k9gZVsaDjhVPHBjLLMQVBaeY+6ldgqTN - MrB+AyoQ2s6UYJXFTnBAEhkW4Sfk8PhZfoGQzLeBLbwDD7Kf+cXyg23BsvuJKD3G/ryvV0Oup/4N - 1TMTYD++6+725H0GDKzlYkWewmaDDxiCob33GOvVly4rG6r/+KML27mhR0pSqVFUQjz2/XC3uw4j - 6FBbJ6p3fbrLwTZCxNQZi8+ypA18Q8QNPtNzS9y8lId5cI4iTGwpIuahrprlnoMZ+tJj56NC0LDe - Wnuo14IrDpBOmnG2Ygb++R0aF2Ow8IELofbBT2IdoqkY+ePGwwxst+CrZ3O8rduzR3N/aXAEvCHe - JBnq0LioADuyG1G+8qoeJP7pSszb6U63nc+AbvEj7INmAyPWjhvUtT4Ilt0/olQCAcSblpA/vrwp - g5KhuyneiWpFVUF+Y82is+77M5PXxbBlUO9hG2AmeDuaotF2rjNwIJJENLqf4l4iVweIgwzWOQjB - 9ogWB+nWMcZqWR8LOjJSCvsVRERRmAQs8Nn3UMFbQvRz2MY0EW7y3/7C//w8z1hTwGp5QZRLGIBV - T5QW5v7J/OPLMV3gtkHWGuoAHC69RveD2PC492lidv5AlyTr4b7fgs16rPGMYv8Nl5ElOD3JxrC4 - h8cm+d0nJLc0PzZbEaQBPNZjRsrRvDRLbr886WWsE3YSoGvkGNwg0G77lEv5cnL/6WXd3xaiWefa - XRM/rFFDyG9e5jGn7JXPOlhPpf2Ht/GSeL9UUq/yA9+1jBRbcX4xUK+rhqjnY0H/+dHNmwIcfKyb - u5gnsQLyFxvBwpvU3X7DsYO3rJ2x/pN/YBGWTAK5mNnYPa+6Ox2fegdf18zA9k9rh+1Pv0SNX2A5 - hvtcsttJl6KfqOP4nt4K2g5FB8/HdiWG3EjxZpSxA3e9jY1G6ujfZymINZM462l1FzzGEkhsw5xX - Qnw6LEnWHbtVrbBl81e6POuq/re+sn2R3D89IB00vSXX0j6DiQ+ojeylF2dxO3zjyZ64ChJ7CUmu - /2C8zTOepZ2vzOxL4ABNZ4GBP9hMxE45RNfflxuh+fVYovRy7S7PpgnRWHb53E/Qpeys4wC29NvM - 6+vGgiV8TB6E7PuM01dQx/R66lsIThrEKkl+zRZpWQsLw7sHLDuIAzXZzoEHntvIiXDWP70KwkfP - Yu9najHluFaCfHridrytXXJf8wTKk/7CN+d1BZMVqTKUtNwhp0d00kY2l3o4XtsT1ug1oMvAqTmc - QNfNh5WTwPD4Ljk6vg/3GYTmja4tY6bojw8ZXbYNU6vEDtjkI8Z+2p8KCsVT+C+eRabnXcJxswiX - UC2CLVMJmOs8f8G/+svjzNaU8lDOof55sX/4PHwzSHn4ueRPfNJexvDHZ4930zrPM3eTKZffXxW6 - vBQNm81Db+hpq9s/fon/6jGTflhKRPm3Sa5Fb//zmxH7OSvBuvegW7FSBjAPe4MYV47dT3TOC7QK - 6s0/W8bx9nbfIXBbEM4Sl+ruWqfpG+FBL4NkX//+lxQ8eGtB59OH0gyL6i4yopXME6+I5IFuQlvC - eHUwDtqtiTf9d2KlHR9weQ+Uhv7Qx0F/+nQ43k7xcv7iFp5qxiWGV0B3hdY3hFf3RIlcpWTYnPDY - wxz+fJzSrqTjqlo98FhWx8ruj7HoTAJ4q0oXR2/rPdA/vgan4zdg9vVapetrhrTiVWJ/eKUQTLa2 - 0dxfGyJHlQ+4gTu+4Y63GD/HV7MWIdhgvtUD9hVJ1Db9PMqwHzV2ZhJGBpwke+K//HMWTqa7HrnL - CKroNhF3rzfNyz5FSHFmjag7f1w3f2JAY0refj+4Gf/8zF3f/+mlYZ0fHYSrU4U4uGGhWDOmD8Cb - 4IVgvbIAfQaxDN4Clfd64aFZ71lqw8t4+hD8DVDc81/ZRlgQ12BLuL7ZoJDr0l+864tPNGraawg/ - W8DPMPq+3ZV7HErwZfee4ZoLi/VO5RFNfHsNuMhs4z+/C+34gU1NfzdLIbYzRD/7h69w0Sn900Mv - 4th496+G7cJ4AbivaYwDZWzBErqlDclQqgGS+ZZu7HRP9p60PdHTQ6Jthu3of/ttPnadGq/Hp9fB - O5+/56PzEuhW/OwNemEj7fwNDVurqglwLnVDzJzttFUa6IKKYO9BzK1dvP7lp3AKL396cZg2/wNh - /W0HYnx+KV2CH3bgseelf/jzV++Bq/6mWO/1wt39rwxqiWNhh0vf2si/LjL6ZcuTZOZoDHSB0gJv - n5uGT5fULrilD3VYJ+IRmy5RXG69/3KgdsDArm9+h/EogBCuNJ6xOn4Cl+b0LMJjr83z89V+aTd2 - xQgnz36S7Pmbi2VVre6YBsI1EBnh5xLySSAM3uD/9f/BMnXAP6VlZvf68DLUawr//GD/cvtowpKE - PRp18UqUsv9q46xjDxqodYhRqMeC1FcIYd2uj4DzOqJNnDvn8AGvP6ILX8tdLWm0Qf2cApz1iQOW - Jj+l0GNVMfjITV5s3y3ggXmUWhKoS+ZSrEo7f4MBVsyyipfPsxmhUKYfrFy0gu743YJXWaVEfwkc - /fN3wd7lfEa1rA5CBU8V7MBJxmqGX2CzryYPj290n/NdD62CqpdQ778G9h6L49JPs0aoz2FOLj/+ - B/75qR2vE6z14xavWUQT6ZKHh0CAKQMIcxh6KLpaTVzjURafNnpUx70+hK3LAQHaVSOE1zN7J8aP - P4EBenkOgjtk/8Xn95Z9k/+lRwH47ycK3NGTSXZOVHfTf6IDe/XABGwtv+gyXG88lLfkEQjBd6Xr - 068j5Co6ix/fnxlPcu3wkNd+Cbm0/Glg0d1PQGpsBXGEVo/ZYZJq+JTNgHjQzDRhkrkWQgGQIDrE - QKNDNywwFZoHCbj7zaWfqpeh9DZr4p+lY9zVm7zBI73YOOfjD1g/gx+A+Ds+SdnxSrw8tk+G5kYo - sMHxRrHKtcpL0Zoy2MCdUbDa6QzhW+84UnSIBzNrxAz6Hpw5YESlAeOBKhkcbnQLfu+31QiUt3q0 - SRokhhOoA71sxxR8ivJHzv23jhemrGfU5VyKT0cL0k2XQwmt0bWd2YvlFixxry9AtPmMb/ctKVi2 - nRcwR8yReO99LJV5/+Ww+gYBtkWubpbHKbCP9Sm64UBPSLyMN0eCUWp+iWnuc6kv7ndBg+5+iS5P - HuCNdH0jfRO5QBjSu8aN2xrBb1xZM5BVblif8JAI1ygoiMX4L5dbjpmEKsjf5kPJyYAlLL/BJgZ3 - rFvPKOZgHcuo/LUfjLdNpH3/BBBm2uERrLNQFFx0NmfkgKrD5lg1YPN9dkFelh6DQ/NSXW56SDO0 - jDkjp0ar48244RDw1f1GEg7lgKucfoGFVeckGcsY8PsBMvR6TWzAWEo4LMcklAFProCYL7EHlExR - gkSZH4nDxwblhMxJ4Q3XEKshI7ibtcAM+jgWdsdfjKkyWzpYcpiSh7S0xcQldYAS1n5iI+3jgTSU - DwVLrity+witu1xBH0D+RxbihhdCV+lhvuHpl9+Jd0CKJiS20cHQliccTpE8sOyrltFf/GI5DQdh - nhhmP8RgkDjN7GELodoiZX5csZlL93itnNeCZKG+kfgQn+j39zl1sLccjwQYH92luNilpGrMLTig - 9+jy69XUAfu0RZK0rF2w9f3sIHVpGuI9QrvgE3n1YJ28Fnw6ZRc6BfQ3w8DhLiQRrFezVe+bA67X - 7DL/bOMabykZOgC4OQ/oIf7Rz7DaATq9vYa4Q0PB+2n/XkdGW5J57b910Vno+kav7nrA2sEzGmG1 - LgFq7wMzi+n8psJjkCEivcdg/6McY9pQJoQZG7bEUmwU0/kit4jhLIvg8GU2rOqSDkaKIgRi9K0H - 4dtwntRan1cATbEs+vo3JhCeuRir/U8ZOO34lKE+RT/isEdN42S38KT1ubk4uZeZywv2rYTgaQCc - HEFXLLKZVMjkIitgIo2hy1ylJpppKBMfm7R5D+OphH/xLr6Ll8vzrJkj74gLbDrXzl155dlDkjiv - mXf7tumu7dai7FfWpKDdp9k+OujB9/2g82EQArAEN5qg7jKUJJjwMqxaY6cwu337maNR3mzl8ZMf - mXfHE79NrkDAhzBC/kk/k1vd5tr0vfbqv/3mNBfYbFpfbpJrRjNRfndxIOF4HpGhXc2961YElk8j - brADzh2b06oWvPx9RDC+cgMuoSlqRM02HqbC8xFA0rwG4VKGPTp82AfW4veL8pIIGVhX4hvfsuw8 - 0PfavQHIKw1X6OXsr+U6KlA5W8WKmziFYNrHBTAnJGH31MBhgYsrQouXFGL8ehyvQxm26HUebXwN - vmdKtqL24BucXvNSFaCZNkhVYD5WY16RILhrmwU9nNyRYty/l5gGydhCCw7pzLX8qdkgfOhwXKQ5 - kJJ4/pdf0cibHbY2xdOmTt1yRLC1Ej+pdE3IRP2F1P5hBtIjyOKN6YD093xJaZJPs2ZqZIJmJi9i - ObdTvD2HSwavU7ViLXJkbZtChUVtFFVEr2UVLNhkF/RNtFfAOcGrmbQgnOHf/XvSyGnr3+fTkZ+x - u12+LqtanQRPGiiJ8VvjZt6UlYXSbx4wfgwiXdhXJyOjkhWc3q4a5dWumVHJspAo+q8YVvVTqZBi - fcRxlBfuqFq1hOzIzkikMKDYVrUqJemhEKKVcuIOS15sUvCQjJnPJRSTbfu84OiggvzFw8e/KzIS - zyHa46kcFq51bBiKi0GK1T0Um/5bHKT84jt2F6AXfFbaJUQ01bDLl0O8xDzooJaq71kIFRZsnGXI - UJrFDzF0lDaEdfL5iIW2JEbp5oBbXqIHx9bfsMmOXsGP4VxBwI05sUrfAYJ7/urAU6M7Ph+XiAo5 - uo4IxTLE6eGbulxgn3MEjFtMohQ37npnVRlFhmfgTMpybcl/oo5Y8wGDMitVynrp2UP1I+GIMiUK - IJdtTVFoq1PAQTEehNuReyO1NiJcTmYZs4HhjvDbW7tjxL615Z7HHQL7VKX0dm3AsjcihZ99pog1 - mKrL40ZkwRh+v+RBGKpN8CQlaP1mKY4S+B4o/3pX4PVNY+JeT5eYNfdeZLg7CQTv+CKIX6EHpXS3 - 9hMOVrw1/bKhV3c5kIukT3Q9hEwLYYkt4hbhrdmMdG0RCP0bDk5z5G6N9vXgww97HLcAASqPqQza - r/jD+fnRUbJ3Q0PvE0sCNHwaShzMyFD2bt7Ma58ELKE98YB5yD02JN2nbK+ZPBSwtxJDa+KBZk0i - Qgr8Dyk41W5Ic7+0yKz9K8kF/hCvs6iXyPlkHil1VY955/mAMIXTDVtYWoql+2kQ6lP4w/onN4Yt - PuwdwfYuWoplKmDNV/0Ffs+fS06hpoOFccwNfe3VDZDyfbh0POgVfKjlh1janBdUFOsXLO5vntih - I8WUGZU3+hhsha/aLBUbiRuIHvPJ+8f3Vka6s1A05RZHh7jQyI7v8K0HbxxIAgDTtbdDeDWhThSD - asVyDc+ipNg5IOrFKOIt0QUIx8v6xFgS+maNmrqHm2YyRBM7oK2XQ/ICR/W7kD0etC1K/Qo5QL6S - y0nPgeCKYgeVSjNnxCW06eXwDJH0GweSjT9n4L+01SG6ZwvRz7Ll7tfPwCsC6n6GW3PJpTqN8PVN - Ynz7OvqwzedNhPUj5XDwcSq6RsFRlZRuVrFfSxdt1F1N/Yt3rFGZDOs7GV/wzK019pTqNlBweC/S - 4tQGuUlnfRjEc8rAUXZToiWMQalelgzgvFDHbsdHzXasGw+66XwIartqhq2zjg78w6M/fB6BTt9w - enYmyT6z3LApNHk4guiFT4ozFcT7bCYUuPYyH6XnY1h1EJnQGrFBzPct+tuvpjR5Jz3oBvwdtjCU - IrgYx5aYe77aDv2bge1rJCQaHRvQZJUj8L5YJrHeBY1pRAD7t74BOkvHYnPs5xuejuxMrE0Z3fVn - li/psthh8HoIQbG950hECq1tfALvRqPe2algLDpb0HDspdjWX6dC/VW+ZzH4xe6Wml4Op5aViV6R - tCEiViLYDfEXyzy1Bl5BcQJhK2rE2J/3lsyPAMrjGONKqX/N9jv1HjgdquPMgqAFXXcYe5i8zwNR - 5UKh86qmJXTaQtz581ZMenhaYJoGIzklzklj335mgly0OKJEmI1XbpAlcOPEmjih5xTCIWd1cKU6 - 2vPDcdg6VxChPM7xzg9v7ua/LxU0qk9J9HuZabxwP+mQcI9LQNCwFC1vdhnsLr6LXf10BAt9czPs - M3Um2jqtLpV7OYGy8LoRSz+YA99oxIPaSjwcVJilY7WwCdRFnez7+zPQPJA71OH3Yb9fk/K6nEnS - 8Os68pDTpRk8GG0IXhWfeAGatT3eIDifgpR4F1F016v8UyVu2zas73x/BfT8Rq1adFi5OUyxeNks - HX8wMYmcJ9d4SY9HB7BymJK72qsD/6ef0udD+7ffxkfRmfD2OPjEu5BnsUpm5oHLQgZiIS2Pvzk+ - zlBNeED0i+XGyzE96XDPv/gqQt5d/Uc1//ELEoQ/qI3FxLPAbmSJJC8+pyTs6w7dlvCCcynvtcXP - nAi204slxkoUjb1+Ti9Uk9XEp+TsF7P+W2zIzwkm1a7PtvzRt9LOd4LWd31Kh65ZwFH9LNgoVd1d - 2nDzICi0KuCCiwmE541WkAL8mddRaQqCNJMBgrThGV6IUpDbx4DS8R4OGO94tH6vL1ny7V7GljV+ - wSbQ3wIEsIJgUC6JtjFoeMFdX83MMVHjidMOqfTHX4KwNgrB5oADstunn03u0IN/fOZbcV+M14zX - 1rIQI+gTe8LazQtiWrRRjhwn3LuSb9+4xsU7ByeidsTO8OiuUdP1kKNYw5qk3umYNEiHunW9zmvu - P0EnDvdU9Cyxm6XTWS4EPDIp9JjiQrysfIFVaUYG3kG3zFQfhuHL+yyPvCw5kovVX2mH7PsIgg98 - kfjLPIuRMWUZ3ovLD1t1gjXyFUceHaOy3Xt0GQ3XfAcdTrBy8SkVnHjJR38B54B/Ej3wnvF6010H - DoeIEu+lB3/xKUP/NIzYZlKr2eKDbEql3nCB+D3nLnH2qQlXr8UBw3C/YhvDuQS5mazkfnU/xWiG - VQC25lBjp8zkgp6O9xD88Ze/fMBL/ruDf/GpF2ChlPJWB8DTAkT7Pexm9wtGuOsfrNWWWEyHoBP/ - 4gmnu97d4rG10XJJXWxWa9vQ4AClP75IdFV2Y1Z7SD0som39x4fY1ZFE+J3nO1bGcxWPO/6CqeVl - bP4oC8aInmt0I6xGLCg1wx5fDPSlQp9/4K1p2+2+1tAJryXxj5HibhiS4O9zID2PVsEu97WCWBmc - f3j0wY5awcJ65VjFdwL6R6t6MKjTK5ENKaOL82Nz5N1riq1DXQ/L6kgSBCD9zC8tqzT2BDcGQplZ - sSKmi0aLNs/gVl0TcpLOesNjSDyoR58ZK2vRahuT6NEfXyWqvQTNvPM1CIYHwfqud1d/Xbp//HHH - a7p11Y2H3nTQiXYxepdC5S7CZ/h6EOv+9ZvtVNQ25FArBPW1Xpu1t/0WNL4gEyc6Ds1ioUcLmf6s - EP3Kjs182542SmIux/bXbgridniBYVYyWGk436UJvTNQXN/6DFs3Kja1G0a48D2DnZochu3RqgGk - 3Ccn5q2Whqn5NrrE2DDY9asdU3TbZydZFsZO9hS0dX3/9p4jfIiDHX/mJrdEiJH3xckzVgo+aIMa - 5Nu9xydFO2mcWLwr+IrODd7jDayPF6xB5VlKsK1vt1kkzamh60pHYhv2qG2nTWbRNtMPweHPoYIr - PUdkya+KqA2zDb2pURWddbnEVoJRvFyk+wJlZL//4fGMP08JvLa2x/41Kd213EIGuZsgBotep9rm - PK8QfH9mhM+mxsSLsto2fLD4Ny+Pj1as7WWBIL6G4Xzb8ymXRLEDSnvUcLzH1yY8LRad3yITAKMU - 41+xoTd0HouMY+3cDVOCmhKJm/vA7in/0CXrmg6lDryQy+160dZuVue//EIUJFxdNhHfM3AVk8WG - 2A/uxF48HszMr8byVFyLKRKzCIJPeJrX9hfR9dlWPdz5H745Q1Js5yTJkP7tKnyP9ae7yeGZgW/d - exP7nLzctS85D5qSGBGL+G3B3mJWBpGiCYF6dyZ3cdUggHCdT9h5GM9mvcpPGaqn9j1nl8IsuM+B - ZQFrVGGwaXajTctLDKAkORC7ERjpugT2G/z5q398a/tWTAn3eCfna10XK2VGE0oPjZAg+f6G7X2V - vH/5PFxvOmBJKefwgBeV3IZs0tZDyL/RaW6v2LiE4T+/BOz4ju8f5RYLnSh7iKO+hu9hzQ9T6eIc - xKNY7PntCbbyOGWgncyGeAy9FSuuGweOLd6C49d5N3sFZ0MafJ5xeK3PDW15uvtfTB9saFGaTYm0 - fWJlKJM4cmp3+SLQQvk6NMH1Oa+A8EffgfjruNhnxobOp83mQXYwBxJA296n5MkegkqhBvPOn6mI - lRDdLp/bvKJZLjaRi3jYW+191/+/uBfsWwWu6soTf/VJ8acvoS/ddOwCfnZ7Iz228FB3n0CAUtNQ - 962oyErf0p9+KKj0LVSYm+lKdr4OJla+e+B7sGeszAKI50X+5H/+OHFGwA00Xt8SrExgzX0rD83u - J7ZQ6Ub1L7/TPb55iFQnJ6dWLun25+cYjNjMHO4+u5/GRzAtWAGHXP1zV+QnFdz5J3Eaz6B81T3E - oxVCMotlGhb0aWgsvH1KjLX5OMTbRe1DOB8f+6j1pS+mJbJSKVR5SrymCsGqg1yHjH4lwe+bPxoB - K7MMmoNL//xAukocicDGdAVOE/MKVncZA6AcrsfAUzulGLPIYf75OeWN/wzb4cIl4J4vwTyjVw/o - y2Ql6G6cuL+TL7mr8gQi2J9PwO/6e9r5IXylkjpv6YPR5ugwzvDmXNWAQS8HfJlp6+A//Cp1I6YO - CBKw5z9yBtoaz+DD55A+aoQ9/hPSbc/P8Jt1ComjHLjr5Lo6yEWDI2rx+4BtudxV6H4/T2zW0UTJ - Xz3BXXkPe7cDjEcL7T1lLA8TTWF6d5Sum40M+3Cev0jLi1XI1BS6T5QSwwZvjZXNsgS5YJ6wihZl - YP/qEZyYo33O9i1e+Fh7Q/xqRuLcwQ0sj6LWkXRIeKJ3wRCv9/svhV5HHSLbc9UsF3GfOihfKuwP - XEWnJqIV2v3oQLhE+xSlvf5yN2J/vghh5/7hFTykN4n4ratov7u9jqjP5BnLaF6bMeuaHmruegw2 - T4gAPZBmhNN5DfcTPjydgZglf3qTuO3prC3Jy1Ol6kt/AQQH2lB1elVwuK0bOfmPXFt9TsjhJimQ - BKhoh2F/vkgIjYTo3lGlf/cHQgYN8+vu+O5UPd0S/uGbNu49EhgnWOCdVz9/eNj8ZtGrYJgBPN+d - A4nJ7SVGYMd74pvi4K7vX+n8wzfHC0pA/e/gwFa9dcS+FGZM0aWWYP7FccAWIKRTFKwqNOL1ik+X - 9wD6zlodWJVNRCyn9pv5x241FEIrwf/qLX//F1x0meTMM4ipvnpveDltI3b4shxIeop0aG3nFz49 - FwVsf3p/azJjFh/a2lDpG6vwkTXK7HnhCLbyjRZpGbLvPLEK5/5Ky+6BKFxvAcshiY76XnEMjKHA - p+hY0p1vvdCJyB3xle9Bm5cBzfD6ZduZL1ANNuDbMkxuSoItxb4Xi33ue6iSn4zzbpq0OWcjT2pY - epgFl45uZ4heDs8nL8Vn/Vc0pM7UHFVKBIn7CCyNKpdshuaglNjQSlzwvg8XwHmRTnQ2fxbz7m9L - Wlty2L4UbUwjfXjBFh2MAARs16x//kqmoQd2PKN11+P9Xv/x590/KdzFahkJhPFZn1+vNwZUX/U3 - GEH4Il4f8cP3ftEl2BHJxLh4J810I3EJIXpKwcF+3zSy7284XYsiQLV02XuaHnN45yue+EZlF/Oq - VhWY0oMTrLgIhp1vQbj7wbOw69nVnuZZ+uMrux8as1M5qX/6NlhZ7FJ6VqUSALxaRPtkesEawceB - evSdsXUuhGEDviwj/Kxl8lgz3p2fCZyhHb1/+HZ5D3RdgQJh+5V+2Oe/mzvt+RFelmmYOjUxC8G0 - 5BlcpFrHspsUA4+l80sqGqxO5Ax/xT88ZsP5S8z7fwAAAP//pF3J1qowEn4gFjJJkiUzyBQmFXeA - yqCIMgTI0/fhv73sXe89CEmlvqGSyqPXqHydc+ljU5fIpahq2zZEOtr5OLGt+attsK3mP/9sJotx - HTb/aOiQs75vYvcD0hZcvDK0FO36r/6yXlVmlCKOVsEXvFqX5scQou8numB1Xt5gcMXlh4bk5QV0 - 95PpdjJjeGwkAbtPS6VDPQ+6ZNitjOXdv19CM7PR81i0/8b/m6KhhDv+Ez3OT2BTE9YEZ/7M49Ny - TAv+Y8o6fNQvjWg7nk/sWWdh7xQG9ubw5a6X8QyRoLNoRo2ftHt9VET8d1rIDQxWQdLVDkGjCDG2 - zcIZxktMXn98cdcPcrtI43cDMY9Of37Dv3owUGhjB9TKRsonV775w4tgQk2pbYOY9uCw6jw2GDd2 - fzAWWDj4jEzsCZ609TLeGVgGaUDi3R8crcBuANdcanzqBxeQ92nL//ykeV1saxhfr3MOvdY+kFN9 - FNsxch4i3OvBwcFnCjAubGz/PQ+rF6y5e33oAh91pxHbv6TtejRZFlrGmOHcZZSBPbDSQyqlp4UD - hjsVSysUDLz67g+bs3JPlsDQxj//HV8Hs9EIVM4SUvXsR557fZVlrKMON8ZzcPa3Xk1LHiEUjmQW - vlBKaDQ7G7zUd23m4nJwyV/9KYexNi/r7QXYthlTMFtjQ56sZrj8WW1CYLyG48wkvtlO/L7j/f/Y - UQD/944CTR2ORPaWhq4vOxQRuVmPgA9PARAYHf5gOigxPrmVQOlnIgEULVbCbj+ChOK2CtCkX2Ki - qhfkLrc3m4GVXBE5eaHScu3tKcET1mViaT+r5bRqUoFXPZYA8N8xWdOFT+GlG0OiF0ZTTNH1E4J2 - sj38MAq53UQk6fBdvD/4cXivgN6FlkfM5hJiz5ilBKZ1ioJU+2Jv0G13u1zzGJCb8cAZ39wGcsSL - g0TGGYh3fFyTZRbDDvkCZLE7On4xpmUsQyZ+kFlcJ5PyVh7zkHE5NH8EtxqWF8xU+N36lkSmlGqr - OSQ5KtsHIvJdzIulXH4bkKnqEkM0kkK4uY4Iz0Un43C48wNtxd8IvndWJ9pbzhO6khcDRY/JsDcw - Pzpp2vADi+0csal6qF2H8vuD2cH4EOXVeho3x5uNlu/VI9qCNZel1uIhFLy4+UjOjsYT5djBtHLW - gCxZMyxMZDzgmQEusQy+cQXSZDHilaggxvA8tUtKHwFQFCxiT3yl7dJ+NhuZObxhp7k9iuXntD3S - YjXFJ8hdBr7WOgeZFqLYj1mnWNOEhWiAvI918fIeOFw/NvDo69Pe9V+kVIwfJUSSapB77nuFkGSv - EfJw/JE78ywH7mA6JvKX+REwHX6AIVX6Cn5B7hEvV45gzhEvg/XYScRkJEZbSD9I8L7hDJ8C5gR4 - MWVSCJ0CEkXEMZ2Wztu7ZrohMX7vjvbGhzeRlCEQHG7BCQhyNXRgXn57F1qtGKbkLDPo1IpPUu7x - sJa3SYbg/X6Qx+9gJ/P1tReE5vMDh4djB5YniGwUta+KPMGqF7xz6St4TeNsRuZpAV//8HJQUESH - oJGba8tCkrNwMJQTObv2N1kF9vqCTVG4xNhEt90u9nG/x1nysdWnI2BHiZHA66qfyPV+z4AAgtEB - h+xekUQ4Vy5voVsGI1PNZsBdU7q4Wmgi2TR6ErPFu1jL3piheB7VYNPlU7tFljRL43oeg61xcUG9 - T9ygEbx64mmpARYYRw/4S3h+ht5RLdaFfY2oO453nPLjUVswE82o8+wzSYriTvk8F0q4xw921t5y - t2McyuixLSYp6XWkQ9RHObpWV5/o9SzRaTveRag38X5GOQkKFjO3EQp3CWHDy2ttO7NFD1l0q4Pj - ASQDH2inBl7dZibqzzwkayxNLITLNOPCDMKBfY5LiW41DbBv4FmbdUXbkBC1pwA6k0Unjms7xKex - gk/ehofFuN0zQP0H/Rdfsza+ebSNYYV9q7ALzgxyWTo9qDz3cfhNKLBnBha1fsVeWN7blSZHGfjH - hzXnEjcPNLu1r3/jUTBwdVdDvfb7Hr4bDlJfb9nekEqos4ZETH3xgCByZxVGP5Ynbp3cEmLw9gUC - sdOxHZYt3ZxLVaF+sB9EceMe0GNW9PCmmgEx9SzTaPtTVbSMh+tMQ7V06fER8uiXsDy+PiMmmS/X - OIZmztyI9e1eBVu3MoPyM7RwQGNf4z9TtSHxXajzfOXIQB/EvsCHNg/4ji9isfT9UYWsZS1YFoKp - oBW8yyDeUhCM37AZhPtXs9GRPd1JvvaWtsoiF8Df+3LC8nC/DOs3AjHY80GwCJ3qVty32eB7J8FG - KBzcWX/4Duw85xxsE+y09cseeXi/2C3BS6a2U3t7ilCBg4XVtEtctk8LB3XdXrFkqEKXnz6XUDge - NuJ0a6IJFste0P5+OBB+NZ3yfFXRWxUYcjLWJlm6h9MD/7LvseSb4zBp1VtG5MJ02H6ABowDdMS/ - /EH8OfrQ+TS9RsRJFz1Y9vxNnFT+oTnzXGLzzbFdpf60QD9xLrMYli1gfdUzpY/4EAPmAjzwt36l - bb8X1Ti0hibEbib/G/+4vZt0cWbJgY4ThDO9HyZ3qYPBhNUrsPD5cqzc+YuLHB7Tx50oqfAutkPe - ZZJ/wA7x4l6g9K3d7L/1MzPMQLSt+iwzDJ9zRxyJC1p6Ok4VVCchD6xELAv29oYZXLPsQ56fqQeE - /OiG+hN08bWYNZfrLE+ExcCIODhbn4Tm4GIj/AHNLErSXCx5IAYwarsK4wLGGhV6OwUf6bMQ16p6 - l3sdFxtRIV9mMJU64NxTMqOZ25L591wMwBti78GyLRGOTubc0v5Yv9BdQgq+z/QMBFYqLjC7g5R4 - pc8Wy/kQvtA90gxs9GudUCXkRWR3BxgcRnmiSyUZNnK8qz3T000byIjMBfCz0AQfxpvAkpa5DKvt - xuLk46UtC700RPmHHjBu0LMgQNZK5NSnDwnG6O/MgMzA/tMk2FOqqviBdy3B3u8u5JKVybBeQ6dH - cxa4wSEmziAwfCKCHZ+IIYRVIkyd+oIaY9lYeTB4YKXSZuCFPaKA99+HYRXlWEW+EQUkMuUoWbms - CVD9tnLihexWEEs7m+im7j2X9nj4+z9I8tca8GJkgFWLvRQYllnhx++TDJPf/Bb0ycUFmxZ/olse - FxAOSs4Q/61XBTkPTCcFbabj64uXwJq09Q/+4bf6mFAxoW2xkTIyFrlx7KkY3WjZkEHqE9njTxsf - 8TlDO38j6fSMWnLgLBHt+IdVFxyLX3Ha9/zPXxNrzKxqL1fLdOQc5JnIXP4B6/OVZGjHG/LHx8jn - rc3opM5ecP/Wj2IdjoMKPFs1iVdwz4Ri+fv7t/5vyB8SmmfXF9zjK0CR/AIL59UZarir/bfe/vBN - R3/5Ndrz4zzmpwVeq/N+oveqgMlO3ABaSugSXJ+OgFraXYf7/BG74ze6/p4YQqH7fbBvT5W23lbk - watZhERv5Tjhq484w2rYb6V4R2WxfZ6yyXwxhCQdkqvL+pcaolTtvti2blyyor0ppRlOdxL+2gQI - rieU4I6MI9757SAYdVPCD7S04DvVr3aNriQEw9FTcdKxBzAvm85CmLgdtu9fN1nKZe/6x7tuED6d - FUxQEwL0x8eUR6kMQrRACX6SX45P3xMB8ylsL1Dq00tw+Ptedth+EBz4nJzqZaJbUDkiuNSvbe/Z - o2icUgAWtsdKwYHHqu00HFsVvldw2ytMfDLWrQzR7fA7E6O/eS13huII30eXBDTdjGFrxq8Ed74z - X198DqbNB7y0nQcys0v1AKMZHiVY86+JBMC/a4S+USl9ylyb4afjteUw9A3Y8yPx7HoaFob1ZPCX - X8to5rS1qb46HLX1hd3D4+AuFsAQbP3vjP3p+tbGX/3hwSuf00AUvU5b72rFopCtDaKB7aFNUq9s - MHUvMdb0r+TOH16PQXT6tcS4FrO7+kMQSrUcQhwctBasqX9YJJfTKuyGbg2W42RBeOe4FruYjoCA - YWqgUr0SnHndXVuqYXwA/M25oH2Z12GSDC2AshTX+46Xz0ALDE3YvvIWq8kqux0XGjZ8jG46Az13 - NTaswwuccVmT/ObxlGaNpQNNeqAZHB5Pd+V07wLvkWJgrxaqYfvAUwnp4uTEl2Bd0BGZGxwl60xO - bnWlPBtFDmzeP3ve/PcbrF8tfkEjeZyJ0Xo62CgEnvTHH08HZQHz986aUAqyGwlLRhj+xhee77OI - jRMfUcqXPgPfP0bFtj34rnB79svffBDnM5wpWx60B/x7XuotZUtjN1SRTGUX6zA6t9tFyljYOmU3 - c3fJL7g/vTh8RoJ1316L8SblIQh+vxzbL/s3bBd7fUGAo/GfntnCBy/B39g9AymLU3ez0C2H8XYB - wfEa7Wcys+SH+KvokcjJOncbBdeGEnmNJLJYni5HqzWho32eZM+PgBq8nKLjMQT44c9GwbEMNuH7 - ikqiPa41XTnnVMHTY5VJRo+1O32rAsJ6gw1+fOpLstLmNkN2a2Vy0p8/upkHYUZWZ353fkJ3h/yo - ApJcCmKary1ZW3h5QHJ+JrNgLi9Kx2icYUsSPeBbjkt+d1mG0FZPD6wdGXFYplaaYcnCEFuX0h1+ - vFX3f+t7Bu+T7O74n0MDZVFw8zbSUnWIIHgm8IaNj8XSqf56HaQuqmaYSDihRnLad2W2OdEtpaOz - md578BePZsb2xebi6wWmbhr/44fU0L86CI37DytfTmt3PHfgYzylgYC1b7s6kRDAPjuJ//jHdFcr - HpQsE87Snk+mku9y6M7mCeN5vrib/lJTVKrxj5gm/LiTfxht8Ic/OOIsOh7smkHip5YIfuYvd32K - xxgo8GsRz5k+dHbFxES3ThzJ4xlE7SL96h4yD7YlKquFw5++hHizn7jY19N8EMiGatGUsT6itt3z - 9wjZ7uFiq7UMd2mKwwamWsoD6vonjRfWwwyncQPkbzwWiEAFztoTYk+xh3aJkDFDdWEVoqHSGbj1 - 3ldQ1pglWBlaA3J8ZDx8dFlEbrve5Oiy5Si5urcA8S9vWOPRsCW+s44z9Olb26TnawSGMw/EHCui - baJ576HcDgsxvFzRNk4ZS7D7C3/vP6wHob5AQkaMsy59ATpJfYlyFdUz6LS9S6XVmpJVcGNAhXOl - reK9yKXUmi1y8p1X8ec/gPjelNg5Kidt57cSND9pgJWTueuVXg/RY9vMgLkmSrvVFWnAV1gU4t2f - A91YdSlBXrVfokUpHdZ4cy6AIAADQMJfsV7BawRilbyCgyTU7cwfbQnaZZ5j4+Auxbj7D5Cqrxj7 - kWQNlPnQAHnDNcIW6B/t6tl4hm40noPtmoKERk0ZwKf6NLGzt0feXLHQwTEz+1l8KJW2Nac2RdQM - 7uT0iAGYSte7wD++oYNSHdbPYveQKbwMR6qhJGyUYxvODZ0C/m352lbxUY7sveSW32FVrK/9xMUe - P+SmPx26PDR2/tPbM7vr0+2d3kZIK3ogOLon7tDbzgVydVeQUzgdhnem1BCVB36dN3VeKWFtVkc2 - md7/5m+pW5sB21t2ZrT7GxuzhCEKAz/BeGwA2OYMbbAiZr/zNTlZDvYXQnXhFawsqQqWTQEy0F+Z - FPAVERNyi2wT/uXPM6eb2nZfqwA6Plvge2iqLlfngwgMMaTYul6ChKZK38C03SDGux9B1Xe6QAA9 - TJLxVhb/+OKOb8Q3a6Pl4vv4kKIxPZHHzzBa7pA2C1IDOyd3C12S9eoYD5SeATunXjC3xL4vNuzQ - pSc4t71hQ6+z+Of/BWs8OID1L18GoTnisOyafLEIUf6C6/El4XM6vekKXDuDQVg8giMNpZ0PVBUY - mndEXD4SBmosmgMtvAk40LC5n3mrKgTiMCGp1qZ09x9GaL0+F6xWxyv4jofKQ698TP/7/DNcRoSZ - KiFF4c10XD9PCe75mpwauxl4+xbw8FNmGn4sXugKD0/ZYHyvSnxpwDZMvcuYwFjlO3kKALXL8FQa - JNLYwca1CNzFttQM/cQhxeZtdt111qMHUNnS+/MvB3b/P0D4oSNBvVWAl7/yD5LDDRNdueYDrZ7F - D+z+YQCMsRlI8/JYuPtR2Ep70xXsMePhc1EComo9207hdHOAvXpH7Of+mIwd/kIA6kNClHhw6GIX - fQiDSFqJPj3Xdh3ehgPP+gETzXvKQGgst4Tnp+fiGN+GYqnX+geVp3kgjvpswNPqkh4djzGY1/oA - 6fZaExP67OWLPbv2W+EilxXMBTvHT0vk22HQQxmaZxZg4zvoGhU1tMCZjWTyOF3CgX7CNw8v3pZg - 54Wmdp4fEgPTdoH4KSZ7D7MpspHJRC9i7PO1je7XhB4OS3I3cOCuLGOZ4I9vvr39nua+rAI0QNbH - 3jsv6GKL3x7ufImcmG+537yRQERGUw+afbzW+qu/YL0xzczu/uPu/6lQrKJXQILrGUxpc/7BZzmq - ZNezLv8WwhCGivAM0qymyUhWK4cyksZ5QL9bu9RpFf/FP9Fb6awtEy75v+/f9U0EFjl7VMc7rldi - 7Xi/hg/xAS8sQERlUVJsJWvutwDPzQxHeqe9GhAV7H4xvj/eisYy8NNAkZfOxGYvz4SqdpFDtXvr - xNLimFLc9h7QVpJis1MI6A92DfcelzaxGl9NuNwLLzBdf/r8p3fJ5+2OYNWSU8Cpd7LjVciC4Rio - 5A9/FzNcxT8+Sx7DcXDnk+r0UOZOGlam7EE5aXmHsMCiRqL6UNIt0eUcJm+5xqH8EQqCsXqB5Xx9 - BPZDqdx9vDtwmZgOK9l2LkZv6hyoGvmKXaEl7uhNswN3PRocXodqoBczmoEbYhX7n61vVzHYZukX - BgMJ1LOW/P7Wiy+7C/a/NZOQ5oF/YB/PWTyhV0utUxzCp7gIJDshvV3+/u8Jbyo2vgkuNhXmFTxv - 9ZWUPjXcZQh7VrKDOibBgy//8kWAHNb2//kz9MBhERzvgRmwrRwXg6WddeiWHU+stU6L6bC+bMi4 - Apr5eTwlVNZ+ATw0453kzyzS+D++azFsjDWlH8Ha94cH6M2LTHCZWtofvoDcwcI/vbs9a1rBfafi - 7j+plD2GZgdL33WJf0u+LU1DxYP802mwv8fHkt8zCQLxpQdsnPd0/eNTf/4oW3se4G5t2UAfmTFR - HmXdjtM5luGcvmSc/m5ssvxu3gWGAU7+8fkF5hD+8Rtcaumbbsi/zRKuBo/ogj7RUQZn86ierXXm - Eu2WUGjaPfCVr/YvP237JnzoBmQmBqeb7obOXICC28ki2CnfdPHpkEJ+ufABKiyifQPjm8HvqKwB - XfYTIzMnjvCKnSP2VndrpzADMRzmg0FOym1qv7v/Cq4Dnon5hU9AdWbg4fvNv3Y/9lSsxgU1ELvs - FOx8iBL+iC/wuNTxH94l4+4fgb/6g9kpmNLidWMglx++5J8/WML2hcR6EbBm3g/t+hh5CerVRojN - ZHLC/fl7ux7BmuhQdw4vRQYc7f3Ezv01DPM9qHu0v09Qj/xC1+14lmAe8t+gBdvDXeki5eDqD11w - MJy5mMrZEaGbcD7R36963z5LGvCSihSrQ3ahu78loRF0fbAfIByo3FUbitTcJ8onYgtqSWsD9nw+ - s0MiuOtfvYygI8ThGf2S7c8vHJpPFEgPvtW2HPEqRKgP8d1vNjClvrChMQMqCTy2afd6hgpY9iti - Qwjlgj94/ij1nyoh2YhQ0fnGhwW+DW7YvWXfpHdc0EH/WTnEkL+/ZEsK7wUPSuUQZf7xCZE5pfzH - l9TCUV1uK8tKMttZC447X1rX8eRBzC0/4pqdUfRPcHOkg9I4ePfD6Zp7WQr+1pPMp077z6/484cU - HhUuXXVHgoLCnwK2OLy1gfhb9pdfZ8kdf8nyGZ6ppJ5USNyxUbX1YDo63P0AbChAoWstTDFY2WcZ - LIejCdbaDGSYnxmLGJVpAUHUuOUvvnDAnBSXbXX6QvibccRYO1br26SM4e96V/b6YkTn4TD3ULvF - Pxxs8nlYD6Zqwo48q7/63ECrd8bCKPxIM5gv1sCmL1T98Yng9h3ugHqfvAJX6XUihVXZbs+K2gPc - 0yEMnp4LhsFzIhvtfhEJrhxud37IwOR6upFA0Vy68qBaYN1PbxK8iltCzepSoi8fzwGSoJJwTfU1 - //Qh0RuTDOuf36MdXgVxOfJt16rFORzPoYA99/MbFl0/NECKpjdWr+OiPXRLhvATqhg7pyrUhM9b - G9Fc8BeiXcW9fghJBRJVATPqxDf49r1QwrY5htg9S9dC8DjqQctzxoAPBMElb3bJECEzxmrjkoKM - 0WtEvx6mWFVY4g4ouaew1IuGGM95A18pGUTwteduZsRnSumo6SooYdD9+SHt6zNcU1iKXUHk8/s8 - LMVv6sCe37EH1a7d9d0oPk+kJH94tsZyqwPbiiFR6VC0W0jLBXS97+x+5q2YzYa/QAcfEuKyXEfH - Hd8BfiVXrHsLHKaZgAb+rRf7E3yGv/oZ9I0kmKW1/2ibx4QhsvnLFeuBymqzbtkQXi2xxY6pMu6o - nBcPFlqDSXASmZZGvKSDaLycsHfN6mIU2JsNMvAsg2b3EzbJDXh4aqUn9nd+vx3jTAWK4os4OSkB - pTtfRUiSDVL4B2vX888U7PUQvNfHk7nJdBmW4qvAumm9tHWuYx4F5waS4J6P2uKgmEe191uI9oxe - A3+aXvNf/ZC4QKndlQf9Ap6f5EXU9t6B6U9f/R87CtD/3lGAnl1GzMJUNfbpiSO8LD8JB4i/FdR0 - tQcMX/GDaOx7SOgQxAGqh1DDWcobdBNCbUb+8fointCSgTT1y4G5IT5Jcl0CjcO+rkMreWyzKD0F - unnM4EFqNAh7J8YqaC7iDD6fsU8U7TIBauoVi/rM84mvHcOC+NDwoPy8h/ihpXKxfPSzAy1o05ns - Z08IzTIbpglPgmPDy+2W388N5DXiYkd/xmAznMyBYCbtfgZzoBvpOgax8u2K74bgDgvTHzs4bVlE - rMIb2pVxAgeG7eOE5eglF9z3RQJw6BBPTiH5aYQpDjmsxnc3MxGntMtHvzvwMLwCcgdvXHDbR9Zh - Hx1crBeZVggL8vczX9mZxPM7TDZxvDnQkJ5forGZqQl5pHWIEqHHGs79hIP3xkTye0xImgf3YtPc - /IV+HDuRSwoUuiB1CcDmPsSZGxaN8mBdL2hS+JaozE91N/7nxMBLnSdWa85whWP35ZFixgoxblgF - wySMEiy+DymAbZS16xadQxR5zDOQprBzt97dK8bM+MQ3bvUK1h9XBhE9VXEhhBXYNDd+oTISz8S8 - kQtYpaIIpMlZXRKUn4hSoy9S0DvGg2Tu6ThQ0C8jkpk8Jfr3+BrYmvQbetWvNw7GYw8GGa4jglMm - k+z6lJIlMz0drjVsiGoymsbTV2MiZcNX7LC4GNjn0gbQc8v7vL6hTNdVz2TIALUK+HhRCnZ62xfE - 1JVDtPb4HWh66BsIcRSR22UyXMrbfgXUdT4R6+50dBnG6b/xl32cuiVnb2IhP2t8wL/FuKDaQX6h - /ssKJHsWyzCHTqfD9tNUAZKTwl2O9+NfF/4Ia9Kl1ngPRCISLS4l+T27tZsUxy/0YObnzFdW2dKJ - TTpkVIcE2zpzKwTefbASMLWGaL9HOvDPIvyhbeQBwe5Fdnk9QgHknGdBtINzcTdPtDvY/74euQXr - TOeCj37omDkFcWigaLQMTw4A/CecBbHPhi3l8Q9G574myW840tUu1RiCOwjmVeKsdlO8s4xkIUX4 - IehfbalCVUXDcSDEe5c3jXfVgwyVGapEnTeuHe6XdkFVYoXEztmKTt3TUpGzdPoskMGiwv2GdPgX - L8Z2DgArvC8h+BilhbP994TtuxEZvfoj/gNdB6HIDw38FY1CNAYtdFFYRUeb2oT4fLryBZVMu4e2 - znPYGh5quy7xdUZnrbOxoR7CgRxEnUfXo/TFJ99nio3zIQsHJrQwHhTBnd7wIqF8FTiswqNX8OVN - WeAXLD3J9edGV3o9SHB4nrXguI/ncj3UGRw+wRLQx/nmzqwVe+g99F+iXKLbIIhtWIHc0B0c/GpN - Y6vL2INSfwPi5txc0MG1Jci0Doux8f7tXXQSHn5L7kWUWydrbOGOMTzmnzwQLpui9S0nQ2h5RT5z - UeINyzzDGMpd0weHbCoK+rpnOUISuWE7/VbtKKe/C2zmeJkPERhcNs2bB3JC+YLdj0s0chtUB/0m - e5i5qIroyp7EAPLt44WvVnpwV+IiE36u+orP+mIOgrScN6hf9CNJoi1xucrJR2jrakV0+Bm0+Zwd - NvhqMcZ4upVgQculB4pY3ojznvNim+Spk+q7m2Ibhj4VavGWQfcWG0Rz/V5b08wrwfQ2hBkdtrSd - fmhIIcO9KVa0RSzGwMs2FH+6+A8PEhLVegCFcUxIqaqRJjBrE6DlsJyw3KRD+xqNbwf3fEkUal5a - 2qJ9R9X7edvXDzvsz1ugy+5dTo0GAPI5sQ/IlzeP6Ev0S2j/TTwEgwckPuefwKJ98AbB/RgQ9eca - VHD5SQWn1YhxvgpnwF6/zOsoH+l+b7oUa2vwSyBsY4tibfxy2nrokhfa5x+nIDy0m/f+mFJeYXc+ - X6PPsC5DwsNt8CKs5LIzLJdbpaKHZeXkRLy62OZbPYLUtSdyDo+qtvHp9QGrOAT4GrKdtsyusgAm - 9RfiVyAGywsML4m9ljO556dmWOuuspES3xnyvAlKstzePx7StFTwpQ9Fl/bfwoMi89OxLuYG5RiT - ZdH5ru9ncJjGpZoaMeDNpgN2uFvdjv09CoH00z1SePMVcBlfLYgE+oGYddQMnHYdGziDn4zlpg4B - Z/IyC7ckfeOndGzbv/hAt2S08C1YA8BSvr2gkkNMMLaROPT+eISgiN4Bvgy8n2wvPxWR6O/4hM9r - st7KxkNOP17JWd2GZJs0XgLksirYld9lMeWtx8LBul/xWW/rhJM5nkX3At7weVhaSkWnzODkUJeo - OG3BZk6yh4wKJfh2bg/u2j0tGTohP5A75ZlisR59iD7ig8We0OKBFwSrgey65sS71H7CT5csg+h4 - 1MlDeNBhOyh+BUYTnkjqD9eCd46aito+5UihEDVZv/VZgjsezxxB373ztuYgLnHv2NHHzGXRcvnB - v/FOKhO7vNfcN1hDM8P32s7c5fy6ZOg9/L7EqfaK+Lk4L1D9qtZMoy3RCPuQNqhPdogL87K5NPe3 - GAL3gQOIYZMMf3yqjl8qybmbMghh/u1QvnIcuez4M/WKYCMB344ktKMH3bCVLuiy3nHwJXTVSPOI - eVicSB289/FctsUekdbqHjkvr5u29SJTwjufGDP7fPft7O33eMGe9lhNz49i/cnsDCwauERGb1Hb - nkX22/fMX4lfpprG679jCm9mMQWV9Qvo+AuvEPrXQz2jwB+TzT2HFRKP+IpNzcjdjdbhAsXPawoY - /FNcLhwbFhoH8YKNlp2GtdIWHdUnUQ4ASX97hYov4emydDi07qd25w86VB1LI37qWi3bSbUDB+t5 - JX6+bMX2Pp0cVN/Rl9jPYCg2rzkv4KmcnthiuqTdjryhowC/OlJ8T07BTxojAneY9Ln5UNnlikMe - wuH8MrFqMq1LbOckwyFEM1Yt104ICjYHtKsqYHt4ee6ia+cQ5vJzw8pBHelQcbYJ/U4PsX19SsUm - G1IOY0mLsPLaJLACpfzBdvEa4gXvq8v/1ihH0dT4RP0qBzo2x3KGWYrP8/hW55aqz5qHZ6FwiHk7 - e8XMp88H2MeD3Gtb1Ma29Hiw83Ginco44S8rlQDDfejM8qrsrvQqSNDVhCxgfnXrboBTe8BtZk28 - Pb5W5xE5aOev88/UtGTxFzeGsujGxN71APeVchnqn6kl58j8FP0jvW/Apa+VlMFb0NZmSXWUyGEa - MKu534ri1yV82XqAjc0+DVNkLxcwExcETHMQkpVOvxE2+pIHUzfa7Riz+5mnZVoCpB3DhFzeNfxb - 77PoTb9is+iPh85HVHGic+rAJkp3AXv+xiZ7SwbavtADoj7MieU/D8M0PW8qrFzcYjn0d0e5v/0g - RY76T49MlDUuIL9vLlGXdhj+1gt80+aCrR2/qXSKGHhLZuuPz7Qb/YD8Hz7hIYAJySPtBd45n2Jn - 0mTA2dnCwLnDycxCdkkmt25KwLD0QeTzagzs3acO5PrOISezCDQB710rh4O0BuL1mSeUfZ8z6CrZ - jnfDjU6Td6jg4M83IpvmI1mlx6uBDbEMYkSmVQi500OYugc14MZrQtfX1kAYFl2JtSpJEyr60yjt - 3xdIkSrSLXioMhJvT4bYR9HW1vjMvSRHuir/9M9SBOgFxkHZMK5KrxBK9Omg0FtgBrQmYDZDaYF/ - fG/Xk+6251Op36pjsGalpI2JMqdHeK5S4q7xGyz13QqlWRDQbI73e7IVEMbwas0ROVveGdBDpMgQ - GNgnrpxd9/mJflCRhXTXO422KXhOQcPDDYe4M1w+NvMLnKjXYX/nG6u+VD2yqOeSIgy19jfSPoTR - VPn4tr4+7viH71FvVcT9ZGyyLq9jgAaRy7HVNmOxGM/fDOWu6kmiCp9k0aHUAcUMlWDd+RytSo2B - nhduAS/zi7tehMGD+nJ7EwXSBiyrn/3g4Ek6sSVjSkZVf47At04ecWepcumlpyFc1pydpQMzJmv+ - pM1xG1mA3Vt0B3N8an8wzBeLxEoet4u7GjPsrM+BOJX0LlaxvPfgaQf8Hz8etmRMRmAO2nVePfNY - bFh3S9iunBfkXtBSusiXB9yolMyMEM3Jqj+wA4KvjnZ9BYbxT+8dK/5O/vjyHNRFCmxy4DBGuq7R - Oax1pF/MIzZww7ffZF5HiPo4x7IyUjCfSq6DO98jJ7ubByL60wxPP00i7oSeCSkp/SG5NgCR9/ka - pOW+wEsI6h3/oz++m8HIcxxsyl+H0kf1nFHifsEsTo2VcEYSLvDWoQIrx0hwvwaaRMA3PcZnZ1Za - AZ3LDQLIxsTZ8Y/FF32EP7+IsfUhUUs5pZHQT+vPszi6pbtly7GErMHrf/xnWN+BWEqN0RHi6PHU - dslXzZHGte78i7tTQdUyrKRsmgYcvBY/YXM7ZwB2nvK8Sn1TLGh59JCH9ptkn6tJdz0gQsFjJGyr - NKPjVz7lMH6Ob2KEx8alm7imyPgE59l8VmO7Tp7QwKtddSTOTdYlvz5c4HqLK6KeTtrA38tClzTt - Qne8ZZLx5s0zLB21n49mnxTUPfIp+r6mCLvfC2434fQ1YYPPI/El5QFoFsDxD3+Je5dnQMygXMBX - P7VYOx/EYqULlIEyuId5nWzQTnl3ceDBvz+x/LDM4quyMAfLt7Vw4JhLu/n5MiP2e5yCaP/98oWL - isSHOBDPu/gaqThZh7f3ZcXus2S0/fsquPN9gmXlSOnVsBf01d1d1V98jd7m4gHay+iS5ODwbl9+ - jiOwTrDA5SGWW06IvyOMn/MbG86sDIPpug/ojbJLsu0kuOuL3HqoM5tEfMo/ik0cIxtazHvDvqmH - lHv+gAin4XIkMtEvdNvb3IIdrwkeD1sxf2dZhxio84y041Jsa8+xf3olWFYsFH98SGpPtzUQgu97 - mF4k6tGaHx0sr/hacIk4/f6bL6qELb7K+fyT4uYgBaPuR8MezyP407fuGhuUT+1vAB3rkmL7/svp - Yr9VEdZG0mM5/ex3gRpaCl3arRgfpDuYl9PThO/xwcxaO76K/qSE2R/fDJbfz0vm56OG//i0p3V1 - u5DolYH6rSjkvuMn7czWRuHMLX/8J1miLOlQwzNbQGGttJwuuLGk+7kwS9l6ArQ3qhnG9hBgjWfP - xWKOHwalp0YLYBZrBXuWFE+SDeeNo8kuWvqL3wxsuIuNfZfr2j/9gqLDwQu25dIV/+KRFeUSPzZB - A7wD0xTdnsZM7HgwksmCQwm0UmXxX34c/W/VIyZ/bBgb2Vvb/YwedAesz8e74VA+dGYTosjXyDO2 - w4I/Z8Ii7fiFHaEfhoXPIhU9zZcXMGG0aouqP2eYuAMgJjyf9vGvg3/62oqUIZmT5PpC1SanOOOG - 5k9vdX/5ElvKiQF/+g52vjwSZV/vPK2zDTjAvWD7ducGAkmx36JRzzs/9dvpmC4v5PmGhXGum+0Y - wkCXgso5B8yxDChfVwsLuW/+nFF+atqx3jYdXq5OG6xvWNElJ48F4m+mE5tKI6DRfZPQFbTJDA+x - PHSbssno5vUE3759CFjg3GNobkedeI47DtMRsBkq2+aA1QaTdr0rdQBn8ylhe7uUCac/sA0tRO/Y - j/O1HWf3tMDHOTgR75L22o8GhxDWaFJJWLYcnWvXVGHelT4OVjPQOEmofuh3+Pywt2432ge40RF/ - gn7A5gFKloUtH+AzM3eiooSn871MTFjdLjE2z7bl7hXYGFGvSwMA9Yu23vu3DKPgcyI4NYdh/dOr - qnPMiPWzhGTXzzLa9ToOwJCB7eWXEuTisz0r/l1v5/x+roDegWDe8WDgCbwtgHCzMat83Q273osl - oOfiXzy5W1AXF/jnZ+ZtewPbIT/9/vnBt1z6Fds5Oyxw12Pzxs8DndW+eUCjpj0x1HenTbufIuU3 - lxLjpSTatimSDDLuk2Gv1ZNkkeT6ApX4yQTyWDQJmda9J5L1PmA9KSd3vHZTCHvcuNi4l727JV8n - g33k8OQP/4Q5/Jrwj3+rL9kYSPYReTgpahCAj4vdv/yGmvujIE7cfZNdL/RwVK9r8OLxu9hgWpfo - 1p3MAAzbg7K/E0ql/f2I7g/X5CucqQz//IDbl+PdFZ+kDX5LDROf+7AFOZtuIElOSvFz97NeFPlQ - GkQhn2fHXIblMUUinF/dF58aXh6+aZBVsDv4OnbeNbfzRa4DEZM2JHsqpjY+fnYHq4FGeOeT7vDc - Tygt/bwFYMejLnW3Bh4VSyfWu/q0axrV7D9/fx3nVSP2BCq45xuinp+vYq0sJYPhLCyB9IffrK80 - 0KTT7U9faBsEUwiWUF6xKZfvYZzecgp3/3c+NL6q0R1fkc0H4sxYIHTn+DT84FVNXsE2nmqN3/0Z - +CHbFrzYgQXL7i9DI3SYf3x/2WgsA8GnmMgvo253/8qDu3++10Nwy//FZzj5JsFpstDtL1/s+Q77 - Xsm3C8f1IpgL6RMcuo/XstnpnMFxOz6J0ltKwR7sLIeK+LgRW3b0hGrqDYJk0pigfyqmuw5V8kAb - GsJgZe+quyrZDKHAvmQcaZeJbvfnK4Ns5SxBn7My2D4XMwPqr1/xWc0qMEqftwMt5rMF09OBYGnO - v+CPLxHXPI106R1hQ4z6MYlqZX679cE3hd57PQUPq34P9MB8Svh5OHawze+lIOmXG6EwX9DMm+yg - 7f5TAFkFj8S50Hn457/cW9YkfnaA7SBRYkLlql+wv99dSYQf1x2zs/IjZq9tdOydwwY7L7hiX7nX - GrVXnYfim9GwX6attjW/1wJP0fVItDuYh4m7oBIqV3HAaYItKpwSyMB7wdyIC35DstS8lktEuJCA - 2/XmmN7BApK6v+K4wbj9DcylR77lekTp066ddKl8wcFOxoDMrUqn/Aka4LI3k/z5zUsTZ6L0Zi/D - fkLs5a6tV3ewISDHJ45Z20nc5Af6atcggKq6ahSMqoru3GMhzipwYElrQ4euGsyBuM8PJ5b3H9z1 - Mn4Kr7u2Jcm1g7Jhv7EP3iSheL8F4/a0ZnwKmxfYOtxuiKkbZwZ05jT6lx8GJrawtfJ9sWbHzQQC - 7/3I9X71NLpBp4f7+Wy88yWNNtclgFbc37GMsphup63KkLO8dKwvuuHST6eI/+oPxu8mDm9lbXXJ - Y+4O0e7zt1hEoOh/70tM3vkWdL2cYvDj+CmgbNa5v/n2HUFsepjoMhjd4U/vnC5bt/vpx/bD18oP - sIo/Yr95asnCvKvdTzjFxG1WvRX+9C7tzIH408C4sx7tt0BZ8n8AAAD//6Rdt9qzvBK8IIqXYCOp - xIBJAoQNxrgjmeREEqCr/x/7a093aiiUZmd3Z6XtiDVRt50UXI/Q/4DzLL1gzBbndVrg53w+z7Tc - ZcMG5kpBSRPr3/i7AnP10DvkBGrsn54xzr7x2Ay+eCfkmZohPb7DCGZOK859kdwGKs2GDG/eXSdu - PG/D3AMvhp4RS/OePjEQ1yb0oSfuHtRyZdouceW8YdJEOr3whs7YPdneoPC/hQrRpxrKX77tXF8D - ap85XvvycwovtC6oktg4+1jGNf/pVT6/yRygp66oIE6NlJiPUA1FGdIZJMy5zcD5KuLJXtZhyE4H - 4hbGLlxjGTbQ6Y5/hKQnDH56LVTbV+rLjzkN6Zj4Frw3dUNc+tqH20/fuBz97atHPrVV3XkjbK1H - To87eQ/W+3x04fe8Es9jByB8upcLbV/e+3+N12gs+qsqJM4HkaQJRzV2cR88krvhTq3aNQbm3i4W - nLFxpgpIry1LT7iDfQw+//wjdjE0H875DIktNar21Uv4f/lA9jtfvS3EsKH7dL5Xosik5HKPYVki - 46dnMql+/O1gL35Uaj7CJly/+Jefka6T6NhkbJo/LgcB6l9Uh/ySjVWgKqgO0yv55g/wLC6dDIMu - KGdZeuNhJfa2weVvs+e1fKXZIqqzC396sRfhVzsu0yrC52OeiWNoWrYtKszhaG4lNbSjrK1WsuPA - L7/J7zxlkJT9OwHrnvfn+atfS4+XG8PbmFh+LDp2uPIyhYCc44ng+lUOn/HJGfCnN91xe8xWo33F - 4Ocf4NsJgZpNzQh/8dVx6fZ4su7hE+lRSehxN/HtJIq5D3/5IiErXSA6f9MbiFknUffrrzPp1QU/ - fpubbz5r+vHF/1FRIPD/u6QgLySdHkzNA5v2+D4Lki4Xkv+5h4yXMHcG/iPYqLoRzARJ5jbYWVtC - Sd3+tZSU4Pu4fefQU2zF4XYt3o3ob3pBSzHwNOlIoxxeGmjS686CLUvPloUqc0fo4TFMIb1exg54 - vFjOM6srthrSMstRtwz+eZc6eOGNuw/Eh1QQOyIbXoxD1MEgUSNqxusCGOILC9xlbBLvMIrZClsv - hboS9STd839sjhK7BPeq3ahdmXfQ3813ivL+TclFvnSDcE/JDO5QPRB95Cq2AOskojK5qv72HCNG - zcMoQ/UojrPYWiqWumsmw1o7RcT0bglmzyk1oFFFmCjR7jrw2bKvAHeaferI2hULNWMzpL2mUlsH - 53CUkx0HNYFJlLwQzxbW0h42TxiS03tn4c9fx/XwN17tZKityI5xh5wlkGhQ8ly25uQMQWaJ/Xf9 - 7y0vJwuHTNO2qX1AM1ge4jGARa5WVPsTnt/ylsJHq2/+zdAL1Ez0PS1GXLCtRDnDpd0yrIv7w3UI - qJapdrY88/4NU2A2xDm+NCwyW08RVwJxZni2B+nI1AC90vr9W7+WZYg5ULrpFr1+zjhsfuNlgH/T - u0RbbRUWVUFTzRB19vwdiF3Wq//m46fXPVgxqVwUpvRATfd5ZmsBjhC2i175O+cjss33tEhmdp2S - MtjXA7+fHxt0TQtTAzzakIlo4EB+gJLfd6QZ2KuqNlQ+nInYk3MAj0Nln+FVX1xqJEjH0vXolXDp - 7zG12gsJxYnzXbBd5JVGC7Xb9absEshP7o1k7veuyWkBOSy6+Ub1njXDeqjsAG6V0JCDcnRa9ueE - KSqyo0byvGs0Hn8GC/Yjf6XulogD48p3BLfLbiVG+BjwyqmmA4Jq9EiGJpzx0daN0M+fe6puZYtF - QOMUujsgUxvvL4PEwDajhBBGHc2bhi1oDxUcO0uhZdXstDXc3EQ+JvWNqkF6YmPzakZYcotKfdWm - bOGNqw+pKE700Ew2E2Rz38CQj6+zUOancC0LoUKKX2vkXh58bX1xq4jMEPL0pi6RJtxPuouqMb/Q - VD77IbV2dQTDG+tmYD9njQnOJUaCkNlU4WytFSwjjGVSPyK/l3Y7wObrjZPFh1CQwNRVJol8laP+ - eJfn1+EcYl7C4hmCG3eip/exYWLdwABGXKqQu+ZI2TrPJPkW3uakUPYnbXU+qoh0krnkkDSYsZlw - MaRjKPvrIOKMX2ougff5vCc+zIZsdAOlQfByjWcJdH4meZbgQnUceqprat1uB67p0WuBH2I0+579 - 9gde34lArn+KEvJJy/GwyW/VP3ytzri8IdqrF6qz8jIwRysdYIAzmJfzdMska1hUZJ6P8rwLHgVg - 8S1N4bk/T4SogjLMuXqIUSzBmmbN2RjYmu8VKLbPh78fixzP2sNSoeP2no+e8/AN8QQRxdJBJ6f3 - JwDsHdg93H+Kly8a+eeHxzMEnvMkx+GxMcqonaLmqYjEjZ2xnZr7cwEan+ckLLzXwOJvo92jXx2J - ecNG1ktrCuHXvvoskHC2cKHfQSGnAXU55ODtceV95OBxIQ4s3EHs2SmHp4nD1J6cGqxim8zgfJ9k - ovdMbQVhcZS9xpc5PfB/k8bs6tnBn/2RNldol+TVvaHHwGEWeMcYpEw+xghXFBPbb9RsvST1hu51 - QWhZeXUovDerQV3UP/2tnddwmY2+AYuSa/RsVq9w+Z0fb0Pfort7ipku7ox9hw8BuZNTnq2VpcxI - VnFFTC6UhzHvLBfZj3Ch2k7S2vl07lP4tw632QBmxKTtr6qg9UY+iU3iA4m/Gj1s27Pgi/zXZa08 - cYbOIzZIjoDXstRPI3h33gFJpcgBo4I2FTV/PZ63UNQGEd4+T9Bk+pEUSn8Da5PeDPC+9TU9rq6G - +c45BzAhHqN2NmiMvW5HB81AepIjepdMuEQAQvttmNQ+3hS2ueweAFlXUnqviwpPQCl2QCRKSfEj - cHA36OsZpfa0+jyGGEhrvipIyKeAYlTy2vi8mAqKUrIjX7wOwqJ7PUQSzoj55ef5piQ6HNfLhxxr - XAwC884VrPZnhZq3CuE11iMeNodyIebuibFkh0IAGy9CJH6ESsir4RKgfpDlf9+FfR05qKsqnhpN - z9p1PIMGvLvHmdy7TGvF5tjykAuWlbrmUdfo8ZlAOaAbpeZVsTQWF4qPktd+JEpdYLwU2oFDIDle - STKWJ034Wywdhe/SJt/1A2uvtBZ4t4Pgcy/ctRQNtzc8G2+d3IokHKT4iWd4d/qA3sS3rK3eywiQ - EY8CtX09BLy49DG0Dx+T3hdd1tZgySxQH5ScBgBXeFZlXvzeqnZoZjgQb9dYl1EvrRZNyTEL2bjA - pyy+xis1aWFnY+qrKnoG+oEWeevhRcqLFGanpiWWPi14rZTJgNB7yuQijRJYy8fNQmJXqbS8vqeW - jXyfwKl9S7MwHHQm6bs792+/vJMifvmEOPAd5y6xIyUFbBc2PpxiyaUH7vpg2ykOfNSL0kDNR90M - 03UhPPCVcaSX88UNpav70VFBfMHfpQtgw4fiGHTGvvSFO4lb9qdKC2Jn79t49O+RbS67BtCQr6XP - evPEtrV3evksOsCXpykF2wdGHNRKuKPX7/6PS7LJqHhq/Fx7/qhtXh0EqOof2e9/tqAyUOGgDTH9 - 2aNVBYUDpoamBLvsDbYb7F1oq11HbicBDts+iEagyuk2A6038VrBPx9sf43hswpjvA3xQYUgMa9E - MYShZXtN06GmGcPcG6kVitXwt8F59znRH7///FXU0Vc8f8YC4g3QMoGFf0qIya02ZuBmB6hJrTeN - +g5nIsejHLRw5/z8VbA0y+wD4VOBWURLPSz3wbRgrsqR39fqC1DMWQnkhFgiOoM6Xh+5E/3Dj32q - joC5hWqBizw2JNxnp2H8EzhDDvnoSv3D4wCEU/sRwZd/ZhDjwyDKzHXBdN0+87549Wzzvo0mN4dT - 6FHjV0ap9m1jouJqXpJWAVQ47SDw1DqYd5XFs7eExQCWbs7TcHe7aFLqpzE0z6ZM/aMZZpt42Cwo - s/pE8ONeZ1vavXLA740XNc3RCvm8U3xUsOXb+PN51yb60GT4xRvFIBQY7ZtU/WePk79tGaZ0Obnw - PwAAAP//LJ3J2rI6EIQviAWjJCwZFBkTBVTcET9UQJTBBMjVnwf/s3TjInZ3Vb1gWubBcZ13tTOv - 8wpCpnsUlEo2SOP7M8EgrwjStvBNplO1a6FTWmStV3uYwtssGbXo21R4mu0wZrp3hqEnMDRL0RuM - iAURPPr5jpGXtfAhVMblp+/YQX8nIEVeo0OndAhGsK8GdmIYwa157llk6m3Ic+ivl6jlCAmWrtXT - 04qLnx9nHsoeIXuU1AOrv2N25wZkliA/w94lF2w2/Ysvmyx4wEb3fGZfM7i+JD65xlS+PYaz7pRy - b5hMyJlt4N98lkrl0BpipxkYcWchi5iK0ZpKMPOCQhz4gexaIKgvxvBGt1JVqAUTJkMUMLTW23La - 9xnMYhPjnXd91RwP4wixmBHm8HtfT16zdaGXBAcW1dWOKJjpit6dRAf/8tq3DWsE1nmOg6ccO7KV - v3p4gE2BS+XeEEqQ/oGhUcksMtdF5kS5V/DovinDw7Xly7OSEqC0k463yjCRBfNHDw36VzErRmM4 - naq4gcJxRMxUj+Kw1PNZgbFdHyhc8+QkpG770zscrZ+XggzCP/+WbbTB4S9XWuBIbnu2r/dfh+p8 - i2B8z3Rmx/El5Ip1lYBkd2/mSt/7sHg7but52xzxX0vDUNJffQKGMvKw93Y157XqlVGckMWs+4OF - H6PoI5i/wYjj66urF/HufcB8mxp2VdQ5HVc/YfgtjpEmKx2YjfPNhRVKCiq92z1nzYVounv5JOg1 - R44jHyQTQoUoPlI+9jhMSj3b8Kgad2wthwfg9t/WhS/lNrBYZHo6KlOfgU6vfaQtmkt+8xUK40dg - Zr5b0ln8NCX0SXLDZimVhJd709WlMYEMq4+Qz+b0KcDqhxha8xA9hQfP+OMLRDdfOpAJMTuC8bYI - 0Ou01YbZHJfG0OfDlc757IRLqw8UytJ9xJb0FOqlho9kXQRr423516bDTx+Xs4RYebmeCZu+uQfX - PICvmakDTtMmAOcvodhrqz7klmBnRrOJRoZgbw/L3dEbvd9nEosVO+B0qJYzSK3dyCIgtvUSqbEL - EScas0VfcdjGcVzwbZqM3U7gRKZhwyYYXFoBx0A4DnMSMvfHC3BUMdNRUvmqwyJsBxzQ0E9/fuDf - fI2vkslVYnBPB+Q4YlTIHpFbfDWh2pibFTEp9XRAoQCff4qL1HW+SMC7SpAGfzJ2T5NPZmWt1Mt2 - ifDP3655OjGCc/fEDqcWmKTjo4eptR+xI17ydJbkC4XeR0TY+hTX+t88iSWpZNmT+ID+9ETsdIPW - RVMAjqsuAlHiQmxV3w7MSjI3RvMsj8yOBJFPpxcvjEVoOlZmEeEciXMPj175YUH5WcjyPAcCWPKw - w04AmnRaxAMyODMNZgsvz+H7XTLCjxC9kWwvMvl+WtGFtRjaaAlwWv/L4/NE0FrPb8C2yuQa+33o - IxE/0kESZWELOYMFM33T5zL7zNTA575Gm40sAY65mYFerg4sVOcm/T57DRljPIoYfc8Vnx06SEC/ - jQldVn++2LokgdMJiWjdWJmqXIs0kE39hqpCqtf0ZvuZsVEjhnG+e5PhfVsEEMJXzpxtogIeH3IJ - PhLVxMHqr+QHVCNIxr8Ub13z5UwqHrdAkqDMCtJTMP2lz61RV6cc72++z5cX0uhPj/7Pd9/w/k9/ - GLbbLh3WfA6saLwzUxV2hJ7C3ANL7nd04fYxnH37W4Gwana/fkznP/u5rsFZBvoazqCepGBnwtHc - nPB9qrxBtjJVgT5fr+X3rq+BzdcQ/fIVts/SPlxAW2hwKJGHf/U/zpHVQq5vQrzdaKEz33ABoece - bEzmtiXTrfEQEEsY4eP1NP/jTT/9R8Zn05GJTJsKfsLNlbn38RB+5fEWwUC8XJn7Wv8iML4/C1x5 - GksKy+KLtacRtNK6pIc882u5MczW6JKFMqc/+47y94UZjNTmzu5OcEk59HoPfgT0xrtB6sM5En3b - WPkWUutzQzgueQ680a9xmfUB+T48bwRVaysYH4eZs0drTcbk9DlzKwLJ0vNrCW88++Bz5e4Gaa8d - TSMI6cSsa/NJuaVnD2jseI8etB1r6tbhDcpbK8cXYfad5ZdXn/z7YbtnkXHOdKv88QtWbutdyicl - /cDlOYk4I5a0np+J4EstPMoybay5VOUu5Nk2ZqRZQL30rO6NIksa+qAXOaWJ/m6M+dritT99QMtI - cOFYPPp/fn16Sjf7931Idg41kH+8iqYhxdF+1zjcOSSL0XhTjrfHrOGzkuMDvN3AGwmhIpJRcz8N - 0KhbIXiJOofp3aMA3ZeVzLnIdPhOe1DCQxS4SN2jDZ9n3t7gGzcqMm5LQxbz/hCM/CYABO5ZHY73 - 13wzGtptMDGgPfzjWy86YtSCV03GfChu8LZcQwply0nl7PlNfvqB0eUT1yq4Wgf4nqwIoxZ+0+nX - /yf/PLPAFs7pHObDCL/Vt2A3aYmc3zyA5LA+Ylz9zuhjMYDSF12xdwOFM8XSXYM4dmKqO7eG//ie - Lg2XO4sO6MXn59vOIaFZjMa7enOm6akUxjktCI5vyzblQrE8gDh3V+asvGJ+voP893vjvSTbqUTd - /gG7lIbYX/PAKAThemkUSpAGLc/p1vNW977t4DiQVTA3htfA4RPfVv1xa84H0kLuvQ0k9L4Xtont - JMaPB4br+clKsmnh51kYVHSCC1GZ7t/gyndYgLNkmM9e0IObrWUUnsApHftYWiCsQottH9cv+Zyv - wRa6z6HFoataXD1avQCjjXzBnvxO6lnfzw9oVWhd89TA9Hsv9iPc1W6EfYh7slzZpMGU9CUVd3zr - qFraR5DsQcxCh2DeFdMVgW+oE8q72gHqdegnGF72Mn2mwcK/5L5sYXkhL/o0Mwt8dqfu8c/vOqUZ - cfVlbCJYRA+Pnentycc/79JDQc5UFndgx1f+28PDxVXZzq0Fhw3BJQGarsX4WEErnFW7KHV5Ti8U - mrrrTNdLXMHKrNOVh2shP1+THFqB/ELGGd4AVfZY0Y+qeGe+HbX/eJa++ju8SzYFWdJ53MKV32JH - 3jzWefuqwPuWVyx0+rezDORMgV4aAWqFuuXsl4dX/8biSzEQHupFCxTjpdHJx89w7o5BAtQ8GLB9 - rdN02Vy0DByzTEHAwDfAEQwVmMbgjF0JKnzxGzhCZSoOCKbTxeGlOBbgIwZbpGnSN53H8/MMYnV/ - Zm7TFoO68mZo5e2Xxd1I6zGWLjp83rSOpQd1SKfkclBgWLU75smKz9XnESpAFYwOKU+VpvTzR5KN - 71QWAmmVgc/T68yNThaZSrv5y6eN4ucQhmebmac9ApK7swtDIZLP7s4V1VRrnh8ol23wOy9n3sWv - FuJLY7Kta+5CJY+t3DiwiVEBUSeVjMHPIHgpd+Z19t3hz+Mqln/9Fzty/wELCz0BHvd1iLJdeiez - L8NAL8P68u/5zLJT8+LH53783KFjQqIfz0OTDj98miO/gUydApwwWSHfsU76X70ivdo2Kf8+TqUu - 8kfK3Mi//vLCx7iDbcPiQzzxRn9VCfzCM0PvYOOs+WHXAPdMZYbC56emL6SNYO0XvFt59aifQg/Q - 4C7TzWN/59Nf2m11omw7CpKHRpaBlCNc8xe2hMsOTMFCe7jlYsrQmu9mr/gcYKRtdKSK5iOdZMuv - gCqIHQ4IuQx8zc8/3vR/v0XbcoJfQBLst2I98PpxyUFr3kXavIEwzKejG8BTllZIifLDsBQNK6FV - RQ6CAk34TGcDwQNsC6qa9OV8p7BLYJ/GNto8pNfAFesowb70F7bqJaCfQyfAMPK3OK9PLJ27o50Y - pw3frXkoDqnOIwR46k34svZbv+YnkBejw0giuFzRRfmhn3V8xdvehWDwtOf5p7fscNKlla8d4PoK - SoTjvz1wZo3GNzA3tUkNpo717MuSZyjGW8Pemt/H/nTIjSIhPXOj9xv8/CZ8bqOZhbOnAm4/Nvq/ - +eH9Bce0Ti9PD3qJd2A/vzYLG6cB/wEAAP//pF3J2qqwsn0gB9JJkiF9LwFBxBmgoiDSB8jT349/ - n+E9ozPc3+7+kMqqtVYlVe21JNjOul8+KoHyhMatvxBZ0vtqaw73D9A/DvFP8sxXNMvVFnykd0Tw - dJ3oH/8SawM8iVklWs7qYZT+8QksGTfkLvExUf70Lj4LhRQxq2c3Iv5Ohl/lSpdT/NmvDJp9vevN - d77rnQCxey9JB+sg2q6SPwNzmDCWRRDR7c/fWfKMzk9ZTAd6Ol59pJBCxgbzk1w6OK8LbAZlxRJf - JoCpUSlCqdId4hqtU3HuoZ1PuF47kjP0rU48djfwp6+kVL9Ea8BYEHqRdCfuK1bdNX1a3J9/5y97 - fYRpt5aBUGROe/0oqYbLqi5wr/dh90JwxQJgObDpzRi7rT8Na6QMNWTC2MO2jHxKUiiLiEe9j9Uy - 25v4pgwHpZsHsXWs7hH1Pi9ORLydY+3dyyqTdw8BmpbR/NVT6TIuYwD/+KYtaqO7vAZsARQ/8b96 - LH8LD/9+HuLO7QcsyTUOoJ6lEb4Uy6wuWmO1cMdPIhPVUJdMTkcI8mj0N1cRwKwFVEMim7nY5Vft - H7/9x5/MJv7RTeRECfanVtn97hJQvUkPcJWEG5YjJKhs3Ekikj72RPysCVVu96egYA6fmWMgB/qG - QwIctgXj3LtmOTHlWhDLCyv5y1oH7noNfQt+T500C+VtdLfTbUnQn570ueDsbpZdcqhyakj+/IPN - ia+1uPtDf/kc0PITQZidNhOb9/LhbnciCPDuOjfihLOlst+0KKH77lts2NnkruNrZMTxPB/9iu0t - sCFbef75kTseH1VCwTbC6vqe/Dlj24hWQ8vBQaspufNGW5Hme3iKez7Bvp44w0IHN4AHAx/8pLI+ - 7jgfDQ692fpDkvR1HWamNQzI37Dvcy/wyxcrt2pY+fNM1ArSajx1gQFlZFHiTGntti0Y4//pSgH7 - X64UhN4NS/BWAjLVyoIOvclio3S7aK76kwUNqyyxpex9bD7nUwYlcX/Hi9Pa3X7DMqP3ywtIYjxs - df3keQ9ZpymI//QfKuUbMMLYlh9YUS1EyVmRY/Rwle/MRqw8rP6n96BTGovPN/IXTORyPohec1iw - XK0Xyn5kCUK2zgZ8vr0Wl1bXYwC19OwSWcssl3ksFQNRqTvYv6exu0pjfoDBSsxZbLQX2IoCpNBT - QU/OafAG8+GbH2DHzAdfdJ9GzieaVUC3jDt/feoSZQ7CpUHw9FB8ZjPfw2SYio8k5zL5NGTFYXEM - /wO1Tc0w9m/LQB/lIQbsr96wZyl3l+kbKYCdlmhYn6p98qEfbFDpD5C426Wtlu5ORdgdnMu/X68u - CC8I+7j3md5e6aSXgwEPdPqR+GlldNWNNEWBKQgkexIFMMeTl4JsmkWi0HCNVgDNDW0JaxLZYbSc - poK9QMOdf9h/2vrASHnboDa+tASP+6SnNCFPeHKJ7LPV4RatPzZQED8EA5GQowxsE/gMOFRvB5+D - y8Vl5OKSIYTnfbmjDNj1lDAwGZsrVvpUdtf+wXFwHNqMPDP3CpbX+mhhpoojOX+2AoxKx7SoXYMD - yUjzAaxj2Rs6CScG25rGDasoShq6m+qNWHYiVczxpKUQXFsLv/r3I1oox3+giF0ZW2uTAza7F/tk - 5iUhZl5/1NlxMxGSyNzm1t2mPR6FGb1xNROJY9FAO2gpkKsLk+BFlartrmQQ4mzUSa72ssrmR1qj - jhkPJEqa0aVaFouQinGFU0E9DuuRfDXUOLc38ZYzBGNR34oTNyjbLDabo3KDcfdR9KWifyi+W7Ta - 776F3K33iV9tQT6eumWB6DQCrNxMbVg9CzSQr0cJF3nguxwBnCM+4vZGUjldwFIf+Sc8yRCTGFA6 - rHysGyg784DogqTvk7xwAVSmd8j+/dWllyIFihffmo9zoVNmUbz0VGjpk7wWtdz7MF8zpM9jSLQr - dqqFaVcPtYGoYb/Hsbvk9gWipvdLfAevY7Vi1MzoFnkPctNhUDGv9dqi7PF5EAsrl2EbZt+Cm6P6 - /poPWrT9rb9/vj2SJG8z2jsGzJC3ZIvYtT2AdbvkM5TeKodt5ZGDziXcCH/85YIlcWlcnmtsAbxP - V0Q0R2Zz8ilLCZrK28Oh0HzoVN4FCSqyM+ObaLnRv/1d0fmKjbuUR4vaBQ3a8Wgm9tsc2AdlIJLK - McVh+j4P7Ck+iBBx8Y2EvYtz7q5cDrC9Pcn8u5/PLmG/gQbGoc/8+Z1UdKNXaYFFz1k4i97naikc - b4G0Ek8+70yKy75oIMDnFc/EtI83QP5+/kOvs77R8SBfB7x8kKhnmOjk6OcsP6YaPEeMPqOzTumS - ztsTcYO0EfW6qGo90yVAZ0b6kHOHekBfZXaBjsd7s1imZ0ocVjXQXWWDmf/6mrv6i/eE2D/32DAe - tsse4rIGE9QW/7bGW0VvmnhBnMNccab5b3cTELpA31l7fGEU1WWfzLxBc6tbbLb5JyfV9XgBC966 - uQlQ7C4C9/YQeP1uxH26qrryY2qgTjy4xE8/rsuYQlIA69Nr2Goi4tKQbwJxEkYP38/Dqm54W0vU - ZIpKLHc7q7zNEgsij2/mYyts1bLlVIKv9jNh3ArbbhlcLgipGoNfSnIaxvch4tDEfnTyqMqfyuiP - +wG0k3An+vvT0+Wdzg08WaXoi7OQVkskVgn6UNkld4+13JUEi4EORsngeH9ysb1D5QMv/iL4p9p2 - Ac1CoUDac+WIt4CfOuVH2qArF40zgiVSN7aKObjjP86WwnP5sNcU9LU3Z+Zl41ft673AVDw/cdGa - ftRf5TyF1zEucH5bm2orlHJGRjujmU68NaybPlxA0JpXfxkjpVryauoh0FFDzpt0cRlP75/AODo/ - 4h4NOeffU/oEg5Dy+FyVpsrHWl3Cy4GJSbp/T45erQXYY0vI4+SW7uIItYf0j8aRRwiKneK5HLRx - e8e6H/o5I/mLhVyjzPz6L5/cpduIWqflcGRBJ2eaY54BWs4bds+x6vJ/eHn5SAVxc7XImT0+oQiv - zDy3jZ1zh+3gAeMa33FsMDeXST+KAmn/qHFQYC1a740H4WmGNTmfXMllInFIoNwT1a/oU1UZW49n - eK9mjJ++bubrV6lriPkxx5qyErqF96BFa1OcSR685YjTnTyBD2DIf/wBjM3XmiHZ4IozLXlR/nS+ - MKi2yxCbblmB/rRqDPT7tiNXyPnVMuSvEi546fB1Tqphz18civTvg2Tts8jnHN8ttMfHfNjzF4fv - cIZe1HgkN7cjmGpZ3EBbv3Si1oHvMiZv1TB0E4Gonm0N7O3ub/DYIIc82njvs99oB3i3Rcmfjf0+ - ITyXDawq40f06sBHs4juPmyk9IKL29mg43UeZ8CDh4XvQ9LRyQX3AN1OvUkMbgbR+vBWBs133Zi5 - N8O4P6WJPn/xiLUL67vswzdnGG8fH6vCLRroY0wFcC06kYQOU0drq4gJcorZ9ZvqcQCbkyILhiCL - SfTgwopxnvcSTVa/zMfIUlRWMpr5Hx+DQ+xFyyMJLbSeuBV7eqW721stA9gCecE4ZeJoid/KE6la - M2OZGbJ8CU9rCS/gOBCMttN/1n/O0oU42Rqp862tA5TFuo2N6nGglGtsEfTMgSderAbq1o2fEuIb - W2B5IJ9h2yqsgW0ZEmy8mdilU+0s4Df1JdkHtVTzvh4R9+LHFwxYu7Q42QlMy980l1cTVRsleQHP - N8nDOtYCddnkMkH9cNVJUvvLMIan0wdt0SAR5yOnA3fNGw1l4ofOzWMdKa1AXSKOaWtyv3Yl2Jzv - t4Y9qD5EAfitUre9tOj8aHisfU5evuGzdwGvT1T6UXE26KqUj/gPT8g5vt6j9uVkJXSDV49tkFnu - Hx6AnPFkXBTMe5jDe9rD9X7c5tls1YGym5iA86Pm/dOOb1zGfp4wk8YvsS+FUnHb3PVQ1qK7L+q8 - C7aI9SGUs7LBxi81cypkagAvn5tI9u8Nli0HEqTo05PHVa0qGoTYhwdKfkQWGgUwHePX8BdLhNjZ - q6OTNecGSMvvhG2PtVS27dATQi/A2CXCqNJThgyw803isBfWJYsbjUB+nBzyWschat6h8wEJ2QqC - tYVz5+9J9/7idabKyxhYA/xq+Ie/dAm/7n5eP1ATNOePHw9bNZoSktFiE+ftc4AUJzlBWntvsT7Y - es4f5A7C7888zPS6VO7CFtKG+FzzcXRdVJeXv90Mv7S+43zHs6VEHwV9I5IRCd4kQIf1doFKtryw - YgRaxTqf4nmaXkZGNIPh1elx7y8w0MoYS3xl0nVQJQM+wBf6DV/o1VrL4gJlJmbnPf7UcZLKDfqh - ImCVG97R9nq+JLEJaoKjeQoHduxZBYLjysx/fJqi9qv94Rux0Qm666+IPCiH+gUrreypzOvmQGim - yx3bzLlXp3nvE31Ybio+m8+7OtdhGkCIW0r0z3Ws+mrEEoht9bFbko9hPtVFChav+WBr0Mrh95mV - C7SL4oKDUEsivo67FLyh1+AHp56ipb0rGmpubjbz8U1T6UcUZ/hsA4ZkYFyj7ZrPBgR3aZyXnhCX - OjKtT9dEx9i8Ol+wFAnyoCGAjGiy9XXX00RH8Kc/bJC17j8+9ROLO3Yfj5Supf94QvmWvHzAW2HF - vvPjKDK35DQz51AEvblsGeDku0z0U7fkM/8RNlj80GuWwq3c+flkiH/5PplYd89fkoLQSxx9NjJv - OTXPVoowP+ezcC2ISlrHSOFXdDVf+JZTtFwt/glT2lJSdMN7GPlHnSHDHX8+4+UrWKdh5sD4so/z - ergwYMGXtgXZ7zHhXX+4Y3cHwt958vlHwOQ0fOQN5A/d/ghy3lzC09GCbHj9ElMEZ3VWb4QTevnD - z7zwkgbq5R8P/OE9TuVLNQW1zsB6ce7YHpE28O8478HZ8SIi32McLSSfArBOkvCPL9DsdRkR9w4k - bJ7FMho+ZSvB09BI2K1bg9IvkAqEPLYh2DRdd1UMpQXuCDUcfq9GxGicyIHm7DvYXPeSyZ8ejdn5 - ShT78QFd4o4luBaDiE0dLsMSlswIF+N6xfakQJcmz9yCz2enYD0Hgft3vqBLpzM29/Utl3fzgZeB - +kRSajdaEvFkwHkbBRJoBxNQmCsM4tPzlxTgquZrfgQ1HF2aYz8nnTviS9kffrFCiH/kv9XiGMYH - yvwH4j9+uYDH5vzjX9ccLCpdvnMK3oWy91GdD/mqTMcaTiNesfI7aMO77n77XITzis/jzcz5qVY2 - +JB/4XzSNglwl0wpwJ6/sTTBKSf5yBzg14MhMdG1G+iv+l1gArcN694E1LUJfA7qV63EbnKOIsrT - 2oKkeBsk8Ps7nV/dqYb92kRY+lrxQLizrv3L//f7aLkstGIPrVzfE+N1LdUtzSmH2rf2Jtn32EZ/ - /Byef35HTLK83MUTJAdedXHD9krPw673NyjdtMZnd76zecfYB8oZNhjf+0dEi9L7wFgpNuKKV0Lp - dHgUsLzEEPu+/ouIJcsbApenjDUr7Yelb6QLGP1W+4//EN4eGTxwTeAf1td92F4P24LDvOrE3vnK - 4r1YB13vUTJv8ouq60sCvniLKoCt+4uL1kKaniCyngY2snoe5ijbr4zfsTif+JtEx+96shCdkjOW - n3oJpoO5WPCTSCxJx2UD9KjkC/SZS+5XKeZVOqKqBdMl+OBXzQ4Rrd4hB+37tyOB87sMVJHmAIpH - 70Hcimfd/qbwCoxG2cSX3HHp4mpCC8GlkMmTzT7q2OxdW1KjkIi2VH5Emc0qIc+ID38psJYzMn5D - 2GZshB32clW3DLM+jF3R8I+1Yle8fGQy0Ib3I9a772XXs7IIW81wCf7TMzmRfBS8Pz3WlcO7WjzB - sv7p/yDcpGrBpSKiqusuWF1CXV2bzm0BzwgPfO/b4zBFqyZBTvta2H/6yKUnqU+BfB4nvPOfYczt - ywFedMmYT5b4qkaTl2p0nLwbeRL94Q6cVPTwfbohYokZU9EtDQX4VrJ0Zk2mr5aB9h/wxwfd7WJV - vKe+PLHzDY74jtCp22V/smPM0MHe/adG05/fdO8D0RdJo4CtGrEC2bdi4is5ztHGwrCFm3QpZjKC - O/3jZ2A6Xnj/CCIdsBUTFHDIpyMOzRENO/4XECoMIObuNyzodV3QZVh9rG1iFZFhsdM/vwcXeltU - FP5AAJymy4l/AUFEH0bVQ5/5noijJa/df7q2//DBXN4LWFguL6ASXt35gOdNpcxhOaAsixOi16Rz - +7k89bBOEmfmUqDSpTurxZ8e9EWzTAaqHtME2tU5x2qUNMPPZwMO3c9BPUMkXwf+wgUZnD7PI3bj - uq1GSX000LyQDstLS1X6it7Mf/yNRPGiJTQ+CvSMLfLT+yuJaPbKZsAxfU3OX+8UtWNmt2Kfhkds - Dd4x2lK5s+CfX2AJqw22Byv4sE5iB/uGeIradzrXcBmX459/M2xo8zJIroecaF4egmU6hx760wN/ - +mN52uUI2TodZih/rxH7xydCkMY4TS6CuuZUteBz5l/k7/utAxZK+P6iOymMpKFLS4vl9LdfGof6 - atVnPUH+7aJjS5XzanKvtYS6g3UhkuefI/aPz+zxQwpLOanU8JMADl8lwlr9ueXzlfmISE8fGtbo - 4kVLdTz3//SSdJCGYX2dqxbteDmvZ1+i/ABPDPCz/Qo1YrSB294nHz6Ec+rTh9RW49/3cKDQ4sv9 - LbvcNQt8cffL/KM0zWBSolrkKjimRNHGMF/RZxZh6cMbia3UGdhfkftiV5dvjK+/xOU/ojhCq/Du - 2PqWU/7lTlYJwvjQEy+QWkD/8seLjRsSosKl2/shQkiKysCGGW1qPbePnX82d6yoXKtSH1cNSOxk - mzt0gup8Ovw0ED6nAitWL7tcSKsY5Uz1mo/226y48f0YwastJ5Ls/tJc9lUG8+DwJJ5qWwMX3oMe - 7dM/sb4oLJ1+3FcACI/YR2kgg7mNphH62dMj5iOII659WYFYzBImYfvR/vkTpzbjo5kh09dd2vOv - /OPjxDfEe9Qtl+MCdn6L85c8qsvOF2BdCCzRy1EGGz6+D394vvM5AWzJT3yCnY/Nh7bpInJ/lgLs - 5tdCzKi55OvBXBywJbzp/4aJB3M/FQnY/XGfD+5MvsYry4BJmD0ik62ju56GUPUVxa9FxwIjDEEJ - dv02c1HcuZsE1kJcq1/tr5mgDIv3Qhb4Vz/A/RTRQw5bcCVEmsmhrN31LU8K1C9n/h8/p/rvaMGd - L2BzmG5g2fU4LPMlIMaOBzTu7wtM4LIRm+ViwF63Xw+Jcrru+WKO6LQAHyyq6xPFiQa6nB5WDEvP - nGcwYzni53VS4OsTlj67KCwgyL+kQK2PrY92f2edwsKCS1Jt/tG/LdXG2fQA93wyM/5mgIUwHwP9 - 7d+fftz9cgMejW2aEXt40+XjtBdQR4/3zHbfzV2GRjTA4bTciPQ9PKo9X2TgT0+exn2OxMgUMWBq - Z/NRyOk5kwGhB5+RIl8MbhwgXZJtYqIsJgnj6kEXcLk16E/Pe1sd0lUEDwjt7uJiSb2d1SlGaQpt - jHi/1Fe72vnEAjMKrT997JLOEj4wAE5OsEUf+Sbb+yyqpmOJdKDnfJmkdoPdbWx3ffrcRxtYHPgm - tUpeL3+pFid8a/CbgLefdHwe0bdZeVBxGoVI/vcTTV13T6Gv+Uei/QibL0+7nUHdlD+88wdKGuuz - wR1fiLbe+eiPP51cDsQ+05amuphPd4OdCF182esZ9bdJavS+2yU5F1Hnjns+gs2U/bByqJ/umh6/ - DHQ4XpgZh3/m2zQfIeBurU+0p5WB9bqRHjoPtZvRvl9bjkMHkS4+4rPmVVUZRaWPgvrkY+10+7mL - lJc1utFMnNufLbjbX33q2ywlDuRZiVhg9hk8JFNITH6e6ag/wgMcQnTGRTfIw86Ptn/6wlKJTtmJ - lzNxH65DtMbohybrsSB2WvchknDfryQmZwmysmthDbeYMgedEeFvVXU/iPIhottW12jn0+QZrfec - HnKmF2cB6ru/2qpb8xoT6Pd9R8znKuVMOosFlOMXN8+f41xtj8A5AC3Frj9Nju1uf3pNqbPQ5wJz - cNd31jFgr1f4R1VTwc6HHIiSVzoLjCRESxS1Pjyfs4z4XDCoi8WLKXS5U0xk3J/zVY2f47963+4/ - 5XNv/gy40sD448cqt+tFeFOvT/JXf6PsNzBgtmwyNm3bBxu6nUbIybmMlZHf+VxvitCJAh6n67uO - lusbBqJ1pJnPTEngbn/4epz8G9ZkS1dXq1w4hKow9LnYufz5oT7Qwrb0hWuB1X96SqnTkMSb+a5G - BsMM3CZ7xFZySdXl+FxHpA98S9S9nkh3v/evvkpkzmsiOq9fCeZyBOe/ekR3TWMI//xaUyGXiDXF - cP7nP3uuh0G382XAp/hLnCf50HXU8Qaf80f2T7RIoz9/Ef4Qh7DqphxdWBzE8Fu+I+yyL4MuhaMt - 0JPeBrn37av6O1+oPgEXu024DmtAylR8HJ5frIRVGPXesfCgVM4pOd9egbpimDJQ6h8OvuqPa8U9 - 1KyESyWdSZgJn4qunl4iPRTv/vau23xrTrwE39cNEmP0pIjb4wU6Xanh265ft7IfUkgiffM5p1zy - 7RvdDiD7vSafdfmQsg9v5VAdvd7+sXkmKnOZZA0Kh5NF5IHJ1K3SXQHuehebuBrUkfgMB17pMBMz - uMf5Xq8e0e5n//GnfDtLhiE69BHj86+a8+kkjg58+fOTKKba5OPqp4v49z2F4+EHFucZltCNkgif - k9HPt9F0JVhC2yFKyR3UVS8HDajvn+ZfwmrNtyn+XoCFsxL7ZXsa6CvqODBbQvePL9LJXjT0tRfH - X5TazW9jlR6gs0RngiNQAZKdphTep5eMbf+5UTKvkwSq16zMvRQJwzyUgwhHRemIMmpRvsqTvkHv - HB+wZyQG4KD7TcBe7yWyL38qGj/eMXiyxkQcpT8PI7fPtfpfrhRw//+VAo15t8R6aGtEOoIa8dZc - W6z+yp+7AE6rYSaY9ey1t6SiTGxwkC5PjB0tO6t8we6j6dvbQnxb0wEb4MpAbBNY5JJwVsVhVymg - 5toacbYhpCt7rUrU3aCEjegKwGoLUYE+TGERrRDew0SPkoFgVgDi1b2orvoQMuJPKyscfK8bWGRw - T+GHeVrEEzIzpy8DlODAkAfRRLq562FvMXMan+N8yuzWnayuaKF8HGOSqy8hX4d0naGGxmrmDynj - Tq+zrKHzbQr8dUg0Som730pWwGM+OGtCN7e/G3B5rz3RecECFGaBgSYcbzh6K1213owjAxxbMoji - nwPAKbXJAbM4Sfgp3+yc46JDLf4E5M6ftDrRJXufStA/o3XeiQdlmFtQQATfJlGHc1VtnfTOEBcw - IrmKSVNtgXGvkZ0UP5IeHz1YzesIRfxVamxTC+XtT9MbZPPuZxYO4zPfTveoh+UGGGzNXZ3zwbXs - 0bzpInG7y6AukfM+oIqfP0SeLzGl+enEiKzV1iS+wgysCKwC+Pv/7TL+5DzHoQTGY32fj8PTUGfS - 8gHsfMPGxjpfh9XNFQk+soQSm2PvlIgqLyFBL74kDb3GHfOALZF/EyLyXK/HfLGadEbkNrr4MXJG - znzvW4GwFwMSyLhVaWppJXTOUCbaI6F0pS/9AqHb2USR67vLWMZDgBPYLJ/bby0u7TFu4dnZFKLc - fod8ub71BDF66hDtUM4DVedZAveXHxHPtz/Remu2FsbNM9jXK1Ia2I8S6nIRkrR1G7BpxiED+8+P - jT6GAz2mFwG1240QL0pll7eduQSDiJl50WBdTdH5K6IDMz3wObvytJMSFsLchhYpgp/kzoly8+D0 - dPZXEvZEv9+CS2FjhzmOhFcxMOGPZvCCQ52Ekhy47D08b5A+BEKSx3kbaGEtM8oe5TYzb3XJZyZU - RJBZNU+CpbAiOtfPFhx/AcQWAm3F7/sJPAADgr/rIZ8rOzmgcXnL2FslP1qztJRQYQU/HD6nSV2g - zcSoPV5u+D72R7rqR/uD1M+jII/wXlTs3Nce4m+WQfA8nYahsJYR1UHqEt+TWLp6SveE81H5EGtL - JMr+flqMUCKEJOBEHfD9gcnAFKgzltbzwV1O7lqi2zufsFybGeByh+yPKh8xViq0DX/rgePw1fHV - 02XA6rfegr/0EuC0fvbRevLZJzoPT0Tk/bwSu38nyLEVY99vf2B4uLVo3x98foWRyx3OcIbC+cET - ScaWy9RbLgFHP8ZEESHNu9hiFRhJ7W0G8kul/atiRpSD6wH7e7ySMI1j9Lc+7xWM+bJZYQuQfEiI - 5c9tvozaaYF6HpvzShjeXaxMb9A3IZBIbqa4//Bvj4eZ1QaLcuEkQOg4J+yDfOL3Yu5FQF1FpBmY - HgOWpj1dxO3StcS5+tFAZ8FI4N/6vCkK6aB+gho+uQuHTS+7AH7twChydfObj/N0r7Z3djmgY/pw - sK6SKV8Gu+/ht390uLDPpsvHycsB2lUB2LGTNqd/5/vpCRfi3nGWL6PQ1MiAnk7+8G2ZSkNBjniH - RJmbo0sQOyZAsWsXm+nl4E6qNSTQrfeHnq+PD7bP7LTQ8vKJSLfLOVqSoniC6PytiSI+ZZV3dNRy - +3nCkoqLgT5q7YDCM2CwcWxGd6uFtIEP42uQ2+P4pbw6NxKQ1eBI8uM1iPj6N3LgQYKZGBmJ6SIr - BwVadpIQjd7v6sIMkgMDawnx9WOZbvPi6ww25cKRa1DFKhNvcoy4bF78+veewFplygzr/iITucC4 - 2rTILsWTfTrjyPy6Ls9xbIIGZzKx3JuGS/9GnZdvacBR4O8l8rNg/ctn95euRvSw0CcangXjM1mm - gvbTng0onXILO2/jCdaLISfIwx+RqO7FHPiuah34jhqHOPEiR9s2xzH6lj+K/dx0VZZMNxFiwLZE - P36/+Xo5HSyoiwwzb+Q2DNRKTi30DZmQQHkagN1u5Qjf3trjSBXmgQpTXsPfxzHnn4U1dU0/9oLG - THBwcmw8d6SY88Dve37ic3gvhim13z7yQSuRtH46OW/mnxKF5xNDvIiLcmYrRR+is5Th6+mbuaw7 - r6Ow4y8O1kNfrbwnL+h17O5Y/7LqwJNhEeGWtVfifU8pXS6qOYvm8fnCIVHlfKVHSUMJRBo2aXio - 6BvdRxh2PSbWJa9p14qMAt8HB2B8zyWwbdnYwnuPInyTo0fEJsNQgpWyHA4MeFJ79FxGVBHpRALN - FVx6rkAA5iQWsfOri5yJ7+WGzuHPJHEkOYAfzzRAoNN74jO1PfDRL7Jg/OoW8uRpQVfIvFp4WZkv - iU6B7TK9n5dQE7uVmKciBd0bww3KsPuQIPUtylFFM+B+Hv3FTzZ1X5/xLx/cX3qVT7yQKbCXkxo7 - RTsO48E8WnCLPx72fhZfLXznQHTRzjK5bvBcsRGrB7DdroTI7HimpIDfD7oJvoeNcctyvoDTB1SI - G3b8twbmECsBcOYekqt08KKhvIoJmlopJ+c2Rfn6xsyCslvakZv7vEWcbIEaDvfPad7MsKTLc6YW - JJclmKnwKvZ2QbRF09OqcTyCfPjL59D2moU4hLzB9njbCsrB7eB3IPqAVX4GHvRaryLJVBmU6T+q - /4f3xIiuOZj0lzGiTO0hxs3HoZwYfxaRdSfgd5ymqRusuhpqW//CfsHAfN1u7Qz335/Z41ePtlBu - Z/H3TO8kX47msNyzWwBl9XLEOsGhumnNae8acH0Ru0dztXwNZxP1UVYIjhFPKTTaTLTQciApRj+V - UEH8oMvKfYl6eP0A36nfGJ2Lzp1HITOj9XccBfjIYkou3KoN6x7v0Eh+DMGdZFXruxoLWBw3GXsH - oQbLbE5PeK33LhvWlETLuCoKeH3Xaj+/erRVyyEG5zI7YeuhhfkUPt8iYtWxwy/O3KptuS0H5O+D - RdQDytS1yPMZVm7oYWX9fMF6UkoJMHrmEOdb/ygtr1sMg81i8XkQezpUpRVAQVA0LCesTJn17JRi - 9vhsO8plLn8qXjXI7YNFbmpkVfzzFRvQaIt0x593vjkfjoG9HNdzc2BfVffZqgvc+fw+VNuv2L98 - GYPMxd5j7tVFokcL5L6ozpx2pO4AtqUHP7t4kqxox6p/WuUTXjQsE3XHUyq/fh6kXJ1i7G9ooNj3 - U3jkjJm43+88EPlFfMjDBfpCfPoMy4PWEvpFyXUWEuudU9VvRTiNhwTb7+rt0hO8pOgZkf0V2CUA - HE+sEvy8N4t1IcnzzhbyJ4Qk43ymMFmXSsvg/ds/aaBBNBUKddCRqtCfynJTtwG3HgQbes/HLUvU - KZKDJ5Si7YJ1J73T9XJfFXDyWQsr1avMx0YOBRjw2kLwy1rpJsafDYovxSXO9SipU5UpI7DL+Lnv - jwO41JucP3wmUnbswAasWwtz45QQmfcXd3rqDQd2PuR3WLlUPHhZEGpornwWqjwY0dgz0JiyyIfp - NtNOoJ0FWeCrfh5LebVKTlnCcK4dcmlPvksZxm6h4DYzdgiRwdp+8kJsGS/CZ/IzI3Y/r2iPJywJ - v70RplAy6C9f7vE90BBZCtxcpiKvnU8OL6puUCT7KxBB7vOlsbgnnDdTxFbWRMNCkjqAKXg2RN/g - NNDs0I/gzUQ6UTK9UJeN6QTxR8zYp/6xVtfb/PKhclCov1rIjVY3d6ST4U2IOHq3X0mtOAcIDLqQ - ZN6+wyikxgGuzusyi+OZieaTXkFYjhrBWfLYckpe0wZ3vPIPk7MMU7zEKXK1iBI9ksNhec7AgotQ - 2z7qU6saD7FygdDeJGL+7lFEixZxsDE8DWd7vp4PonYA19PcE+dQIXcVcqmG4/n4xnIFB7DO/egB - I/kyGF/kH5iEXGrQX3wa4ybmyxLSFB7tIZ1F1sYVvd/UBN5YO/O5+jcM660Re9hTQSXByxEq2lWl - Bf/O9zml7LDv5wddwY/65VGPKzrXSQ93PubHFbNG2x7vUGUSCStqXUfTGQklZN/JGdtj/6JTQdwU - lurdJJIk2dFWurElKrWD8Dml14pbh7L9l8/3/DtMQ/m8wD/9HVbQBROUPhwMPWnA8c6XKAtmD8Y6 - Ff13bzbuSr4ghq9TFWDXs5Z8ZbvIE/UFHojpYGFY7mJmQMX/GESRrlnei5OaoEiURBKbJzaaDlUY - o7/9Dd9uD8ZGvgsQiNDHwc4Pl4N5dFBscSPRlbOrcnEUeZBsAcLKxNrRpn+sGJTbicEF/3Qp/+Wq - Fo5vguftIStgm040gLrIMdiVAt2dTF9KoXK9utg937DaCkGfQaG+3HzmKXAubb6kBaHwlInWXLC7 - PZllRr/Jc/1lH0S4450D//SMji774C0vbaBpHvU/vVn9xncPIVfXP3LuxGO1vMu2hOXFO83FrSJ0 - We83CNU18rF85E3AyCBM0dFoZP9QH990u7DXD7quBM3gCT8uNfErg2tlgZmx2ID+/vA6ed1Kf/PP - C6DO9FCgekw1oqvkHNHzEZei34w6dgT9WE3P+/MDvV4f5/d6XHN6dvfBZnaSzMt66IfF6bwM3GdJ - xzaIFEBDbWwBChaLvD7snM/yO7b+/j52W3XMSSutLXqTq03kb4fULZyEA5I+jUZMJL7pZoP95z2X - Gk7WQI24hecMIPs3FVvmfiUmkAoJ3l9eRELmgd3t9ugsZOfpZ2YC3IE9vltwjy1M0jFhqx/2/Uz0 - R8bHjyv/rTbQXgo4ZuI+qjoK1KlxFwP1H/lFMu+bg9FX7h78Tp+cOHZiRcsYmj4UU1HBprmK9J8e - qb32QW5hFw7EZFQNHleL7nxTo7RTpxhyXfmZl1OqDmzZXmN4k+bNZ+XiWfWYz0YIT5fJF1MQ7IPP - dAUsRCj/8X2OTX/F3/ci6W96DNTEtwwofWTM5X7+mK19LvBWFy8c9w9E5wEIzz+8xHKBSTWIgeOA - 89H74+9+tbaf6AmIzv7+6eeNlXwR3G5vDjuqO7ub24cG6tnVw398+e98o+Tt6eS667+NXukG87oU - iR5uG6WmuTHoLz+Zr3erUnBhOXifFX3eZMOgXD2en+AvfwSOq+ajctA1eKufL7w3t8mXt/Sd0fP3 - 0rAa30PQDLj1YXB/eLgArhXxzvSQIJJhgg1wIsPyyN0C8OcAk9ee79fjljowo26HZcdzXPYblBD1 - uZ/NqwjG6rlS3UNXWr/9tzmPEaUjSeHWFSb2WdpG5N31EsT7oJ1Nru8qLa6GA9Vv6f/5h9XKab8P - IJZwx898v0I1qR4DWeCpxBH0V1UdFvCEBN1kooexNIzKR+CAGTIQn9lNHpid/yJTVR3/+GjI0B0P - UwOtHL92f6ACW6tfLjD2IUskdvi51Mw/Hxh6yoAtiQvBulqdCI/py/G5Ua4rypinFG7DNfCFCrs5 - 3zLzE8rbOhL5vq50OZLnB0hTa83xDAxA9a8XQPYdn32WFR7VymRFCk6PViQX5fqtaLOKG9z5znzs - pHb480cBsFKCH+EdDuu34DJQ+i+XGDbwo+ke6gvqAyshDy6gdHuc3s1p90ux2g3v6E+P/+kzfPkE - cvWPv8e39oktjH4u1TpkQb7QZHL98z+5Z7GBS8fyfhNdc7r4Sugjbrn9sJXfSvXffl2eGo/dUBuq - yZfGAP68iiUuM9Zu3UVeKsb+gcVS/B7zVapkA4X1HGO7+H3BCNd8/svfJJHeEuC9QK3RXz7Ws3lW - 128nl7B8BAjrprjfXWOzFkzLTgCCl14t/Uf14J6vsdJbHuDa9fOB8kZHLEXfBcyRVmQAl0bnn17c - XaV3Ne9hSh+neVW21V2Ph28jnrq7SLyejV3KPeMFhuPU+Jx3H1W6JpoE39qSkvxghMM//ytpHIn4 - 8s2OFm6iHAyb78PfuFWrJiN7cCDjvNwHsQQqSlzO+Dsv/mfoibr9HNcBb4/22N1GYSBnvodQoJpN - TGeSK/Zi2DGUSBUS/7vchn73R4CVWha2Ii8bFlv+KpBhRN6fy3dKiYlfqbjzwZnWxHAnX6oD5LKd - RmzZKdS/7wnR+lKJbV/fw/KD7QF+pqH1TzaNaP99LRIyTnZE1J3ftWGsjvDeaQ7+80/Gv3z+8Q4T - VsamrwachR/0NBYbP52VA2uhUAs+1wT42+73zIn/ieHt2k5ETrRrPl86y4Dyw2H/419rp8KHuNUB - sYufTrfGZySIT/BFLl8vrFZjO6R//rLPPc5bRUX+XaIV52filZ4JWEMMWgi24xvjDztHY9OuAdjz - Jb6otRZNadfE4JNwNTaPdln9Qq3uwVisEfHwUA9z0h1q6B3lH8Z8EuZje4x7OFTZMrM58ugmcF8D - XLilJC4LFMBPhWiIE8edZ6QmN3W7sI9/eDNDg3TR8OfH6KOqYDu1IaWPJdjzi0Nn5mmW+Qyg40A7 - LMeZrlygcsB69aeiPAR7/qjoFh+HAohfNtj58JDv+e7550+SP/6yTJIk/NOni/Js6Ha5phlUokIh - vqllOdXDojnt9Rv/AMp7Ti1y8SAKNosY8/atRjCXCvRt443PcvIZtr0+AXY/yj954Tr8/ftw97vI - xSXUJeaPPyDZ8kX/GcDQHc7uZYQ7H8W+qYl5j7Y2hWr6/RJZrrZoHg+jI/qosnf8MNVOH+7cH//y - hVuvDnQqRA1aD0MlxhpU+cRp5APeaan7n9dWgQ25sgL//OXsbRwo/atHWVqW+/xzAcPwu2cedLa0 - J3t9I1rw4/uER/WNd3/jka/pR97gXm/AwQt81YVCo4CMk+jErB1JZV0zi2EuH3PsgpbSUVjI5y8/ - 4YfMxHRRQ8jAV3Jc5+M4fdXWvu5N2d/7FX8xFKvJ+b2eItWkEzlHdxMwm3Vv4aedpnmBMR023U97 - GKviOlP3Yla8cvwo6OwsCnYLV3bXzt5EZNzbxhf2esNSvGAMuYATsSrITs5kdnBAUr7WOH9MasQ4 - wcD988NsJMVqd1+tBna25eOH/xiHDQ+4gSIhBrFCz9j1OO+IykGiMwSulY/MLS3gWNCI6NXtCxbz - cLqAPX9iCaSpyjv3KQGdYRYY3+aUbl/PSkD6WQ4kv34edOGv0eVPnxH5fDIqjk3JEwav73Xe/TZ3 - 4eShhmdJ8LDRxq26iL/zAuC1ksg5r7lqYW7BEw1tPGDfiu75JkpAgyzSPV/0vjndZvkmQnegZ2wL - H5kywJ98IMzO+V+9cMGoDFDRog/R3V9aUaOMUvh58hz2xCwH49mfW9DJ9ptYx4dDmaH7bfAo98Y/ - fJl/XJSBvT7mH74LP0z861TANb7Z2PirH+lMlaAA+Sd/9+cB+aSzIv7lZ3d8P4e/+ELCbJ3xhZGq - YUK462GfXF7Y8mcrZ3GcQ/hNJrjzWxJt6IA4+IK1Qq7dR1LX5Jj7kMd2SnQr/UZ7PNcwnBuH+ELY - uGtqdz4cb2yKjXnTK5pP8wh7AK7E/gUXl6h+KyCOpnfsGcSOOMEpONH0TX9uW7ehez0P/vMv1fAh - U+4WJgocyuvBJ4q45UsjhyIMX0v6zx/aSHsMoFdizp+yKw9a43YXYJqezL3+tLibGy4lYk7c5sMT - tAdqH+4S3OZMwdZhPOQbuuxPanrF+zuP1YwuUw1vr73rqjc/VHJqPyMyjmGB1e+hBtuuz9Em2sqO - X191qKRzAbZXExBT+5gVvypxAiXtUGLjRjVKTzGVYP074/n7YwV3+vOz/Zh9+xsucMVaATRg/C6v - WFa2Vd1ua8oAraRgBucbdokHIw6abNNhyc0+7nTVNOevHkrOcWuAReHKEl0VHOCz/l7zdWFEBxaF - EvvcVanAVJ35BgrUsAm29XFYO0TGv/zsC132rTbHeQagPQY3fBMGVe3kZ+BD+WGxvvirYU61jrXA - Xzzoz1MBNiZUBIA+Q7jHOwDUzPvPnz+860+ibhr1W5j7grrjUUQr+2It4KVrLcFhkEezXw0KDNkc - Y/VxndQ/Pxrs9TofPl9Hui5v5/nPn5SQXrpTx7kOnI/7E9X9Ihe3aPIFQeByxLvuXRCbAXngr57j - 6YeCbmbo/W9dCvj/0qXAs28YC4aqsodzncCXaixzOjRQXRkpv0AUFxK5CG4U0ZPEiTAsiwfGto3o - AqHTI2Y8n8kDbt1AzU/2hOlqjaSI1yBaxUyv//68D7VPWdE+9zbI3VgNy3mEhyXI5RpdpsQi5998 - dvk06SQA51bDzisS3LozegZKqdNgL16DfBPR2iPQMjZxlQtwqdJIBcpXJcYO1+yzAPPgAB96oM/s - FFnq0miWBwztqxGfub3VNUhqDr3enwb7T2nJaWzQC+zem0XkA3ustvIuZpDaLPBP4pOCMcjlBqzc - bBHfdHk6DibjwCF8XIjRnh/VmMAHB8VH+SPaMw9yPoyKEgrAvuIXJkbOvp19tmok2sRNu4FSlm0P - 4NaqB6IMsIo2Ztwa+JZOOZEAp1EORouBtDNnkzvckEtOnjCi/lGxxDO3k7u1U2HAqHR//naQmYry - Jv7AC9dMxNw+rTvrabbA62huvns1o4HFvdX8+17BrT5XJHS3BnJ6G8wH7W6qbDQxJcoHVsHWiX7y - 7q2GBXrcrym+PZ4u+D/SrqRLWVjb/iAH0kmSIQIi0gXphBnYICAiTQLk199FfW/wBnd2h66yqiDJ - 2WfvneQc0Y4XD40LPOGEcTJYfs9YR4DtXa/H1Wwv8/y5Q2GPL0RgXA5m8I1bONnOjwZ7XA98ePQj - lESyTe8FXw5CMucBAlpS43Oeg+I3n/IGrvw9pCf9qTH+5PMjxJZNvDV/NfXcDx8DOZSN2NOMYOCG - +9uDqWwAbETyI5zhi3XQMRxGvafiF/zT9Ffk+X5Jo5IehuX4SHagCNonfYbfN1iM48Kh4gBr+iT7 - bzElHXKAeuZnGk9CbK/79TfK+Sl/YCPum2GNbV9BdWlmNMNxW8z6/q4CuehNeuajgz1F6q9HO8h0 - L90dYMhKl41ImJSYPrprHi6mLaQoYcmR4oapgJ+IKiA2WRPWku/P3uqkGujD3y7UCr98wZqPliNx - 1Hmqnp2mWC/RuIIQph09Pm+lveDjfga3fDToq1aNgQxdtMJr+XxQ1aZLyM7bKVrp2t6pvZ1SXMog - TpCOx44+6t+XkaG7z3L5Td7U0jO5Zr1YzNCis4v9mF1roRIDgm6jLJMdLzxq0ayHGW7P6+28NQ6J - Jq8p+tgrptvza6yqVwO9E32kZjh7jI/LI0TO63Ej/hvdbT7dsRbtGMdT5yWfAJ/MLxOeD7aFTwG8 - FWLpAgL1yl29vTkkGlOnPgUmZ91worsnID5FFMjvyHFxvo3P3P2sJ6LzRLE6NzGbOWlnwQ9MD9hC - 1LP55vdR4O1U8RgrWg1mqhvW3/vj44F+BjGM9xZ4iItH1VPRFb/r1lTxtxoPbB4VwObulFVQWrGB - TV13ij4fRx9emCd6ouf7gBM82ICTMnTeUsmuxg9+VcHX53PwuBMPAEuTnwrvx+hHgzI2bGGguxF8 - 3I5sW3CqNiu0DdB6Ux8UX8tqWFJBiZBXtry3slphXCY+Kwh60mGDP9Ni9ZRfi0pz6w2mes9iFdSD - CrFrbqeESVmI4S+Q0WAkHkmz1g85Yf1UaJhIiI/l8CtImnxkOIpAxE//dAHrs9RU2DanCJuu7A6i - WW+9uJ43z2OnLrDF1ULlv3jSk+sHUCEfAqhd5MlbdkEwMCEffFjZ0hdjuz/ahAgHBwRtkGKz44Zw - 2f1ABGXd1+j9vGbaXD4jBcIdzKhrlEY4PuoiglqP1X/zx1pwWdEF2ga2L4d0mKnuWZCvtAs17VYP - +V3yauHZrEqqD7GqrWo8W2g1JUhx/PkC+q7KFQ159qb+5fe11+9ozRB0wgWrx6S2P6B4SpDsV0qQ - ydF6vUjxCH0THEjTqhNg04PmcAwrTJVqqsPuOXotUBL/jAPuV9j8li/gsu9vVLn8zhq/4RVoO+jj - Z2vJoMfkpqNYijhcPByJsbNhP6HFXQQip3tsM/eFVqjvngwb334u2uLatigORI9qH76x11MarDA4 - cxV9ZNqrWFDs+XC82CM+XsKaEUm3IpneE4OaHWcX/cFBDRw/VMOuADu7+/XFDM/Oh9HoYUJ7cSPL - gw+4e+Fj4x4HblByB27j4x28z0ubK8BFyLyvV+zJQ2+vaX/soX17dt6UafuQvqtuhfWu9egxq8JB - nLdCUU/zVFPt1j2A+PyWBvLzxPHWS+kOwtfMBNDsaxefTseZsXfVzYgrZAlrtw5thYN5FcXUwNSM - 34NNspWTAY39BPtvtD3vfHUQ97k8KFaDb71SSS3R+XifaSpzN23Nzr0CsuPr6DnOONjzgtAdnofj - D9+4sAnFq702yB/fCn4VxcVm9rO5oxzpIn3JUgh+WViOaMuHWH0nOeCcAK3yvSOEPvxDBNZKMRT0 - +Sgczk+jB7j3F/vgjVVAdUnNmHBbhhZeufdMTzXYGqdAfQWKDRqqGx/N5hIYC0gtzIrGQdaH1JHZ - E8XuAAn78I023ytVRclX33u7FEeFwF+OHdK1BeKQzdYgaLKcw0BoJpwemFqscq6MyCbchNXLjg9n - nM0+ekbJh/zUYsv34SWBM30csK5vhXSP5qTKaA9Sb/+mQTE3FYYwvTyvf89TfO7eL0ecMX1p/pxt - e9znXITKs+LQR+PwGlPRvoGnydnR4uGkbPJXIMCfFgVbIbURjIdDqSBrNmqKd/lo0/WZBTJsvqIn - o2tbkHehERS8nC+13CWy2axIEKiW4FLz7U6MptbXgyhJXtTTYaTx3SHf4pO74Ff8ObPhCIkPs3Dg - qQ7FNlwXT4Eois4a1r2+KxZpTBVoSm5Pz+96sRehbTm4P+4SMoDOG5Z2P6SwPbY2vvxed7bcUO+D - H3I3iXIwQyHKQxm8uaoi3FzcQ8bpjfTHD6lmjtWwnIOziuhMKXWOz3lbX/wdvPadQabjcGDUYdcS - Kah3KN7Gl0vgg4N9MSSehIff8Pd9uCND7cFOnweaylcIf/vz3mOjtw7k2HAKwMJ0os73rYVzko4S - zKgk4IAXe23kl8hDvPYNsZojpxAeQCHIvVYxvoCls5ePMFuwtaqcXnh7P8ycLqlQ56uGKl9/DgkQ - fyrEWO2x11oyI80BRfL3BzB1wYu3F0FWG0jjIKGGG77A0uOrA69i6dLQDxVNvO9cGSgvL/EQOLoh - 2SU5kYWnzGGr4v1ineeyBYyrGFVev4DxxWMK4PdGCmpgrtbGeFR6uf/KKTnst8Yngse1sHohmV43 - /BFK4ZKDMSwxVXBea4up2SZ8MZqT1b5eBlZxZx9eKqnC1qoN4fixUg7sD/1Mdq9Qr8mPTCks9SOl - 3va5O85iCyOLe3p8nR6LdTohC1ytneA16Yfa8+nO5fBMFAO/bokGVl2bCDRqRf4Xb2KuPxu5wzlP - 5p+xMMqlZg+1Lqpxsbda1gtWAOE+8AcadYlo/9LeK0H42K0EojaqiXp4G+DS7nl8mgReW/74RvtB - jJ7hwwPUHFwCpNU1yGHNvHr+9cUKU6VSsfJrDgPTbHUGMLszqqZ+CUb+t/r/+NtFcWbACsi1sM05 - St23ZDNBDD0Htuc3wke7OtdrNuQQJjmP8Pmt6NpyfioceJ/3Nr1UnTGwSrlY8DF2Fg0mFQ0rPO98 - aBJ08qqUdDVbg9fzQOqaYDV25mI+J8MMBl7PiBSFd20W2Y6Ds668adKTkq1pf+mhdKcvqoRo1pbK - 0Ql82K5IL/IzBFzD1wRu6807iKMKhP71K6GmcghfMibbExuIBaXpyHn8JMQaW1Q1gP5CAnp8CXAY - R2n2ELkmstdo73WYPh5NoOG+Q/piV6FYp1+rg5LkVzJv+mX57HIDpl/5TM9nUS24Lm5H6ECF4KI1 - rOEff9q9uyMBagC0SfKPCtoXjkzzzu3tYVxsE3JL+iK7z9dl80qWFp32xUo3PhUykaYJ2vQY9s6/ - 19AXJkjgxmept5tpyFI528mH36HEp+dp0oayV8m/+BceV9XmvwqcYS1WT0L9GRRzXQq9XMH6Si+d - ohSLVD4d4KnmnipmErA/vgLmIW2x8+OdYXXqiwOW+9HH2lmMwYrJ6++g4JPMY2dpS3cIVhSS65ma - 7v3DlmURZqg2xZmIBFjDqqlRDumnGUhHTa5g5JbMMOn2T2+XZCub81IapVtODHpyTseQp9I5h1wd - 7Ih82RRu83g/oe6rO49jiAvp9r5QjtoUW+JhrEklBiNSD03qCbr7AazicACYN4rYFXMMliokPYwG - k9FLmNsDiaiygyQPLwTR2KpZzm1VRsRMo8/8Kdn0V7UjrFxm0D9+3r84yQCzs5uxOyev4S++ke05 - 2sYfbE1Y4l4CqO04fEkcOIzxaHbyzWk++BH0Ppt3/Kqi33s2CUht3RZ/FRlh7EOKNz5qsyKze6iF - dkCP+08Llud4HOEzzaR/63WpkTLDUCcePc/LaC/QKRT4p7fyVo6HeQ2qBqqn/RW7G17yWWH18L3/ - fUhfjPJAXyfzDulx31Clj0rG2m+rgstj8rHzvFwYu+zlFdpKLmH1VnA2ucVRg4RzRb1sMuSQeZMi - o6FZdXL42qxmcVEm8DcCm7AoPRX8SelKuPkB2AjarZHV92DBwyUQsNGTEqzzQ+fQIrVf7FjCr57O - 1seAjp5EVHt6Rr2yF73D0+GuUL+E35BxqdLD49MX8OURB4XYXRUPHrGo4HPAHmwSX44MGzmGWD33 - o8bs5/iUg/oKsJ1LtJ5PBy1HcOYOnjRo1B65rtvBGX1z0o2HuZ4/VipA69pP2OWmJlz4B2hgPJ5W - +jd+y3RlHFquXuWJkbGGy+GaJ/LhNo5UsSWLMXmRN7wTEXa39bPMDlfC9xXP+LLxIXJz1x4SbJ5o - nt5VW3AM2YGdWyb/4n31W2TJvneycKIPU83UrXDuVLWUmkY4a+wnqE8QaTMhnHvPiuV67iCMUPzF - 2v4+FWsy5z7MayHEKqtLMN8uPxVKoqHijc9qG9+/Q+F9/nn0qKeDsOVLlODxQA3onzQWuLYM1qfh - /vlFw5LMNwsKKOnxORa9cAbyIfmHH1SWGFtOSlkCF+6nf/izjOOhAntvL3mHLT/QPuyFv/nxOHIO - 7QUUiXRA10eN//CZ2a2Roo9OXtT92opGq3OtQ1+aVZq/lj1j1iM0kStJObYPmcXItAstZBAbYcX4 - JPXSLTMHfRWesX7WejAkStUDWzoHWNMiE4hMC1s4D3nr8cWisfHrsDuIhWeM03uvMLEepwgMOHh4 - /aZ3ueEMTXF4NSs23fsJ8PyDtfCLnCO+RnhfzOBmpsjwOQOr9vVXz7ZllfBVFydsS0YyLOKth396 - 15s2vjoH0QzhfiZv7LVypbGNj8PP1+2o2zixLWbayQfufmfh8yF7FizXkxbEqfrykE6tcN2rwg6m - xXCiZltW9nScZfL3Ptg9zXu7maxkRtl6HvCfv7O4c+bBUxNQ7PECGsjmz8E4VV70FOYvW1wujQcl - 7Q7+rkAUvORfVFB+nio+RXhgva85DpTF05vM7xcXTubWu1rp7sAjxXjWlo9HI8A3LvJAE5N6Qm9T - B3tL/mDVptdwaU6cjn7KjfcGtxTsNYns4M+PoOrjGtdr+OY5JPhLhJ/uogwC2Fs7uB/emBo+n2rf - xBc59KcHL5v+W7SbkcDnOhf4KgyxJr5XBOHf/Jqb3/sPv+J92pDZOhPwT4/+6VP/diyHlUpWCayw - djzJ9wS2vEmTIJGZLQ5ZoLJ5ko8l2lcfhR41YGlCtkIZ2hB69DpU+3AU63cPn7rqYwP6H22yTp8V - 5TuUYIff37TNT/BgYTsp/fNPuX0YpH/4uPHPlS1/eo7eTJ7w8rwdwXA2PzS0dA8p/ccebbVsYfYg - KuHEYwN+b4ul0FavN2wyNxvm52i0QNOnOz55YvDnHz/RJM4KzbzzLaTS/O7AHx/QPaccWHF9j2Bb - n95+l48ac9i1QnbcXGjwuPLDeomaFTbyDdLTmcCQRd7DA59AULDulxc2o7e+gyMZAgKq4yec5xR4 - EKOVJ3teDeo5WjNfhqTXsQEK6Z9+A7vQselrRG4oZuusogeEL/qgbA0n7WZEMKY69hA3NRt/q2fo - hjlHrUi+MCb3mQprLGhYzxPb3uK1gSF7Zh5wdw/GXczFAYI6MGrn+Zmtf/7QNj/4WdWoWOv7QwEl - 3hXU284b/vk18j+8LY4/jeJA86B3S3xqkt4ZhO7q6kBBnYPVnnAFDd8SBBsek6Z0xnrOvsfonz+t - zg3P+ir8rXDTx9jb/KjZbrMSXOzq/s/fbX/q0QGfgFOopxlrzZpp14G3w529pdcqtpL6cf9br1Td - +C3Z+ADc+AH2bP+hfVUktjCUq8um92XG5uYtwW1/gNxa42lvfloDk0PIKO5/jc3k/qqgH2E9QRvf - WDjpEMAv8o7UmMorE2W5deAp3OUEq9sVkM0vAiePmfQ0VDZbEphCALSopubuAAs6/X7OX74iVRkb - mviwMwN+o2lHNUM4aotUJg5U16rDG15p//KJ2oYtVj5Hpxak8ppAPwzuHrtYVsE9ej74x8ft8/k7 - TPJlTqGVkgdWDPUy/B7PXQsDFbVejdzL8C8+tvWCb/LlVIukX0uIDb2gmDqLvXrcd4TPA2f/+/n6 - hxfnQft5ZRb/wOrsUw8GnhR4+xtnD+vuwfWot2/HjS8ZbD2TzIRzkwj4cnv5YBlS/Q4/+vjyuI57 - guWVThBsepXI3/QR/qLn3KM73XqND043kBZ4PYiXUKfKmZ0GBrm4RZu/hN2MtcOy6QUg9bxC9fdL - YcsfPhjZHHvoS1xtHbNUgrOuvjHWW58NVS3rgP86Oc23fEO+3E6BzvEmelyruoxX6C+H4k860jiv - Obaql10Ez9sViDIVK23znzzgdu0Jnz+nXzFZfFqiw8UXvJ2pNEO/xJUERftpE4bc37CqFyEC+s0g - hGffxR4TDkhyNCcnqnrsDabi8Qng54s7fBz5kK061AiC2ZN53EPSwjkezR5O5eh7K/cD9vLiIIRp - A3pstTI/LP6QyXDnbY26/r5/m7Id+C7XGJ+kfc8Whb5zlGZzix9puRbNH9/lrUHz3q7gM7LElQzt - Z5V6tSEcbf5ZYhmkxe9Ej8exLf70Itz8K3xW2KqtnD+08BK0Lr6k6ydkoxs4CLznjJ5fEx7mWNsL - 0qc2Y/rc9AO9SD4Hj5hXiOgfJ5vcLb2D74d5xxe4PuzVOnxVaDz4zhMl4Vr8Nr4nS6USeHKcLgNN - mrCBbakeyW/zM1YxNDwooKjHTlI5xQi0pof8/NyTNqMXJmj2ufv3974PGNaMzzII2W1/9+Y3umvL - atQSuKrqF3u/kWlzXe46wK/xk5qXmzLwX/M5wssgLNjUK87uj+akwKzdfT2WfC82q8cpAWgaGxw0 - uR1OXwWu0GftgrWg0oAwN+0Mhffp9+fHAnY0Swv6rFmIpOtjMRMQCOAucJge5TkPF1qmPqxIMVOb - a57h+uf/nArFIsPlINWbHjahXU7utv+S15s/06J/fmuxiAXz3lgHJE17qt1OM5gm6yr9+X8bn9mF - c2tX1iFbTwPWnqMXjrbatTC+tD9s5O5Ss82fg+jlE3KICcfG+wmokFvyF8bsKoQzbxwjWJtziO9H - pQvXH/mkKPffb2qWtNWmYzSMkP/pB+xUHrbnI1xSmHNmh+2Hk4L54hcm+CkxT5i2zsPffhI04/GN - Y99qazLPZYMeY2/Ri4R9e5Fzt4F/fn3xNx5BZjXQvsRnfJKVV72i+Jj+4T1+8LajsUfAFHhaU9+T - qBkV7NPeCTSzhfP2xWFm4jLFFrQvtzPZzXzH5ttHUiD/EiKPTfd3zdirb8Eff9z0NVj2uzJHi9R8 - N72Z2sx4X0xYSLqOH4N/CokazyY8GvLi+Wf2qaf5FLTgNqgFNYubE85beS/4lW4nrKRiZc8vhY1y - 9mreON34z/hmFx2+NH2mHtl/wzkVTh5IIsn+l8/Ztn8FxGkfYUMf3EEgeeWhtTjN2GCczEjSzREs - fp2HAx2Vw7rEvQyBZ1+pt4TzsIKKQVj10QNry7tlS6+9UmDe5yvdfh901WtKIFVOJ3r52mFN//jQ - 9vmffyfw2ZODzOFS6q5ixYbbZ5RB4MkBPSsssFl3VRz0vxwpkP77kYKof24SRBPt6XftJLiGT0pa - FJga76leBcvqkFILpGNNaGMHUHpgzpPih6aJ/o6z0FV4hzQhOC+WW7WuQM9/OXa53Slk0nxVYC/d - 9mS1JlLPCqqeCLm/nJrxo9bm5JOl4JrWDhHTxLWF5/5xB+vbmenFPk42e2W/Dk7qouE0OFaA6XTZ - oVCt355sng723JqRj5SXu6MeJ19rlnw7FaYqOhOeZ5SRC2vXrfdw7X147GvC93YREJbVAxnATbPF - 5tMTyKRxj88Xca+th5sbQNQMT28NltlefnkpoGeS7ugNPFm4gLpx4Oet770FByJj0iHdwSGdCS0s - iIEQqKED9yNX4WC6xgUXF1cDuW6QUT1TWE0u2bVEf+NtLi0qiPhyOJgnj8bjP8MI2KsX4KFjSkDT - k8APbHAbBdnnVaPJqH4GtmO5CWPOIfgIpkFjopruUPHrnvSmqR1Yp6utwE+Y7AjsQscWrlFoIC23 - PXomns9mQxl6+KoyB9v7SGL//n7WdQZOGersxT5KI9xrlwVfH65hi5zw6UCRoBlbT+VrMzkiHRQU - Wnj7+1QPy9fvKigJcktQE6jDwq39Dr6XJqKxLn5tEYLrKGPPX+h9ek02/826J9qRJsZJuRfBnEpZ - B6dysWj4cayQ//z4FEVX50HNgmThPKrxE52qS49TxYiAqEuLjCgKa8I6vWdLfxZScBMfJwI/hT2I - kQJ3wBQiSJ9WcAmXWyZzUBt/IX0peVgIs6KVUEqpQJbgqtskPUURKqrXSq27tJU+Ge4CdI3zhHEQ - v0OSsiiFF+5zoUYfNMNSHmAJqcYKarYGqL+cmHpomy+MJ7lgAi20CP1gvFDflodw/S4vHbrmXGCX - yz/DCtFphnMCY3J4fvyC+wb3FlYWl9B84H4F2dWzjGDLKTTaK3mxevuvAGnkWWRV4nFYHxdFQTt4 - mcgukK2Bd1svAHFeQ3wiIbG/WdJV8AdvC2Hv37NY9+ZioOuxvhBUG+9irIywhGzyEMbuRQl5R5pH - dHIbD6fywdK4685/otl+hlTl07fG3Y5ZB9Ny72G3qONi6upyRqYjMBIR6VbUqYoINMSmxmr6vmpU - +2Adtmls0udgYJuXuJpAM414rLXX1zBLH/WO9rHwwma8SwdRskgC7TA40eP67m2adl0OBNZjHHa6 - BbiDcFtB/Ut+2Cr5RzFDVTMQjgOeGh9Dq2mvDQTS91LhrD2t9iJsheOaWzR7/afswLL/NDuU5Y/E - k4PyU4hy1HZQPLpfj4aPkq33c2zA+HWUvd5W6nr+kfsKO0894xcXdwW7xZ0JOTyp1JUto+C5tYcw - /MEQ65yth2LhlTuIiHEhVXsI7W08CVxir6I3Y78bWEsHBe6t6OF9TkXL1kcdB0h/D09P1s26aEsx - luCBtxg+ng9Hxi9U7GH0PnL4prVmvXa9aqBc2v+wXpZxyI6T1wF1tM9ElNrQFkoxlhGkv5Bav8eu - XsH71KKvIBb4/IdXAThwsE1vJk54CrWp5coAfcEtw+egdWxRel4thFbOo34ue4AbokcEKtrk9HVR - uXr4hJWALvZZwZ6WfAum7I0SKDj94uep77T5evafqNdiHmO3j9l89f0UiTKl1DzFisbe1T6FO/X3 - oc5pBNoUp+sdocy+U0MQwTASJ17Bl/I3fNQfb8A44dMjxOsuPXY9qmcpf7UQy8rBE7unby+nSl9l - 2kEPK59g0Rhv+RVqhRjTpA/0YXHVRUe3CCnYmhorZNkTyaDhIxGbN0PSyG6cOXRWfoi6QDgNXKpQ - A57WJqOavcjhJJxqH16P7wt+Nqs3iMfikMBTZffeNt6aECcFB2/uDuKLe5gGpkyLg1jiyPiqXt7a - qn0/BBIGdZoNfADYQcuiQ+DWCT4qcgxoIHAtOvAmo9HwOhYsbDgCgPNTadovZ3utvgpE1HWMbb2j - cJH5UQc/UWREWOUP620z3/Jp2WHzc2y3U4dlhGA6+NRMhcwWb07noa/AF/h4PxnD2nSXSN5fo7e3 - XFWJLVd94VDokyu2ZyiAMRJ8GQi1WxJOl+aaeJ+8kWuj4Kn76o82d+2nDuqdsPXy3k5M8189RwXa - M2x/7ry9OFSaQUmKDONP3IBptUUJqePljJ0RTMXse58n/FLxhrXz5WDPkqhGyLspHU54r62ZcjXu - 6FUVDj2eju96FvZHCUFhuFFVcx2wpqcogeVXOHkS9/nVQy7vdFgUD4KzOMOh8LbNGUlD3mDTewqM - lrv3DIFwsenruhyHRdsk7xRfzb/5sYXv6+XIEu7v9CFo+dabXlIQ/+2P+PJ4zGC5i2CF5aKbNKvd - 6zAP0SOBxu3CUTzy7bCEt1WHOddFf/gzcOJFDdDy27XU5l4VWL/lcUXvpY2oLqSZtvBlbaD+fYPY - Wme7oNKcKaiSAcO6ZMo2Ofzc7RZu98VFPXI2M7W1Q2aiC9ipzz2b1+HiA4GXPHpDHlfQSug9+Ikb - hE9Z+aoXY+fsQBZKKn7eiWuv1fEZgOKSpzTnSqdYh8xq0WkSIuqUuK/ZLS5N6EbBw3tra2Iv1+Qw - wneXXOkprn9slhQ1QV9uPtFza+GQe66BAJ+WBGn2PM41k56ZBdbwTrH9gCwc//C49DkVx8FwKdg1 - 53JoXxXDW/jMGkT4uwpo9p2UKtXOD/miYA0S77TzDtOHgXHjf/DdRVcaV2wMGYHj/S8/Yj+XCWO0 - HxuQJ52L/YdHB3bXd3eYqvszEVBg2qtzpncYep6N9eb71pYO8xCuljdidXhHoViOv+1We2h77aM9 - Amou8git6uv/yzcrETQLCXd8Jzz4KwOsDSMU060RlTbScJnCd4cscrrSq+WqAzfpZgoX8my8pgfv - YZbI1uv1BipPao2iXtKdDGEW71PslrbA6IniHNihf8LFyaF2t9+KdXdA/OALmcJwrR/JE7rmWmDF - qUaw3vnYguakH/DxsyJ7SUdggIV/diSumFOItG8aOVhHi95T+6KRKnxach+uDb0oJzucMzVXwJsc - PvTqnmp7OAaHFCLxUtPjh/8Wc0qvI8zyV4JPITOZoLnpE16qk06NIde13/oQVyhnsojPJn0N9BzL - JYBce6SK/rRDQS9TAuuLXmOnHrZGSvylgSD8GRQbFqct1l0qYT8pLX0ABop55Y4eepuGjV1fC8EP - O4dUxo6y0leYC/Vo0kZHjhAk/4fXb7O3oBetP3oZ+BXQe7KLoBD3JrZdbg7XIot20Ko+PjVMzAam - h9iBV/doke1Qf00u71aSW0c0sM5RXWMOZDkYcDnga5jVw3zdpdsNbGhTk5A7WAGWfLgrfyN+CdVD - I24QRPJf/nJ3AwCrxV8JFB35SsB+34RLe0K9/Ew7hM+OsBWK3E0eLH79k3Ri8rRZNqMZOonie4fa - XeoJ1I0n6a/mvOUTMqy7epbQ96vo2NefQzg5lqGDwk/vNPqZwsAaU9zBuygPRMjJMeS062ACPAYS - PjpvrZgHzVTgPIxbIWIur6kYpRzE5SpSXTPpMB+amyz/vd8xnvRCfPW7HWg8Y9r4174mkqtD+Zpu - jaU+/LlY9TVtYOo9TGyFH7lmz+FXgtpt3vTRBGrNJ5zcgRpke4rvl5qR09Iq8FTYLeEyhQ1bfLZw - 4CoBm+efMoicOTVw/xkEbM5PxH63s7z7f+uZ7vrAAc7p+/bAPkoBaV7H/C9fUex8lIHtYGCg3llL - 6j9WXLDVn0ygloPhzbFFNFqmogG3z1T5BFd7dZrcAfG03erNjshmX3sIoGd8D0Q6GC97sYnLQc+d - PzTy0zWc/vjBxo+8lj0/NVWaJUD3ar+jhn1mbJlD8Ql/knrfOoFaw1q2Sou2+aQb/tRTljMHrvFI - cGF2uBiF3SuBFF1rfByO262V2Wu2LfTAAzsnK0TDPveg0F8D1W7Xu73sxJGTrV2p0JxJFhOmXIUo - sZ0Fn0z/oU2PO9vBX8aPHuy/Tk2dE8hBtkgX+jrkqB7P90GA31SM6LGZvmz+1JEOjVF/k/1jTcHo - D74HKleP8SnpXvaWf3TIz7JORD8Y6j9+BrX84hHBPoeMM+lowNW5D9Qu9BUspnDKYTo+MFnjk8/m - 86deoUGSElsR19rE5WkJ8yj18KkQ+2LOoqsMBcsysNu3tb2qS+rJblLdqcPtCZj/8uO92MtkBT+x - oGa3rug0NDG9W8AoOPJTSgjdi0fx+9AMy+vBRVDs5RSfetEFbMNv5Kkn8I/vi0imFvjx8wM7jTiA - 9WUpOajuhxE/U8NmwvB+7eDlcD5jS02rYr1kWQkpvt5xsJ81W5D72UBq+TOw3UX8sJZblYdWuGHs - inZub3wjgrPvpR4zjlndSq6+g4an5/QxPbV6pjezk7yb2mGr6rqQZaDsoaeoH6zvFTkkH/dYok0f - U7s/KzX/CKECeD5XsROHXb2sS9/CxXOeODvxoz0rqL+jP/zUdKiHPLiOCTi6PfdPv23r8Q43vecd - YPcexu3/A5GzCk/S3XJYdpdm/dOzWGnzuSbnNDNh5yln/PJWvqDvvFLhH993SS2Bn/ddHJS7uUVP - vvQMqWYtAqqbAmJDoVcgVvvKlKP8OHuSmx1rYXPr4fy751hvtHe4xBB3QDjLZ6z6D1yzuy48IXU9 - A5s1/9GWtupyQPe7K9lp8ldb1cPUQo9XKdb8rAAfXz7coUeUgRo3oQQ/Vz3osAtWB+OMr7XVLQIf - hv54pSow34Alz06HDj3/sK1KJFzfn0SBubDdQlSJFc6ZnUEY8s+YcD9W2Uy9eBLczgXRsIe2zVvi - nMj9Mg/4FZ3UUHR6UMKHJxEaSMtNW7+vlwcWJeS9fWaJwwpSyYGHoCPYOY2FzWQ5i2DE795Yi7uz - zTV8sYNQ+N3++U9//gE8DNJM0BxmgN+HHwOu/XP1lnLruFD2YQDsxXewk467epruvxaO0QwxTtVU - G2XJIvA+pj+sf2yFLZs/Am3yfXiJ9g1s9uZEH3IVNgl9rCmbzrnrHP70hgkev3/fBzhUBXouPDdc - n8Q0t21p+y+f1J2h1B10+jmir/ehqdcMRx2aD9cF4+rRFBSVVxn9PjHxvvcdCT+XixRAIP8u3uH5 - mYs5mtId4JmlebePisEc+cQBv7HM6QUgnq2+FAjwNCc5DtTfpLHryAVw8bwnvjRo3fTbrYG7cxv8 - 03fLXz7IJlPHf3x2OXq6iXqsvKjnkl9NZ66RoUL7yZOZ1IOln346XJ3n4EVGg5mIKwZBzHmE8E97 - 0OY6K1ZY2P2N7DPrVrNdQlPYqbcHdu0iKNaPe6mg5g0ONl6yGvLfc9kiReiWf/x6RI6dQyoQ7A0l - F4KpcXkHndY281p3HxX/4nvGrMNYyVkxb34ErOrDTLVvtrcXmW8M+CoKE2ugmwAz5jWBVQ1mrGqL - 85c/Goj5CniNZuJ63vwVGGna5O3uxNUW+njvoMLlK2GBY4CRGy/d4ZhZAjbfzzycF+Noop06fP70 - dzhrda3CT64DejGygS1uNEN0vnG/v3gcBO07bf6DU9EwKr/DH56hMjSO2HKchi26obdwnmCKC75J - NXHTz7KEuzvOuEbRhJFXZ5TFKKWWM+vDyiLdhLfsvqPP2LPDKX6vKRSx5WPNXvJQ2PQuEoU8x+eT - fAlpGEAO/unR0+Nrh2zLf/DaaB7FjfsJx/iSef/iA/RDWcwyVVe04QfpBgGGXclCD/7xTdP3RLAY - Ox2iTe/S3Dxl2mKapwj++cWLmYtDd83SBN53jz02W6MY5m9wb6Af1qq3mhmx12OcKvD81kdsUfcV - sse6K+GZ5+aNjzKwKMAW4M2FkKa+d2NzUPsdDNx3Qs35+WD9x72Uf/4JdWRK66UIQA8jRipqcp/f - sMRYLoF4nzoScbTRFrf1fFCtRoltYbzVdPMz5M2/pGc3H7T1dPJltL8m743P+uEYB5kFuco1Pb6p - p+JfvhaupkWdT2HXPO/VHnodCky9unULLg3IDphpwuMLco6F6ECWHv7GY1bAqxiymV+hnkQn7xr3 - sr0E+tmH0fNNqO6nQcjHhtzA8gAK6j4SZVjmQ29BCOwYm2UiA6aKSwndwS9pGh3qcNDbVYVECzF2 - kOaHc9F+G1AQ807PteANwmt6NOB7FK9kH/1UbeSvFw6GYj14uyCxmLj52X/rnboJ4DQG1+sMFS5d - 8UvsYf2Vl2sON3zBp5z2YNPHKhRu05NaF6nRWHusdrBuMoidje+yxyqUoMJJRB37Zdjzzcoa+Odf - nzzlwVZPzJ9g8/OwEghuOL/TIodvp73QP3+CHQtv/nseIrNosJdwxRH82z84Pw0XzDK1tisa0Re7 - jVCFvLJqEjxOkoztOApsXoW6A9qQt6inFJCtJVUqefOL6JX9fgPjw/QOSD8eaawNJ8ae3m6FB1f6 - /YvnpbS5O5TWtPOoc1SY2IhVC3NTcvCfHpq3+Adu+l3pEQciGOWjStBBdffYvGcfm3TuLYH1I5Ow - bs8+4E+Vs0Kdixx6rN/tsJDL3oQ39Qnw334Ib/i6BHWz1QlQJbLx6Y8F9D3n4tC5t9pMnMcK3cd5 - JsKW/8dTMFSwvGUJtY5cVX/aoUxQdeVymv7pW0W3OaC9LwHWNn60omxu0N/+gD6eKzBmjysHaeRY - 9AguQ7HxhR7Gwn2i2fPjh0syrAT0gv/DR/a71GyP8xxeoljD+txJ4TTPvfenx6gm3B0261OdoDfj - QjKw7YqJVWfc3/4T9kprCql6MSRQX4ya/vGTVjl+PNA7c0kOUD8y4Xu67OArOVjYSSCq109yuUPM - fEhP0jD/01Ng2V0B1gr4HVa3/nZo08//AQAA//+kfUvPgszW5fz8ii9nSk4EVKrqm3EXASkURE06 - HUFQQEQuVVCV9H/v4POmB52e9fjxQatqX9Zae7OLYuXr8MnL6xpaW2+krhCO3rCLKgX+9H57yc+T - PnTwl18CadE/pmFW7T++c0wfljfMLDmh++HqEvAUOScLn0T9SZ+pvtYSY+jchwqtRDwSENWzx62H - YsPUay84sp7LRbexfQcqczjWufmM5tC35V89De8FSU5pEb4IWrvdCVvv3WgQeTaHP31w8TeP62Tb - QHm+h9ThyDGYcR5lxd7XGb54dhfxFaPN9mQLepDQy5Z/G/EZotPHd2iU7GE/muKCmEpwpWqvHznP - oSD+xf+FL3By2J+uMPsUMJBPtuqtO6UncOM5Od6D7JkO5OuUsKOnN7YXvY+Fm7v4qw8EvNzsezGL - lwuS5PuOBn1HlikRvau8g2ZLd4RN1SSHlxDq52XKOXFcY/46T4iU4Qqpbr0Olay8Ng7M1LbEd/W9 - jviaNCYwNvtz0DT02o8/PrY+dgl2Ntqe8yL8DuCHHzXrrUXSmL0aJERmvJR0Rz7VUiRAObrvsWXs - Gm/M9vwffHYSRS0VqzIicOHPwZeMPGX5WjXhos8TJR2aiN1e8VM51JZIF7zNyYInQbzR1oHcyXVK - n8J3gj9+fR5pEIm9ouooEYlC8X58epMwG0+4GuSSiGpIU/KZLyY4ylWEVeCKfFaEQYcKT1XsCf42 - 5Z9H7ypL/g14Cj8VzxIhgYteEPAobiLWnBMFLnp7sLEFyZiTK97AdFtIi/4/GBM6GzHa8XqmP73n - x/+hLCnBom94KQcTEYE/qlEAViszlU/bKIAP6ooB19uqGr/3VoT3r3vCp2Af8MlwKh9117Ck+zOi - xvyLn/GZpNjchtt+dMnmCrfbbg7mN3sYUxZ/BvgURAffj9Cq2K8e8vt9h5ZvPOZcSA3rwBypNm4d - Ll6yYw49V9fp7q1jLq4O3gR/er5nTQEXif+YlFXUPAO2HrR+tAVTQE4krqnjvl4es0GTKKcRWrio - 57fHV/YpB0/1FpFKPW74D4+hlwoKgryURXSFT1f0OZxPwSY+G/1aPo0bQJPoib3Vyoz4tiFPeNA+ - lEhovVwMudQHuvda+dXHPTl4nxp4v6NywY9Rur77sg6/EnsEgit/jaU+PcFbp1jYWPzt60VTg+6o - +FD/JgoeK/P9orfsm0BpzyuDnIt0AkJkx/RQ2Voq50R1UQS170/PNOjOVxpQ0S3Ci78C+cqzKxRu - fUvQku+534ESLPyaTEu9ZyLMyf6/phRs/98tBQ9hlZKS3c/VvCHOFe4L3cAWM3eGZCHWoJVb3YMt - GpNqzu8vAX7eQ4R3cSFw1g1GiZRUv1JfPsUe3ynoCSaTXbDlyF4k+S89VAox0YLZ76kxnfbOFTZD - WmI/pQOYztAZ4LnIX4Slyasiut+38LjCQTCmtzIdyUsQNt6lrvApfb0Bm+5FDo/ISeklfbw9Eh13 - NYKRo1PbP1qcJesw3AadawaArG9Vr/mvDh2O1Rv7TNWj9WVjCEh/fOaANKBLubX6CnDt6Bq2tt2H - s8PjViLzJFyorxhbj5/hdEfv1TejebnSDFGygwTq28zDJvn6FX+eXAa1PNGp4fSeN7dybEK85jLG - kfACIj7rCXIPIqS5EtwiEqPJhJf7TqSqQ3d83W5lEc6FUmCz2umVnDYvBdINHqgRHnswVlp7RaP+ - HegjrZt+3imeDGe1nrFJItNYy2oewFyJVzT7HGVA/N6WIdJOHdVU4Qmk6SzpKDTjFw3JV+xn+FIZ - Uo2Nhy9mXxn8eX8IsHKtDJ/l5Ar4Tr6ZMLRGA4eom8GkJbKvlGlYYfsaw55e2VeEoJEZPezvOZiF - wrqj7DT4NE0p4TN8OQwljwuhTnjovfVby65wdfuQ5boJF4j+p7YRuLAc+/IOA06qJoGkiwuapgr3 - KHfrCY1hs6WuvNzt5sxTh7ohveNdoaTemtlqgpb104K1csXzJ5pgbTIaKNVTAlzD9wGGlsroCcVW - JAql7MDqYyKaFcHToNp+U0PpaBTUMWUrlZfRX0i1wpGmxa1JSVYCF3yFlYE951OnXNx/TLRq3FsA - i0D1pmZzy4BahgF1rWidDm/bltGqEbulpF8acmx7MbQ7+U1j/3OK5hI/BmCRacKWlgZ85Htugm64 - 3bHJzL03tPPcoZeAInorAtuThqa6w/vhrdFzSBAYJaDqUD7EJJCsy9Wbs6bfbD/rV0/TMD5V81Z6 - KFDxuzPZspfk0e57zeEL53uqxscZjPqhSpClP17UE4zUm5LNTQBBFqyXt4ppP7yh6iNN7Y74EsdK - xR5P3YbAfZU0nJFRrbWTMYFVI3fY9FnSU2Y7MaytQg2ky2B6E70od+h9Lg4NnGuVTpbWX6Ennc50 - 302hJ0V7cSnZZy8ck7yNJvApa2T7eRlse3vT88/Gy6G1OjT03r2/gGTvOoDS2Oxx2EA7Xdtb3fmd - N9ZMxfV4dcgTGG2KVaBYeVqtb8u4wCOxd9hhr7U3v9VyQPd+ZWJVvt/6SZEKBSJBpcGszUUqtlSt - 4ahFLnk7y+BCb0QbUDknE+shU6r5zkYZ6ul0w7/9H7r2OSE1aqeglM2DJ97kpIZ7Yl8I01JxiX9e - CWVQJNj4GCMYL4qXwEeRH+nDRA4Q0y7u4A2JGS4O3oozmp5dtNgntuVE5uPrLD2hc9Z0Qk32rHi8 - g8Nmf8JCsGlEh09hdpFR7vsXHH2MkU+Hb5GBRz6p2PbtV7/O06hD6RvaVF/it+goloy+4UfDXgfC - iNarVoC9u/Nx+smQNxwqlEA5u+n4aNI3mMcSdsA+uJxa28Lo5T0FIbynz+DPH+jjxXPUZa2EY2FX - AK5G+wxGyUbHt6qHHls9ZQY3pZ/jo+LB6pvfl7utX8aKyNviVTF3Ji2UaAtp3usXMKC0bZHD1Ire - /Qr3U2QdOzjFT4jNdLA98jqZDF3ulkiPRJT5VFlHF4Ulqant9Ja3ni71HRXOyceYnYjxO1/FE2uT - Prp3BSb4PQtwqyEjWKmo5rNXzzoKNuWOms5Z7btafQ5or9lruis3Wj+t9UsDjY3xpXjZf6n7vExQ - v1iIXYJf6ZpcSxtNpmnTYnneZBv5BDqcbgOCLqbHnY2ZwVP6zHGKNjKf3sLS5f6ZbRprlpPyR+Fe - 4fw8JzQk4gaMEU4CsP5UHjXig+Q1QWEqkDd3n/p78xrxtJwdxLzQwA9UupEEiGDCTeS49HwZaoMf - T14I38LKozv/M6RDcN3aUDpqBXV6W+mZKxYE7m7fnHipEnlSbfsi1MKrHCj9UeHctEwHDvZ7jfXw - OPKxG40AoVV6pLg6Tpy4n9aGz/qVURulLhBvQ6jDrdq8qFrt3F5SPs8GXXa1h2/aeO1ZJyYE6im7 - YXc799H4PB5NNL40Sg/I14Aswe19eYuvxqoqlHwSd6YKfEtXsIMaz5Bl04Zo2Q+sh9nJkJZ4Cevz - 4U2TZmNUBD5CHVycMKPZRdgBcSd9rzAJUI+z7m1x8e2sNvD+DVS8Tx+aJw/9QQR4Pcs0MpXGmM7C - PC1jza7Y0ELSr1GTbOBMBE7TvXlPZfQSM0Svjx0+FUIKRPwMGohb40gtEumG6E4iBEpKvtiR1S7l - iRYr8LK2zgHfa7z/88fl+TQozvue/fZvyZfBvNfWFdlIkQLDUo0ICz9y+sVnJ0ZfVL9w+LllEQvD - mwIfRXbEmWLcvL/42J7xA7tMtaP1VTptYMy9E7XZUERT2mUdBFIW4qusuinb0HkCb1WsCZSg11NR - vzWQfLuA3tKHZkz39yZAl13jUexU72rc81oFz9nc0SuZmEeme5HB+Px4Yi3MYoNbcLOB83HckK+y - k/hEnlYN/eqOArGiGZ/D2G2haWwMnFrObFALTgqKLk2EL11epsx0RAVeSxIHTFHldHi8PAaOmJjU - iPWQz8tdp4ApNlz6aX1Pnu6X/Icn6TG2+37peamRotgDkZuN0cuS4sRwiT90v8Qb7o3SBsqHhGCH - JYzP3ej58Flm34BHoDJIouzj7SlYWl7iPUl7/RjY8IefLPbmnF6lkwKTtExoXn1AShugBGh9zdbU - 794WkLdrHEJzavf4xryhmi5Su/mzv4QAJZ3oWYvRVL1DqvbBxpu28qOGW5Q0wRqlLeeymvjQuK7O - 2L0IHpgW/AEWfE3VZtY9PgyrQLlIhAeS/wmjtUt4A5Qu2eFdJ7p8dMAqgBa6BwReuFFNXz7qEOyl - Iw2IV/a8ka8mNN6lSsPQNaNB1F4bqBmeQpDJ6oqfIrmFS/6jwXYcKqr53xYI3STROK5AP0V6W0N5 - e68w3m4CPq+yLIG3IruRvrsbvaRFh3wbvYWE1BF4ebMN9iX8rV9LXzsuXQXK4OfDbewVtyZiz8jN - oDRZnPqpMhs/e0Rf/SvSuBDWYJb1RwOz2CaL/wXVeJPyzW991KhODlgHFY2hpZkrvNzUBeY58gTI - vJNBd8puC/iVbiZIL0zGTrTWI7ZKsw4Ig3cJZpKTdLYceQOOZLqSlWJwbzge00ZZno8NU7E9zmwn - ARIJ7zgnA03Hsc0ZbM+A0EO479JBeB5MsMQrfN1HacpgqTgwf5xzan/cczolwFPBVCNGtevhZrC+ - mmKYsWymuPrcIv7yn3/+QC3ZE6t5t75P0NJ4SIAEx4q/ItlVlniNXf/qpVPHaAwGK5vI9yIcwEyz - TAY/+4uqWun5ljAbLXiOOkU9VVOQqDp8HKMO703ZSNln67nKQagsqjOnAjMYLB1G7vlCrx93Tscf - 3hRSPaQ7zSqjdq34MfT23wRfHcQ4S0wMgTJ4BQ4c4oHJcDwf+nPs4qho52gaSkeAsRnfaEaSZZDk - aAXw4xAWTFYoVsxehwxdbUkh3bbZefJavjXQPMHLX/yaabxv0Qg3Dr4XNyedpiyXYd6HPbYsnwO+ - Q6QG0eaxoodC3qQjepxU+JrOMllZd+XnPwGK3MuFxs1opOs5Qi0Ea7QN2qIOq/nLxgQEj/eN+nuc - VWM7bSEU5+d5wSOZx/yXc4JD/DLJl73OHt8SxYSZVYRU22vrnhu2JqJGiemSf3qDiTuJQLF6O7T4 - +cPrShooAACxLXusKg3zmsMuPUXUrWQh6mLjHoL96pCTTcr7anwGTxUGmb/G+mV6gvEMNBX98KZw - 9VceO+/k4cdXA641a+Nnf6B5XAK608pNNJOXAH94ip4+gWhMzGXTz77xPjx8vTYdLQLtAe9ooH0v - nEvCcrFGdlfoPtRNQPnhIQKtIDXRmGMBpkZeDtXkJmJXlV9ptfBxuOHnN8XNVBu8OG+fPz5B7eJU - 8inIUx/kfnAhsm+/qqE8r0sQ8/2JOtqFGdO9vyfgtx++iVxOvXc3/J237dtaz26UTVA9jxUpSe6n - 3/H9HuAxdAHVFA/2PNlqHdq8ngcaMmfH2V463+GnLS5U8892L47XjQmrfepRVX4jMJQuOMHWyjfk - 6dTVH99Fl088UcvpPwb7UuWkPCjYYVx9tim98+IJa3awccpes8FPQVbCzPZe2CgeB29dW4arYAnm - OHJkGlG4tLBcnFOG3bQ1UlGRLhvoXUMR39LHyxtdik6/eBpw/1xWPAGMoHM+hfSozUXE50hq4UOV - RZJu5280s3iVwAWfklX12UZU4fkTipFzWvLvy+ATvgVgE7ku2RQ3P53xbWBK5AgSdhb74Y+6fqIg - iDp6SG/6Msj3UcK0aS/00IAhnVfH0AQ///eW72ttBGvwQK5B3YvQg2l1VmW05DuCFvxL0mojQn0A - T+yY8juajMOYgMdxBQJOHI1PjToMyg/vazLm6bCZZxftruiAbynf9+KCP4C+zT26385jxI6xnCiv - 6bJMQYiexsStawsX/ol9qgNOfnzxdJcPpNVGpWK1cWPgt/6T7z6j6Rj7Hfg9f+98nJTf+eUJDuo1 - X+xP52w/6j76XPVXsPG3mcHbrSzDexCdCImtY8+X81X+8NL1kBrzIW0diD1twtjvImPxHwH6c+IG - ZZpovaz7VYs0tT2SrcP8XrYs24H9tcB4wQ8GVXovhsnQP6guVWrE1DN8grrMX9hb+BFRIyODsaV3 - 2CCR7Q0qbgfIHXWP72GHPI6yMAGtOX2okb4Mzr02naCxvlzxYX8XwBxctnew8GfsZBaq2KNuStjf - vGfASdR4I2g6BXxVMcMJmRLjKyt6gpyoewadpln9dYq9TglntqFeKK1S1hGxBp5TYmqEWej94jVY - Po/19PXmXwnNJtCOux3VutIy5s20zVBtr0tqWWUWsRbRBgR+dqJnzWoi9uM/u+vqEPzZ60//ujhw - pOa2+/Jpio0OnoXyH7w1SLYdw7Wz3BroIAamyNEJPAolwnrvomiGbXKCi76Hb7LaRrSI1BzIg7Sh - gS/dop4tF4WVQ9qRWiJqL13WpxC6Bxliy0cb3oSRDpHcxZjuzN4wGlkYJpjuagP7Wr6u/vDWqPcD - tjWrjqYEiTYcTDOg+2itRmIrlgQSD2b0jOZzNK2K3RXmXmljx/l06fzz51MWUGz77FTJ/UQFuNuJ - NkHS/I64+z3myFU+OnaL2u8nMJoDipP9jL1QV7k45vcA1mX2wkaz+fbs9/9yl2DCtTDoaWSEJzjw - 9zVQ4jju6fNSK/CH5354t5PNAMLPJQkJ11IGhkcFZYU8ZhHbse7y8XToZCj0hU0X/cLg+7nuwBHJ - w8JnXh6bwnCDlviP92yNKlak5w6eYa6T7Q8/VDv9ibrD2ca2n7meqOInQVejU6lavL6Am3pkopa/ - n9Rx0IlLP31CE9gGm3vtWImfVQP/8p2j+SOfn2nRLq+BPpb43PSTJSgh/PGFSNo2/JevgCHBpYXl - YRldoyV3hNT6hvdpa0XjvTu1yPjUnLpbSCvuNs87LOSrSH/56PPTK1l2c8nsPzuw2GuuWM91jXED - 2mhIP1UD9aB6UD+tnX79yI4x+uU7u4tUQ6odtIGKB00cFD3zhi2vS1S0xYd8kPVMuX4MTHBWr0Gw - kmYtXUtgO8D96SBgZ4kvc1pPT+i1xg2roVunS/4LYIK/kObSXEVcj1YuUBSGSOuwoPp9Hi34hdqX - dRON99EgMPK0lu7lk+LNC9+FjR/sqC9Br+KLvgx+eGrRLwBh+9kEi74TjHF44Wz2SwcufDtAWjSn - pFZbApySiFj/3DbpTO7f/Kcv0L2/VQyWjas7/B6PJ4oXfMvEw0WGssIt7BZB7U1Xqpk/vkndZvU1 - +nj1zaEkqCtqmFQDo6y4MfDruaBuI9Z8XYW7Ozz1q5hi7Svx7rTb63CIRI59VPrp/EZzDd1t8KR6 - +rIAbeSrDVdO9ln06bL67MUHAQh2ByKy6e6xhb+Dt/aUsN6VlTdJau3Dx+o7kOd263NpQ5w7XM6H - BuG+jdie1TFc+B82Y52AZgqvG5ikzwTfU+VscH5aQ0ilvKX4Qhuw1AM66B6cPhDKlebJS76Aqx1L - qEq+cT9H/tuBkPU+9TR/x1mezD48njAJJARv/XBvXyf4PG8FasX6wNubnDcAln5C80iowGwKU4js - b25Ql2AtlcPr2wRgvdoGPzwkLvkSLvkbG1qz9thYNyrsRqunh+rsVeLLmwT4Ph7vNG+WQdM0t5eW - meRCrZAgTh6Pmw4di0bBpBW44t33moEbeY740qyoMTP8CMDt+XkF7+KxN2btLJdA7vUnmaVtB7qd - HNmwrdwu4KH96UdUQle5K+cdVi+TCdaipXfotPVDemUDSNlPf8YtLYNX1Z6jtR489J8+SXfIUlP5 - 8Lg9gbXCDd75x4pP021k8Bg6gOLtpfT48DnboHLnkT6YqkZMFmqGRPvzobYWOhX33uWir+ssQKjs - ItIIYwjauPLJUdtbxuKPT+h6bE9tFNo9Q4Vxhdf6NdBYaySDJ9t9C1GYCMHkP33ei/uPDZXsqgTK - kv8G9XRj8B6sltlnNqvmtxExuFfvDQ2calcx0zbkn39jT/sizvfrYPrDS0naeumUp68OLr/nh++i - 9XqjQ9jca3fRN0RjbKydCz/uLqZ7v5r74XkjJ8W7nsQAoI0M6p30uqKxCBJqVLVS/epBf+vBRCgM - tnoKDFROaFJs3e+AaeH9CsquPgVyFQzewkcV8AidCseKJ/czN68C7ALtQiZyKXn7in0XKMFjJKjo - Y2OubVOEOFFcbFYUghHdnwJSrdOI7c/zAcYhv/pgOzZLfaqoqj88lz8uOT5o3xtg3y50UeqcYuxd - +Lua90TbwGmS/QXPPDx268IW+nH8pcGCp+pf/i7WOwu7oa7ztWi5f/61+Kfzwws1XPIJmapTAGZ3 - 9AM4w41KE0lyvbmBYAP7S+Jgb5ZUg9eW4QDFE0xsa6PQz7p1dZVf/WTR44xfvvzDO47Za95gOIYP - Ijn1KZbmV8oXvAyaVBjIXCRVNb4iM//L7/7v/Gd3mlAoviB1Heb2g2jeIeyvD4wxEVYGP+K7D7fk - dMK+f6gMZphhDovvfkPd3partjwMC1/PZ6wx5w3mhW8CYZBabDHsRUt9wUXWU6pxzk6hIR2v1IaF - fBexy17c4LZ2vsNDuI6xkz4MY1rwGRSCxxsvz++nED99mM02xiaKXxHr5ncHc0Bf9K+et9gHzIOj - G1SKGkeLXr2Bi///4kW06F0NFCfRpOfevaQjbJPwh0+omb408OOX4IjEAS96Cufu4JawLfMb3lfP - NZjM3eG5VR47j8zhEYP2x5/eL22FLQRv1TwfXjIifjcG0HKowXCOfRjO0wbvm5VnyKtrVSP4Kq1f - /upnPcyuv3poAOQ275motj4Ul8rMZtEX/tZvGoqBVRmff/WFf/CD2sylMesuC1Ecc0xE1HCDB7nl - Q/8Edzisdl1PPvLd/PkfWeKNMVlmnENwmXJcOJVV0fJo2fDinjF2ulxN5fVWb9GCd6hV3Vg0ObPS - QbFeddSVpNL4q/fBWQD0V++Qb6NGwMACOZC20mCMn/WNgVF1DLrUf1MxK8Ireo+iiR/d1HosOu4a - WIixRhoHJZy8d3IAI+lkBah4Ys4PLw6Rlsc61VDR97wx8PSrZ2KDTZm37K8Ir/ZawR5LBv6HL15C - pGPVssJ0wT9XBem3mQzSvEsleX07QfbYpXjX5WXU7+SjidzO9bHlZ4PxDRNO0NOc1vSMmtmbslp5 - gkBplD97nIzDO/7Vx+mO2rAfXGIoiEGP/vBwug6qTwJ36+WiLdR4nvTDa6PqGmQVxl01sciC0DzT - lAZXf2UMlaP74J+Wgn/913/9j6VB4N9N+8jfS2PAmM/jf/5Pq8B/1v8Zmvv7/Wss+DcZ7s/83//9 - TwvCv79923zH/zm2df4Zll4DEaF/Jhj8e2zH+/v/+tO/li/8X//63wAAAP//AwAfYHydugUCAA== - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99ac8df5fa8a-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:51 GMT - Server: - - cloudflare - Set-Cookie: - - __cf_bm=TyZMV15EPWqQXIbTEKRZ1Fggt3ZrMRC8fwKSoQemN28-1727451711-1.0.1.1-1IuaZeYu1bHwXOBs1o2p6JiWcDrm8vtRUTauyAVfYSTF8WGo9o_nFHMe2VvmzxUrbomkaIIk65ZyCtLLaSGbrg; - path=/; expires=Fri, 27-Sep-24 16:11:51 GMT; domain=.api.openai.com; HttpOnly; - Secure; SameSite=None - - _cfuvid=8VnZjkM1FzzkfufFNQobFDcIb4Hs0eOF3OZxxZTYh.w-1727451711771-0.0.1.1-604800000; - path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-allow-origin: - - '*' - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-model: - - text-embedding-3-small - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '128' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '10000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '9987878' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 72ms - x-request-id: - - req_db08890591133dd81002e0e3135472fd - status: - code: 200 - message: OK -- request: - body: '{"input": ["ability and Transparency. 2018; pp 77\u201391.\n(20) Lapuschkin, - S.; W\u00a8aldchen, S.; Binder, A.; Montavon, G.; Samek, W.; M\u00a8uller, K.-R.\nUnmasking - Clever Hans predictors and assessing what machines really learn. Nature\ncommunications - 2019, 10, 1\u20138.\n(21) DeGrave, A. J.; Janizek, J. D.; Lee, S.-I. AI for - radiographic COVID-19 detection\nselects shortcuts over signal. Nature Machine - Intelligence 2021, 3, 610\u2013619.\n(22) Goodman, B.; Flaxman, S. European - Union regulations on algorithmic decision-\nmaking and a \u201cright to explanation\u201d. - AI Magazine 2017, 38, 50\u201357.\n(23) ACT, A. I. European Commission. On Artificial - Intelligence: A European Approach\nto Excellence and Trust. 2021, COM/2021/206.\n(24) - Blueprint for an AI Bill of Rights, The White House. 2022; https://www.whitehouse.\ngov/ostp/ai-bill-of-rights/.\n(25) - Miller, T. Explanation in artificial intelligence: Insights from the social - sciences. Ar-\ntificial intelligence 2019, 267, 1\u201338.\n26\n(26) Murdoch, - W. J.; Singh, C.; Kumbier, K.; Abbasi-Asl, R.; Yu, B. Definitions, meth-\nods, - and applications in interpretable machine learning. Proceedings of the National\nAcademy - of Sciences of the United States of America 2019, 116, 22071\u201322080.\n(27) - Gunning, D.; Aha, D. DARPA\u2019s Explainable Artificial Intelligence (XAI) - Program.\nAI Magazine 2019, 40, 44\u201358.\n(28) Biran, O.; Cotton, C. Explanation - and justification in machine learning: A survey.\nIJCAI-17 workshop on explainable - AI (XAI). 2017; pp 8\u201313.\n(29) Palacio, S.; Lucieri, A.; Munir, M.; Ahmed, - S.; Hees, J.; Dengel, A. Xai handbook:\nTowards a unified framework for explainable - ai. Proceedings of the IEEE/CVF Inter-\nnational Conference on Computer Vision. - 2021; pp 3766\u20133775.\n(30) Kuhn, D. R.; Kacker, R. N.; Lei, Y.; Simos, D. - E. Combinatorial Methods for Ex-\nplainable AI. 2020 IEEE International Conference - on Software Testing, Verification\nand Validation Workshops (ICSTW) 2020, 167\u2013170.\n(31) - Seshadri, A.; Gandhi, H. A.; Wellawatte, G. P.; White, A. D. Why does that molecule\nsmell? - ChemRxiv 2022,\n(32) Das, A.; Rad, P. Opportunities and challenges in explainable - artificial intelligence\n(xai): A survey. arXiv preprint arXiv:2006.11371 2020,\n(33) - Machlev, R.; Heistrene, L.; Perl, M.; Levy, K. Y.; Belikov, J.; Mannor, S.; - Levron, Y.\nExplainable Artificial Intelligence (XAI) techniques for energy - and power systems:\nReview, challenges and opportunities. Energy and AI 2022, - 9, 100169.\n(34) Koh, P. W.; Liang, P. Understanding black-box predictions via - influence functions.\nInternational Conference on Machine Learning. 2017; pp - 1885\u20131894.\n(35) Ribeiro, M. T.; Singh, S.; Guestrin, C. \u201d Why should - i trust you?\u201d Explaining the\npredictions of any classifier. Proceedings - of the 22nd ACM SIGKDD international\n27\nconference on knowledge discovery - and data mining. San Diego, CA, USA, 2016; pp\n1135\u20131144.\n(36) Dhurandhar, - A.; Chen, P.-Y.; Luss, R.; Tu, C.-C.; Ting, P.; Shanmugam, K.; Das, P.\nExplanations - based on the missing: Towards contrastive explanations with pertinent\nnegatives. - Advances i", "anations based on the missing: Towards contrastive explanations - with pertinent\nnegatives. Advances in neural information processing systems - 2018, 31.\n(37) Jin, W.; Li, X.; Hamarneh, G. Evaluating Explainable AI on a - Multi-Modal Medical\nImaging Task: Can Existing Algorithms Fulfill Clinical - Requirements? Proceedings of\nthe AAAI Conference on Artificial Intelligence - 2022, 36, 11945\u201311953.\n(38) Zhang, Y.; Xu, F.; Zou, J.; Petrosian, O. - L.; Krinkin, K. V. XAI Evaluation: Evalu-\nating Black-Box Model Explanations - for Prediction. 2021 II International Conference\non Neural Networks and Neurotechnologies - (NeuroNT). 2021; pp 13\u201316.\n(39) Oviedo, F.; Ferres, J. L.; Buonassisi, - T.; Butler, K. T. Interpretable and Explain-\nable Machine Learning for Materials - Science and Chemistry. Accounts of Materials\nResearch 2022, 3, 597\u2013607.\n(40) - Yalcin, O.; Fan, X.; Liu, S. Evaluating the correctness of explainable AI algorithms\nfor - classification. arXiv preprint arXiv:2105.09740 2021,\n(41) Hoffman, R. R.; - Mueller, S. T.; Klein, G.; Litman, J. Metrics for Explainable AI:\nChallenges - and Prospects. 2018,\n(42) Mohseni, S.; Zarei, N.; Ragan, E. D. A Multidisciplinary - Survey and Framework for\nDesign and Evaluation of Explainable AI Systems. ACM - Transactions on Interactive\nIntelligent Systems 2018, 11, 46.\n(43) Humer, - C.; Heberle, H.; Montanari, F.; Wolf, T.; Huber, F.; Henderson, R.; Hein-\nrich, - J.; Streit, M. ChemInformatics Model Explorer (CIME): exploratory analysis of\nchemical - model explanations. Journal of Cheminformatics 2022, 14, 1\u201314.\n28\n(44) - Lundberg, S. M.; Lee, S.-I. In Advances in Neural Information Processing Systems\n30; - Guyon, I., Luxburg, U. V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, - S.,\nGarnett, R., Eds.; Curran Associates, Inc., 2017; pp 4765\u20134774.\n(45) - \u02c7Strumbelj, E.; Kononenko, I. Explaining prediction models and individual - predictions\nwith feature contributions. Knowledge and information systems 2014, - 41, 647\u2013665.\n(46) Shapley, L. S. A Value for N-Person Games; RAND Corporation: - Santa Monica, CA,\n1952.\n(47) Molnar, C.; Casalicchio, G.; Bischl, B. Interpretable - machine learning\u2013a brief history,\nstate-of-the-art and challenges. Joint - European Conference on Machine Learning and\nKnowledge Discovery in Databases. - 2020; pp 417\u2013431.\n(48) Lou, Y.; Caruana, R.; Gehrke, J. Intelligible models - for classification and regression.\nProceedings of the 18th ACM SIGKDD international - conference on Knowledge dis-\ncovery and data mining. 2012; pp 150\u2013158.\n(49) - Bastani, O.; Kim, C.; Bastani, H. Interpreting blackbox models via model extraction.\narXiv - preprint arXiv:1705.08504 2017,\n(50) Gajewicz, A.; Puzyn, T.; Odziomek, K.; - Urbaszek, P.; Haase, A.; Riebeling, C.;\nLuch, A.; Irfan, M. A.; Landsiedel, - R.; van der Zande, M.; Bouwmeester, H. Deci-\nsion tree models to classify nanomaterials - according to the DF4nanoGrouping scheme.\nNanotoxicology 2018, 12, 1\u201317.\n(51) - Han, L.; Wang, Y.; Bryant, S. H. Developing and validating predictive decision - tree\nmodels from mining che", ".; Wang, Y.; Bryant, S. H. Developing and validating - predictive decision tree\nmodels from mining chemical structural fingerprints - and high\u2013throughput screening\ndata in PubChem. BMC Bioinformatics 2008, - 9, 401.\n(52) Plumb, G.; Al-Shedivat, M.; Cabrera, \u00b4A. A.; Perer, A.; Xing, - E.; Talwalkar, A. Regu-\n29\nlarizing black-box models for improved interpretability. - Advances in Neural Informa-\ntion Processing Systems 2020, 33, 10526\u201310536.\n(53) - Shao, X.; Skryagin, A.; Stammer, W.; Schramowski, P.; Kersting, K. Right for - bet-\nter reasons: Training differentiable models by constraining their influence - functions.\nProceedings of the AAAI Conference on Artificial Intelligence. 2021; - pp 9533\u20139540.\n(54) Ouyang, R.; Curtarolo, S.; Ahmetcik, E.; Scheffler, - M.; Ghiringhelli, L. M. SISSO: A\ncompressed-sensing method for identifying - the best low-dimensional descriptor in an\nimmensity of offered candidates. - Physical Review Materials 2018, 2, 083802.\n(55) Lipton, Z. C. The mythos of - model interpretability: In machine learning, the concept\nof interpretability - is both important and slippery. Queue 2018, 16, 31\u201357.\n(56) Harren, T.; - Matter, H.; Hessler, G.; Rarey, M.; Grebner, C. Interpretation of structure\u2013\nactivity - relationships in real-world drug design data sets using explainable artificial\nintelligence. - Journal of Chemical Information and Modeling 2022, 62, 447\u2013462.\n(57) Sheridan, - R. P. Interpretation of QSAR Models by Coloring Atoms According to\nChanges - in Predicted Activity: How Robust Is It? Journal of Chemical Information\nand - Modeling 2019, 59, 1324\u20131337.\n(58) Rasmussen, M. H.; Christensen, D. S.; - Jensen, J. H. Do machines dream of atoms?\nCrippen\u2019s logP as a quantitative - molecular benchmark for explainable AI heatmaps.\n2022,\n(59) Smilkov, D.; Thorat, - N.; Kim, B.; Vi\u00b4egas, F.; Wattenberg, M. SmoothGrad: removing\nnoise by - adding noise. 2017; https://arxiv.org/abs/1706.03825.\n(60) McCloskey, K.; Taly, - A.; Monti, F.; Brenner, M. P.; Colwell, L. Using Attribution\nto Decode Dataset - Bias in Neural Network Models for Chemistry. Proceedings of the\n30\nNational - Academy of Sciences of the United States of America 2018, 116, 11624\u2013\n11629.\n(61) - Bach, S.; Binder, A.; Montavon, G.; Klauschen, F.; M\u00a8uller, K.-R.; Samek, - W. On\npixel-wise explanations for non-linear classifier decisions by layer-wise - relevance prop-\nagation. PloS one 2015, 10, e0130140.\n(62) Sundararajan, M.; - Taly, A.; Yan, Q. Axiomatic attribution for deep networks. Inter-\nnational - Conference on Machine Learning. 2017; pp 3319\u20133328.\n(63) Zhou, B.; Khosla, - A.; Lapedriza, A.; Oliva, A.; Torralba, A. Learning Deep Features\nfor Discriminative - Localization. 2015; https://arxiv.org/abs/1512.04150.\n(64) Selvaraju, R. R.; - Cogswell, M.; Das, A.; Vedantam, R.; Parikh, D.; Batra, D. Grad-\nCAM: Visual - Explanations from Deep Networks via Gradient-Based Localization. In-\nternational - Journal of Computer Vision 2019, 128, 336\u2013359.\n(65) Smilkov, D.; Thorat, - N.; Kim, B.; Vi\u00b4egas, F.; Wattenberg, M. Smoothgrad: removing\nnoise by - add", "(65) Smilkov, D.; Thorat, N.; Kim, B.; Vi\u00b4egas, F.; Wattenberg, - M. Smoothgrad: removing\nnoise by adding noise. arXiv preprint arXiv:1706.03825 - 2017,\n(66) Pope, P.; Kolouri, S.; Rostrami, M.; Martin, C.; Hoffmann, H. Discovering - Molec-\nular Functional Groups Using Graph Convolutional Neural Networks. 2018; - https:\n//arxiv.org/abs/1812.00265.\n(67) Jim\u00b4enez-Luna, J.; Skalic, M.; - Weskamp, N.; Schneider, G. Coloring molecules with ex-\nplainable artificial - intelligence for preclinical relevance assessment. Journal of Chem-\nical Information - and Modeling 2021, 61, 1083\u20131094.\n(68) Sanchez-Lengeling, B.; Wei, J.; - Lee, B.; Reif, E.; Wang, P. Y.; Qian, W. W.; Mc-\nCloskey, K.; Colwell, L.; - Wiltschko, A. Evaluating Attribution for Graph Neural\nNetworks. Proceedings - of the 34th International Conference on Neural Information\nProcessing Systems. - Red Hook, NY, USA, 2020.\n31\n(69) Ying, R.; Bourgeois, D.; You, J.; Zitnik, - M.; Leskovec, J. GNNExplainer: Generating\nExplanations for Graph Neural Networks. - Advances in neural information processing\nsystems 2019, 32, 9240\u20139251.\n(70) - Rao, J.; Zheng, S.; Yang, Y. Quantitative Evaluation of Explainable Graph Neural\nNetworks - for Molecular Property Prediction. arXiv preprint arXiv:2107.04119 2021,\n(71) - Yuan, H.; Yu, H.; Wang, J.; Li, K.; Ji, S. On Explainability of Graph Neural - Net-\nworks via Subgraph Explorations. Proceedings of the 38th International - Conference\non Machine Learning. 2021; pp 12241\u201312252.\n(72) Shrikumar, - A.; Greenside, P.; Kundaje, A. Learning Important Features Through\nPropagating - Activation Differences. 2017,\n(73) Montavon, G.; Binder, A.; Lapuschkin, S.; - Samek, W.; M\u00a8uller, K. R. Layer-Wise\nRelevance Propagation: An Overview. - Lecture Notes in Computer Science (including\nsubseries Lecture Notes in Artificial - Intelligence and Lecture Notes in Bioinformatics)\n2019, 11700 LNCS, 193\u2013209.\n(74) - Baldassarre, F.; Azizpour, H. Explainability Techniques for Graph Convolutional - Net-\nworks. 2019; https://arxiv.org/abs/1905.13686.\n(75) Hochuli, J.; Helbling, - A.; Skaist, T.; Ragoza, M.; Koes, D. R. Visualizing convolutional\nneural network - protein-ligand scoring. Journal of Molecular Graphics and Modelling\n2018, 84, - 96\u2013108.\n(76) Rodr\u00b4\u0131guez-P\u00b4erez, R.; Bajorath, J. Interpretation - of Compound Activity Predictions\nfrom Complex Machine Learning Models Using - Local Approximations and Shapley\nValues. Journal of Medicinal Chemistry 2020, - 63, 8761\u20138777, PMID: 31512867.\n(77) Wojtuch, A.; Jankowski, R.; Podlewska, - S. How can SHAP values help to shape\n32\nmetabolic stability of chemical compounds? - Journal of Cheminformatics 2021, 13,\n1\u201320.\n(78) Mastropietro, A.; Pasculli, - G.; Feldmann, C.; Rodr\u00b4\u0131guez-P\u00b4erez, R.; Bajorath, J. Edge-\nSHAPer: - Bond-Centric Shapley Value-Based Explanation Method for Graph Neural\nNetworks. - iScience 2022, 25, 105043.\n(79) White, A. D. Deep learning for molecules and - materials. Living Journal of Computa-\ntional Molecular Science 2022, 3.\n(80) - \u02d8Strumbelj, E.; Kononenko, I. Explaining prediction models and individ", - "lecular Science 2022, 3.\n(80) \u02d8Strumbelj, E.; Kononenko, I. Explaining - prediction models and individual predictions\nwith feature contributions. Knowledge - and Information Systems 2014, 41, 647\u2013665.\n(81) Erhan, D.; Bengio, Y.; - Courville, A.; Vincent, P. Visualizing Higher-Layer Features of\na Deep Network. - Technical Report, Univerist\u00b4e de Montr\u00b4eal 2009,\n(82) Weber, J. K.; - Morrone, J. A.; Bagchi, S.; Pabon, J. D.; gu Kang, S.; Zhang, L.;\nCornell, - W. D. Simplified, interpretable graph convolutional neural networks for small\nmolecule - activity prediction. Journal of Computer-Aided Molecular Design 2022, 36,\n391\u2013404.\n(83) - Riniker, S.; Landrum, G. A. Similarity maps - A visualization strategy for molecular\nfingerprints - and machine-learning methods. Journal of Cheminformatics 2013, 5, 1\u20137.\n(84) - Humer, C.; Heberle, H.; Montanari, F.; Wolf, T.; Huber, F.; Henderson, R.; Hein-\nrich, - J.; Streit, M. ChemInformatics Model Explorer (CIME): exploratory analysis of\nchemical - model explanations. Journal of Cheminformatics 2022, 14, 1\u201314.\n(85) McGrath, - T.; Kapishnikov, A.; Toma\u02c7sev, N.; Pearce, A.; Wattenberg, M.; Hass-\nabis, - D.; Kim, B.; Paquet, U.; Kramnik, V. Acquisition of chess knowledge in Al-\nphaZero. - Proceedings of the National Academy of Sciences 2022, 119, e2206625119.\n33\n(86) - Bajusz, D.; R\u00b4acz, A.; H\u00b4eberger, K. Why is Tanimoto index an appropriate - choice for\nfingerprint-based similarity calculations? Journal of Cheminformatics - 2015, 7, 1\u201313.\n(87) Huang, Q.; Yamada, M.; Tian, Y.; Singh, D.; Yin, D.; - Chang, Y. GraphLIME:\nLocal Interpretable Model Explanations for Graph Neural - Networks. CoRR 2020,\nabs/2001.06216.\n(88) Sokol, K.; Flach, P. A. LIMEtree: - Interactively Customisable Explanations Based on\nLocal Surrogate Multi-output - Regression Trees. CoRR 2020, abs/2005.01427.\n(89) Whitmore, L. S.; George, - A.; Hudson, C. M. Mapping chemical performance on molec-\nular structures using - locally interpretable explanations. 2016; https://arxiv.org/\nabs/1611.07443.\n(90) - Mehdi, S.; Tiwary, P. Thermodynamics of Interpretation. 2022,\n(91) H\u00a8ofler, - M. Causal inference based on counterfactuals. BMC Medical Research Method-\nology - 2005, 5, 1\u201312.\n(92) Woodward, J.; Hitchcock, C. Explanatory Generalizations, - Part I: A Counterfactual\nAccount. No\u02c6us 2003, 37, 1\u201324.\n(93) Frisch, - M. F. Theories, models, and explanation; University of California, Berkeley,\n1998.\n(94) - Reutlinger, A. Is There A Monist Theory of Causal and Non-Causal Explanations?\nThe - Counterfactual Theory of Scientific Explanation. Philosophy of Science 2016, - 83,\n733\u2013745.\n(95) Lewis, D. Causation. The journal of philosophy 1974, - 70, 556\u2013567.\n(96) Tanimoto, T. T. Elementary mathematical theory of classification - and prediction.\nInternal IBM Technical Report 1958,\n34\n(97) Rogers, D.; Hahn, - M. Extended-Connectivity Fingerprints. Journal of Chemical In-\nformation and - Modeling 2010, 50, 742\u2013754, PMID: 20426451.\n(98) Mohapatra, S.; An, J.; - G\u00b4omez-Bombarelli, R. Chemistry-informed macromolecule\ngraph representation - for simila", ", S.; An, J.; G\u00b4omez-Bombarelli, R. Chemistry-informed macromolecule\ngraph - representation for similarity computation, unsupervised and supervised learn-\ning. - Machine Learning: Science and Technology 2022, 3, 015028.\n(99) Doshi-Velez, - F.; Kortz, M.; Budish, R.; Bavitz, C.; Gershman, S.; O\u2019Brien, D.;\nScott, - K.; Schieber, S.; Waldo, J.; Weinberger, D.; Weller, A.; Wood, A. Account-\nability - of AI Under the Law: The Role of Explanation. SSRN Electronic Journal\n2017,\n(100) - Wachter, S.; Mittelstadt, B.; Russell, C. Counterfactual explanations without - opening\nthe black box: Automated decisions and the GDPR. Harv. JL & Tech. 2017, - 31, 841.\n(101) Jim\u00b4enez-Luna, J.; Grisoni, F.; Schneider, G. Drug discovery - with explainable artificial\nintelligence. Nature Machine Intelligence 2020 - 2:10 2020, 2, 573\u2013584.\n(102) Fu, T.; Gao, W.; Xiao, C.; Yasonik, J.; Coley, - C. W.; Sun, J. Differentiable Scaffold-\ning Tree for Molecule Optimization. - International Conference on Learning Represen-\ntations. 2022.\n(103) Shen, - C.; Krenn, M.; Eppel, S.; Aspuru-Guzik, A. Deep molecular dreaming: inverse\nmachine - learning for de-novo molecular design and interpretability with surjective\nrepresentations. - Machine Learning: Science and Technology 2021, 2, 03LT02.\n(104) Lucic,\nA.;\nter\nHoeve,\nM.;\nTolomei,\nG.;\nRijke,\nM.;\nSilvestri,\nF.\nCF-\nGNNExplainer:\nCounterfactual - Explanations for Graph Neural Networks. arXiv\npreprint arXiv:2102.03322 2021,\n(105) - Numeroso, D.; Bacciu, D. Explaining Deep Graph Networks with Molecular Counter-\nfactuals. - arXiv preprint arXiv:2011.05134 2020,\n35\n(106) Krenn, M.; H\u00a8ase, F.; - Nigam, A.; Friederich, P.; Aspuru-Guzik, A. Self-Referencing\nEmbedded Strings - (SELFIES): A 100% robust molecular string representation. Ma-\nchine Learning: - Science and Technology 2020, 1, 045024.\n(107) Nigam, A.; Pollice, R.; Krenn, - M.; dos Passos Gomes, G.; Aspuru-Guzik, A. Beyond\ngenerative models: superfast - traversal, optimization, novelty, exploration and discov-\nery (STONED) algorithm - for molecules using SELFIES. Chemical science 2021, 12,\n7079\u20137090.\n(108) - Kim, S.; Chen, J.; Cheng, T.; Gindulyte, A.; He, J.; He, S.; Li, Q.; Shoemaker, - B. A.;\nThiessen, P. A.; Yu, B.; Zaslavsky, L.; Zhang, J.; Bolton, E. E. PubChem - in 2021:\nnew data content and improved web interfaces. Nucleic Acids Research - 2020, 49,\nD1388\u2013D1395.\n(109) Tolomei, G.; Silvestri, F.; Haines, A.; - Lalmas, M. Interpretable predictions of tree-\nbased ensembles via actionable - feature tweaking. Proceedings of the 23rd ACM\nSIGKDD international conference - on knowledge discovery and data mining. 2017;\npp 465\u2013474.\n(110) Freiesleben, - T. The intriguing relation between counterfactual explanations and ad-\nversarial - examples. Minds and Machines 2022, 32, 77\u2013109.\n(111) Grabocka, J.; Schilling, - N.; Wistuba, M.; Schmidt-Thieme, L. Learning time-series\nshapelets. Proceedings - of the 20th ACM SIGKDD international conference on Knowl-\nedge discovery and - data mining. 2014; pp 392\u2013401.\n(112) Kenny, P. W.; Sadowski, J. Structure - modification in ch", "ery and data mining. 2014; pp 392\u2013401.\n(112) Kenny, - P. W.; Sadowski, J. Structure modification in chemical databases. Chemoin-\nformatics - in drug discovery 2005, 271\u2013285.\n(113) Tyrchan, C.; Evertsson, E. Matched - Molecular Pair Analysis in Short: Algorithms,\nApplications and Limitations. - Computational and Structural Biotechnology Journal\n2017, 15, 86\u201390.\n36\n(114) - Griffen, E.; Leach, A. G.; Robb, G. R.; Warner, D. J. Matched Molecular Pairs - as\na Medicinal Chemistry Tool. Journal of Medicinal Chemistry 2011, 54, 7739\u20137750,\nPMID: - 21936582.\n(115) He, J.; Nittinger, E.; Tyrchan, C.; Czechtizky, W.; Patronov, - A.; Bjerrum, E. J.;\nEngkvist, O. Transformer-based molecular optimization beyond - matched molecular\npairs. Journal of cheminformatics 2022, 14, 1\u201314.\n(116) - Park, J.; Sung, G.; Lee, S.; Kang, S.; Park, C. ACGCN: Graph Convolutional Networks\nfor - Activity Cliff Prediction between Matched Molecular Pairs. Journal of Chemical\nInformation - and Modeling 2022,\n(117) Langdon, S. R.; Ertl, P.; Brown, N. Bioisosteric Replacement - and Scaffold Hopping\nin Lead Generation and Optimization. Molecular Informatics - 2010, 29, 366\u2013385.\n(118) Turk, S.; Merget, B.; Rippmann, F.; Fulle, S. - Coupling Matched Molecular Pairs\nwith Machine Learning for Virtual Compound - Optimization. Journal of Chemical\nInformation and Modeling 2017, 57, 3079\u20133085, - PMID: 29131617.\n(119) van Tilborg, D.; Alenicheva, A.; Grisoni, F. Exposing - the limitations of molecular\nmachine learning with activity cliffs. 2022,\n(120) - Fischer, H.; Gottschlich, R.; Seelig, A. Blood-brain barrier permeation: molecular\nparameters - governing passive diffusion. The Journal of membrane biology 1998, 165,\n201\u2013211.\n(121) - Liu, L.; Zhang, L.; Feng, H.; Li, S.; Liu, M.; Zhao, J.; Liu, H. Prediction - of the\nBlood\u2013Brain Barrier (BBB) Permeability of Chemicals Based on Machine-Learning\nand - Ensemble Methods. Chemical Research in Toxicology 2021, 34, 1456\u20131467, - PMID:\n34047182.\n37\n(122) Wu, Z.; Ramsundar, B.; Feinberg, E. N.; Gomes, J.; - Geniesse, C.; Pappu, A. S.;\nLeswing, K.; Pande, V. MoleculeNet: a benchmark - for molecular machine learning.\nChemical science 2018, 9, 513\u2013530.\n(123) - Ho, T. K. Random decision forests. Proceedings of 3rd international conference - on\ndocument analysis and recognition. 1995; pp 278\u2013282.\n(124) Martins, - I. F.; Teixeira, A. L.; Pinheiro, L.; Falcao, A. O. A Bayesian approach to in\nsilico - blood-brain barrier penetration modeling. Journal of chemical information and\nmodeling - 2012, 52, 1686\u20131697.\n(125) Pedregosa, F. et al. Scikit-learn: Machine - Learning in Python. Journal of Machine\nLearning Research 2011, 12, 2825\u20132830.\n(126) - Moriwaki, H.; Tian, Y.-S.; Kawashita, N.; Takagi, T. Mordred: a molecular descriptor\ncalculator. - Journal of cheminformatics 2018, 10, 1\u201314.\n(127) Chollet, F., et al. Keras. - https://keras.io, 2015.\n(128) Wager, T. T.; Chandrasekaran, R. Y.; Hou, X.; - Troutman, M. D.; Verhoest, P. R.; Vil-\nlalobos, A.; Will, Y. Defining Desirable - Central Nervous System Drug Space through\nthe Alignment of Mo", "lobos, A.; - Will, Y. Defining Desirable Central Nervous System Drug Space through\nthe Alignment - of Molecular Properties, in Vitro ADME, and Safety Attributes. ACS\nChemical - Neuroscience 2010, 1, 420\u2013434.\n(129) Ghose, A. K.; Herbertz, T.; Hudkins, - R. L.; Dorsey, B. D.; Mallamo, J. P. Knowledge-\nBased, Central Nervous System - (CNS) Lead Selection and Lead Optimization for CNS\nDrug Discovery. ACS Chemical - Neuroscience 2012, 3, 50\u201368.\n(130) Polishchuk, P.; Tinkov, O.; Khristova, - T.; Ognichenko, L.; Kosinskaya, A.; Varnek, A.;\nKuz\u2019min, V. Structural - and Physico-Chemical Interpretation (SPCI) of QSAR Mod-\nels and Its Comparison - with Matched Molecular Pair Analysis. Journal of Chemical\nInformation and Modeling - 2016, 56, 1455\u20131469.\n38\n(131) Hassan, M.; Brown, R. D.; Varma-O\u2019Brien, - S.; Rogers, D. Cheminformatics analysis\nand learning in a data pipelining environment. - Molecular diversity 2006, 10, 283\u2013299.\n(132) Schomburg, K.; Ehrlich, H. - C.; Stierand, K.; Rarey, M. From structure diagrams to\nvisual chemical patterns. - Journal of Chemical Information and Modeling 2010, 50,\n1529\u20131535.\n(133) - Sheikholeslamzadeh, E.; Rohani, S. Solubility prediction of pharmaceutical and - chem-\nical compounds in pure and mixed solvents using predictive models. Industrial - &\nengineering chemistry research 2012, 51, 464\u2013473.\n(134) Boobier, S.; - Hose, D. R.; Blacker, A. J.; Nguyen, B. N. Machine learning with physic-\nochemical - relationships: solubility prediction in organic solvents and water. Nature\nCommunications - 2020 11:1 2020, 11, 1\u201310.\n(135) Loschen, C.; Klamt, A. Solubility prediction, - solvate and cocrystal screening as tools\nfor rational crystal engineering. - Journal of Pharmacy and Pharmacology 2015, 67,\n803\u2013811.\n(136) Diorazio, - L. J.; Hose, D. R.; Adlington, N. K. Toward a more holistic framework for\nsolvent - selection. Organic Process Research & Development 2016, 20, 760\u2013773.\n(137) - Sorkun, M. C.; Khetan, A.; Er, S. AqSolDB, a curated reference set of aqueous - sol-\nubility and 2D descriptors for a diverse set of compounds. Scientific - data 2019, 6,\n1\u20138.\n(138) Durant, J. L.; Leland, B. A.; Henry, D. R.; - Nourse, J. G. Reoptimization of MDL\nkeys for use in drug discovery. Journal - of chemical information and computer sciences\n2002, 42, 1273\u20131280.\n(139) - National Center for Biotechnology Information, PubChem Compound Summary for\n39\nCID - 1549018, Jasmone. https://pubchem.ncbi.nlm.nih.gov/compound/Jasmone,\nAccessed - September 26, 2022.\n(140) Sell, C. S. On the unpredictability of odor. Angewandte - Chemie International Edition\n2006, 45, 6254\u20136261.\n(141) Genva, M.; Kenne - Kemene, T.; Deleu, M.; Lins, L.; Fauconnier, M.-L. Is It Possible\nto Predict - the Odor of a Molecule on the Basis of its Structure? International journal\nof - molecular sciences 2019, 20, 3018.\n(142) Rowe, D. Aroma chemicals for savory - flavors. Perfumer and Flavorist 1998, 23, 9\u201318.\n(143) Mallia, S.; Escher, - F.; Schlichtherle-Cerny, H. Aroma-active compounds of butter: a\nreview. European - Food Research and Technology 2008, 226, 315", " H. Aroma-active compounds of - butter: a\nreview. European Food Research and Technology 2008, 226, 315\u2013325.\n(144) - Jelen, H.; Gracka, A. Characterization of aroma compounds: Structure, physico-\nchemical - and sensory properties. Flavour: From food to perception 2016, 126\u2013153.\n(145) - Licon, C. C.;\nBosc, G.;\nSabri, M.;\nMantel, M.;\nFournel, A.;\nBushdid, C.;\nGolebiowski, - J.; Robardet, C.; Plantevit, M.; Kaytoue, M., et al. Chemical features\nmining - provides new descriptive structure-odor relationships. PLoS computational bi-\nology - 2019, 15, e1006945.\n(146) Mostafa, S.; Wang, Y.; Zeng, W.; Jin, B. Floral Scents - and Fruit Aromas: Functions,\nCompositions, Biosynthesis, and Regulation. Frontiers - in plant science 2022, 13.\n(147) Tokitomo, Y.; Steinhaus, M.; B\u00a8uttner, - A.; Schieberle, P. Odor-active constituents in\nfresh pineapple (Ananas comosus - [L.] Merr.) by quantitative and sensory evaluation.\nBioscience, Biotechnology, - and Biochemistry 2005, 69, 1323\u20131330.\n(148) Wei, C.-B.; Liu, S.-H.; Liu, - Y.-G.; Lv, L.-L.; Yang, W.-X.; Sun, G.-M. Characteristic\naroma compounds from - different pineapple parts. Molecules 2011, 16, 5104\u20135112.\n40\n(149) Stumpfe, - D.; Bajorath, J. Exploring Activity Cliffs in Medicinal Chemistry. Journal\nof - Medicinal Chemistry 2012, 55, 2932\u20132942, PMID: 22236250.\n41\n"], "model": - "text-embedding-3-small", "encoding_format": "base64"}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '26417' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/embeddings - response: - body: - string: !!binary | - H4sIAAAAAAAAA1RZyw6qzJqd91Ps/FM6kZtU1ZlxExGQQkDEHgGigiJyqQKqX76D+6Q7PXGARFOw - vnX7/vs//vz5p83rshj/+deff97VMP7zn+u1WzZm//zrz3/9x58/f/789+/z/91ZNnl5u1Wfx+/2 - 35fV51bO//zrD/+/V/7vpn/9+UfuU5n6X95gjNyhBU/PEWG7lhLGbwKmw2NyD6k3zHLNZL324eHj - 7XF52mFD3B2lBdVOq1JX7oJ+sREiYFHTB3ZQAKI5dIwS5Kyz8XE8sHo+uF4DLrEzYFu8dIAJ+PRA - cFBe/qJXCZhvEVfAa8YlPlfsPsZLFoYY8Edzxpf7u8oZ3CUJPI9njQY3wwXdzYggGkx/8fnFV3Mh - GkEHorT+YO8RiWy0QW2h0FdmamhfkbX6uwthn+IQ273zctle/ihIZW5NgGeYNX+/miF6HgZEJPUz - gIVoVwLGj7OjybnsamYphwoabzhi4/EZ+hYesgYKW2bQQ/7SchFxZQz28zHFRXJxXf7SqyVisrGj - 5hsdI8nfqimUn6cH9njzzZbZXmwEsIqwUSdztMSalsJPQhvq9n7LWM75JoLzR6KafS2AyFV+A6L5 - 3mPs+l+39XO/glu/iOl9T0Z3+S7BAswjzcnsENOVXsjjYPo1MQ2efAXo5rYVYV0+LljjBM8Qmenx - cEPPMTbiQ8fmnX0qkKcOPC4TIwbSW753MM+Iif1D+oqWzxlDCO+UEglqustjjiPQYPCGjwN8gc7o - vw6qo5fps1d8yYUX3Znodx694pJcCMvMR7cGayRzPsilnb14ENV3jnT6AeQsx5qH0u/WoZ5Q0Xp+ - WZ8G9cZDxWXER0B6taIPH+Hs0JhtUyapyVaGK7586Tpt3CXZ2gnYZK1Ob9/CZvyg2jZ8Tu1Mg+x5 - 7ifz9g3g27IRxTGyXAmOgIceJBG9grGvacyWCsjv6kSdrH/mM+7VB9L3gYv1hb6jxcSnDKoepj5K - qOWK3D4tkbCXoQ83ZW4IyB09aDuHnJbLnY/YJ4UeNE/cEesK1vIFNUCB8nBk2NJG0LPbXvURvb/f - NDOO+349jwK5uHaotvdbl2SAFXBDqitVN2Qw+lN+0WEo8G8aVlwSTcn8EZH+8DE1rvEm745fK4PD - 5If0mH3v9cDP9wYGl9Aj8i3fR6zs5BRedrGGT8trayzXIVPhHro91ud+a7DOXnz4m+dbnYn11LuF - CInAf3z51ooutU1hgFV7veN9aMR5G1QNBzS6QT5ohxCMY/J8oMGPHXosceWKIGscODSAx0drbvO5 - CnQPfZN7TODzUruCFAATpC7IKN6To7GkBv+Atw3a4DibrF6gJ95D11rJsKdEVs/fp4agRpIB3huH - ijE1mRX0TBUda7bnu6LXtSoyOyXA2nJxckFJowq5qL9gY8Ujf94HAXJbuMPW9xC56/vxgUojF+/L - +BPNMMgqqCeWiHcqMFx+oEcV0rb+kjYanFxqtykBanKxCdAMr+b327SE67zho6nXPdMdt4EIcyq9 - p9KmX+Jz4KAtCCm25+vbYHlTd4h9vQzrk/rol6aTE6iMtohPOzQac6B8OQim7Q7vg3dX00solIgt - myv2wU0xiBodW8h/gYddzRh66YK/C3wq0wEflV6rJbB8HWS3B9d/Lq+r++H8IIMX2TBwsodFtMRp - zwN41Y/Y0/J9z2tjHaLZ6TR61uwZLLvjZoI7W1ZpwdRrNBVfK0Rlv7njw76p6zHw5QnBmBUEniOO - DZfeLuA9BBG2WBqB7poaIRC3rUKxFdTRX3yV6c6m4dxfXZqRV4ucEgo0zb0mvx/TWkHXhe6omm82 - eZfSkUCfnjyaAS2L+HxXWtASr0e8FzYYCMfKKtEVijd6KKpPNF63DxE2361LL/yzcecx+Vbwk8EL - Dr1TxKZDmmdAZFNEPbv1I7LkNIDpokZ0vzdgPhkPyMG6Sin1b4lu8BFXOnDTRS02UJABQdROHUTZ - U6HXPQujxfe4FuA8HMh8dCKwDL6tQrR8DXw/3RxDjM+BDbe3YEOTDD97grgkgR/af6lqGDmYr+/M - giMWJHo4CmY/nFzvoaSWzqglfGI2V59Xi/iX/sYWf3CNucOaidhpH658iPuhNfgU3S/chmytc9sz - /dnoUDh9AY0xuuUCazkLxuPdIfPlGvcC9R8PFI83h4At6Ou5tg8e/OmbLAlxP4/E5WD/iHPqhZ26 - Pu/EhMn9aNPjzT0DRqSyAttD2tJdp0cu39hfgjoFV0S+gj5nJ/Uhw7frf6naLHa9sDR6IUZ3GlYV - dHYFqT1X8GTPW39pyiXqOPGWwPtk2jgjyRfQ6HU14cUzasKOyoEJ2UGP0dZ4aPT6Vb8GqTlLh0av - XqkuFpM7IXMMYNDbNS5L3mWiWDxi9DioLo42b93o/EPHQVp9dRpeOC4fFORN0LmCA/Zlcs6l3XuZ - 0N7MEI1tw3SXSzG9UJGcHVxchBjweitw0InVit5cWDCpmjY+YLnfkcU7McAuH88Dqx5SZ9Uvtos3 - BLT8+0ivMteAz+bQWzCRkhP1mLqNiCNAAvWtHeJcGrp+IPj5Qrbj5tjgnmfA4K5MoNC9OXwLXQSG - wch5QHyFYYPF1JiRfa3A++WM9KhW+5p3odvA930LyMa6ha5IUt6BV/RNscepc0SkQZtgk5sPehEW - zRAvrlCh6MnvaKzWrTG/6NGEZ/G0owEJX/UyBcwGt214I7M5LO7ilFkI1eRs+yB9WpEY3U42RDux - XfmSuYPx4CF0FuHoc8s9zpmzfXBo6Q4lvs7zw2VwBCKQOmLhw84c+3mjBhV4PDlMDzdWRbPHJT4k - g1LTg9Bfe3ZK3RhKxZDQ6/R4u+wV0wKiK9mQdsXXHJahD0VPvFOsXPp8TuTPBKtP81r9qQjo/jZP - aBTgRDbKxY2G4q28oKuqOs6CQq2XBMY6oqliYH+7POtl2qYtOBMXUn1nju7MTlGHWo8vaaxrM5hT - 7SmDdT4x9q6IzaueAf1oLzQXHL4nU3t7QWwcXmQafTkf1nlBYsdjap6XB5uEbvKQY+kL1oD9jMjF - /orwVaY+2a74mL26tcBejExsLAZvzIo6W4jf+44vZOY5X8hUNtBxzRL7jbbUU8QlDpSNzsF4D3Kw - zDdbAcJZHlY+WqI5uW1baF2aitrpTfvx8wKcq+X7jTaCeh67KYD6Rt8SyJs7RjWEFhAJQ4FvB/UQ - CdF1UGCryS7O1LIC7GOOJSwjrcAO5aN+3ke7BqL4G9HgMSW16AuFAwfs7vz5aZ/qCU1lATSvVKkT - pbwxvWEkIzOcbRo0m0suTMKQwVsfjtgXw3c9QY8E0KuXgB41VvXT+bhp4eoXfN64g34+K4cMbMN4 - oG7hzy6bP/EDXFWeozhpufq7JCqE3+IC/NeSU4OwNG9g+lJr8mmDyGXVztcVQ9+e6D40+HzpTzmB - tmgDfDEOFZgvVy/evjx3pPbG1sDMtrMDH2L/xD+/RJznyP3wS03RvoJFqPcEyM3Z83mMUNRd+FyF - c93syBxLXD0Ls2xDyxhcWnzNkc1c7reKZN/f1FMdr5b4/hlCONI79qV61fvk2cA6akxq0I+Zj0fx - 7AEEp5zeDkfZJaNHMpifu5RqGz1jkznLOhiA8/Y5ky/cxUuuAYzAx/C3XbkAktI3ARF4GzTUTjmb - a7eQFfMEjzS0Zwmwju9TKDC5wCeHvAwaFFOLuB25Y1fMcS/276qA7iwfKD4g35jNet8Ao5lbWvbh - WC8NMVWQyqPgf+x9yxi/kXUYH16MCLmt9dLDbzulSC6OL20HpV+IdhqQd54eOPzQMSIW0xpU9vbL - lwULg/Hdzyn8svaw5s0asGqSPCA/yrf/wR+zXlD98BFXD5g+bY+4My9fHvAcZ5yPNFbVC419ESqH - 4E5A4Z/cqbGfA5qXdMJn/eoBMb/XBHV8DKi3xVk+aYnKo/DCCzT3l0O+jA9kw5UvsPbetj0xw9GC - rgPP9Lozj/XEz5cGJaMvY+v9vRiTvt3qkHx4iYbuo6gnXw0yQK6qR41Z3OWDBV8OWP2pD+AF5H/9 - QOxcCfnlWZGH7gTOZZWQyUkOxtKNfQmL/eNOy3Ue/s6n2ImYGodDCAhXWQ18SXGIj7uxcOfLmJsg - Oy9PQosuYKyFQwZnt4D0tNEV1tXKkgDh1AMiGcmpZ6v/gyu+scYzxR3naWNCLBg9ocSSo9nJGhUS - sa+oo5YVm5qH58BZrb7+6mf72b2ECci7S0em+/vJpu1uJ0Jd9+744tdpxAL7K8Nuc/J9bta6fhr6 - OUG7k2gSMcNaPVuWJMKfXh2yyMr5MBs5eL6WHbVX/zzVk+XANuP9lU/EFS/XAeTbivpCXm3caZcy - CC53c0P9iGeA8f03RHeBo1QPzJ0h7O5MBt2jeOGSTTRfbqoaoHbbbnF8ILAme722UOzkxJfuy8md - lrAYoJYHPVbPydNl3NevYLNI2JfVzR6w+wnGUKMI4WNCLYM5J7eUf/4GyYkSEc31OiC9u4yq56ms - J2S+A6DSk4v3xxBFoxQAC8aHhlGNCbE7v/ttCle809A6t+7CB1kDri20COrGGiwHXxahoOtn7I6d - 3C9JuwuQzoMn+fUd85avQyTsLQ3bzfXokvEhOEBLvA3pPcPsicpx/N9+Ye/Xaf5XPyJb0db50OsX - c6IBjjPd4D3mWTRag9ZBZfIjqi2vqzEXmpbAEQgXarzTup6e70yHZk9d7HNIdJcPepcQn9yY0FTa - 1Eu3O3JASXcOPjpl5C7HU88B9fw44pt8cOtVr0v4m1e3Jh7jq8dA4Ewe6V895BfuqEDP7fdYc8AC - 2jW/wTUvEy11esayPoBouNmCD5XIqtf3Rf6t721LDOZA2IGfP/v57+muPXn07owYa7oD2HQ+Sh0M - tk5EFCynrlCYZxUdrv7Fnx71BzA/qGSwPxRfHBhGzuYHHTtx45aIWi/3GM1H56WD4/uEsB/Iar7y - bwrN55v5G8eBgNRN/4IHNYyxW057V/LbY7wtNtGRbGmR1nT1g9DYqgcCOfWUi2UoZPDzGhOs+e2t - H58bbCnGnkP0IJQdI+fNhwDhoe8wVrWqHqnfVrA4nUrSbL4WY+VVSGDw8ALCZ0iI6IpXEGztiHyP - wqtfpAvQ4ZgyQP7yoQpqBxTGiazPfwDzl7weyD/mO3/tr9h8a4/eX3/gx+/emK7LQ0bnobKI+OO3 - SsYO/PHdry9b7rVcQhjBJ1Wzp9BPevzxgb7dHLF5CZxe2jZRBkbT+mINb+V6cF9fG9bXosdFV0n1 - dBgq59dfYHN2eGPUbuqEkl3h+C/Ja1y2eysLeElJiPdn8d3Pt0gs0JpXRsVMi+gWfIkMOavO8P43 - vzG+BrC+n0OsIi8xGKdoDRxHdcBHQXFq6RV/SrjHl4Bshtbv6RCRRBEyLqV4513cpVaUBD4OkoX1 - Y+Uz8fheCLpK00wkfSfW34fBl0j9Zi+fT/2pHjz3baLTxTEwnps4mnEyT9BoWOv/8hoPtlsLYkHr - f30p4O/nqYFxdFl88Yen33wl1+T5tz8YUMqrCEmEUucWb9iwcEcZ9I8kp9bVPUfU2T4gnNzPzq/5 - 8OySU3wxgb4Pf33TLl9i82ZCUad7fAzdG2Ppvl1gNN96ssmv5/6nnyiFsMHh5aUCMR6UBt5vUkPk - 3LPyhXNeNmq+wCXtyeGjRWeVgy67RMNuoemRNEQkhuXDO+FI0q6MSewwwLsAKd1jPsoX4xKmP7z8 - zavNr59Z/SO9of0czZjEPsrte43397ceTT//DYCQ+sLzotdCMSgVVIDd0CyTJ2O4lpkDH4oSktnP - fSCa+JpBxdln2OwAMwaxcRQ4O61Gw4euA/GjnCb0F9/+coh6fiOrMDzoDt3fN3c2Jdkxg6vf9hkK - FPZGUpkC5MgxtoIj7el7noLfPGHtAXQgxZVUKFhPvhjfQM/Ys9StX/9F1dao2XIp5Bd0a/lK76C1 - 8sWwLjEcmi1PvVCixpzc5g791c8o5d15KjUOfrlXgQ8XrmQTnGUT2KMn4d/zWbTqYwFw2Jzo2nez - 5RB6FVz9IQGKnUUs00Gx/eHloBRTP6iTCOGp1zPq3c8BE1f9gb++6XnoJDDv1hY4VJQTxuP20M/i - qPMoMvgzvauaXrPXdhvC1X//zSeMZY0HtYuh+JvJ7mphGesMvCX/QF0evRmjuTXBzZLOOJVZ1s+T - 8EqBh2qf+okRszUPKWh7yVKsfZy6pw/lKoJImWx6uSbPfvLpOQC3jwF90N5fLpOTLQSppTJqv+3F - mIPQLH59A/3hczrtZV5Z3//a1xGX3sqqhTPncj++rNnZiHVocMpE90ZyqqdFWXz0OSs9qadla1BN - 6Uz41RZ91ReV0dd2GwDB41rqTPLDWLJAD3/+ZtWTph7k8qDALP/a1JefTb/2by1ckK9Q55u+XHKv - 5b94IO+FOa54eZghXLRjTg1keIb0m4cVv1i3X193UUQkw/jzqumBw0/GiH1t4QTbEz33EXFn6X3g - wc9fG9d9UE9rXwKf/vLF+O41jJ7UVoaBvdvjfbe5sPleTBXaFqWJXS9gEXN51im/38PVUIF55SfY - R25DUIQdd9hvgwKu/gZbB7UF0/jxdPhi+yeBp4Z3pxPrVNj7RYRPv3wCg+wBTOUpUlsPKjB74VsF - 024ocbQrbIMVTzFWgMgGsiA4uNOgqg4YRGtLnTIdDCoAnaDjO0LU51BizKMeTvAJrxt/IqFZz4V2 - iOFGu7zJrDv5ry96wFNoXbDp31m/vPaqg359WxJkZTRtr/FrS4cpo9a5dPrJkC8ZbLnhjTPTHWvy - 80f6yXD8RxzybKnDQoTz/RxhfHZixrQmT6A1Vc2/9X9RFg9WQxus/b7QL42eiSCJAwH79TLXi0Rv - JXjL0sYXHtb+3/70KQw3uoP0DAbwrmL4BB/Th/fqXi94UQuUB6/Bv9+Ay3jvzkQIBG72e3/5RjNB - 7wyWjSlTZ/eV8hnKkw7X/IvdNY+QkiMvMGJJwuourfNpHywcCnqnJkIfEWMA7y5WVv9DjVTR2Jw9 - hgnuBJL/zffzz0/KfnPDvh2n7t+8JfD0gr1fvpynjQWDo8Pw7tiwfmqKqfybV/YqD+pRmCcbrucl - G8iA+9unKD8+vVfbuV7Kj+nAdV+AXR7t2MJMT4RrvsRGIx6j5cFOMtr1bx3/8ghbXHOBz2G60ogk - 33xxrwOEzdcMqN4XsjHYjziDzSJg7Eh7FQiWmpbKfnjsVvyV9WKVCg/OUWP4m6Qt+1FMeVOhT35P - L8cd6ad1Pwb3VTZi39pd3CWScQhdi76IkJ+naL4/TA/WatVS89vJxjLzfQLmJZuIIOAtm4/s8ADa - MQD/3l+t+0UYGrCka9/uLqeTkQEHcu+f3uWiwjoP/vY5Hng966XTlhAV9jnD6dqfSGXy7eBP//3B - eLEve3cZWP0i3okPHjC/PSY/PfDj1vNzMTjLROkv6eW3/6znY+WXv30CdT4Hw5jna9zCaugCIurz - B8zTkg6/fEh3jiXV07o/gbtiuVArK4Kcj4FeoNQ179i6Hz/RiEnhwYDfVjjj3w4TUf3w4DVQDeoP - AesXBJwOivuF4Z17Ceu//IfUCeJVr9z+aJ4d5Mqf3t9Oj53Bs+Mnhc+pm8njEEK2nNqRhyebbYmC - +28+ftiVh72gjdhe/QMN+MiEalA+fZpK936W4BDC/p3ucdSIY0Rvy5yib/9Uf/0p+15PRozWvEOt - 7Ltxh/M+DRTjAbS/+7FFBrUPc5onpLq8HmwWR4eXf35HffI6WPszHz7FOsX+a5L6ie4fL7D263Tv - J4dakLZBg268N1PD+cCaAcdtYa43kKoncWdI3Nd6wDUPY2wFRj6PnRzCSe0WetxmTT+zmMZwze84 - crIjmJPFzn7nw6Z8e0bzuk9A6/8Tya/laA6XB0RSX2u+/N1+jam/9w1YrNMehwvkDRZ+XhP8vS+n - UKpoXs8PJmdycKS6R3cSz3se7mecYnvt28XRazK47luwzh0URnNlSWHXnF1cFruPO6v2osPHIFf0 - fwAAAP//nJtL84KwFcW/S7d2RkAgYYmAyDsoiLgDRXmIvEwgmel37+C/3XXVJaPj4CX3nHN/Can1 - 1MFu9XOZa9Q33q79Owd58PrLD+/ZOORUz7AJf34egEfmTjHQCujjJsNykAU5z70UA27du0I0nS1s - 5XUFkNAJEfvY6vVXD7MC/ON3KuBf//w/ThTw//tEgfRoGDkP0bFeJL5+gUkuVLx8asbY1roXEGRP - SEwzb2qW2VICe0/MUOhkB31H2qRQCsl5EtX0e7CcDBYAlWuCYJztMOLLdrFgt9M+wYKffi2MWjDD - Ut1WKLjePEbP/SWDT3vjBVuHzfk0RWIMl4u+xfrlhRiNZO8Osw1U0IW9YjCLzs0D2r7eEkPIpZzW - d+0OtX25JW7zltkUglMBi0cmIo0etHyJjReG6uVeEz9/U8ZIKgRwFFUJXdDQMkpuUaH8Pt/r53e0 - 6Cc2KG8NSwHPAJ/TA+UqpZSthRTSUuiLQHwZCgdtCvg5EXSSFTaGVZcdiTkq8cjLWa2CD/5OKI24 - K+D9VkgUGqCAuNXpzqg1SjI4ia2OWWkcohk9HAsot+KK/IycGIudIgbvtJ2Ivh3yelo/VzrnEZH4 - mVU5tUMphVLLjSj/8jRa4p2aKabZZ8QD8OUuh2PlwaU4CXi74dEo4DtHlfkB9yQp98Ad9zvRhIGH - n2h/fKo152N3AroWGRjq8gvM22/QQSczGUKwPIFd7BQJhCPaoyOV9fFr5HIDT2KjE/WVbsESK3Ol - JO9xIRpJdMYPo0GVBNQNCdJdD/gXOkGl0rOeoDyrXL7H2FQW8H0ik0pBNG1qNEH+sO2Di530Lp3r - LIV5dc6Ia0BL55C1UAXXbor0zHJHThIsCnl8dQkK3q6+7O++AcOURPjtbwzAGwIzFSUxIpLUl5R9 - MbYN6JrdkYSHaweWfaV4srUsEgmkZeMyTpU1WeqqLynaas9ok/YWDKVaxacKiC5d9uWsrPUMFm1f - s++xCj0oytGV7DnnADjZ7zTFWpiEtDi46YK4bO8wteQ3yfnPV1/r84L9YzqioD2W+bx+H8rI81FO - xWDclTLSwPZj+yShhQ+WY96ZsNx6O+J3n0PEbjQzIdqMGC+Q7uvldacBfOg5JPqH1eOv35TI+chY - Tt4vhsM2aCHupRexpe1NnwVSO4qnjHfi6bLKxkQ1RJiLjo9p1D70xchpo+DdxUbpuU4iYcuVoSy9 - rZ7sU35bL/5ekGE3h0/kRP3HJapnvaD5ZgfkTBZw8WLdZLjbNypJvupJ54cm7uARCT3xvV0xUoi4 - BI6n6w2ZB6fWGVNGC2rcLscGhOdxx7Ebln797PKfr9sWTyGEdnDj0KMqp5ExAagQKnhG3sTcfPIe - 8aBU7XRAwatvataXc6Lo4UcLxOxycrnwFUKFnDIcLPr5kAubJSugYWY+lr+c4tJNctuAK96pWJ7n - BFB6szbwjrCG9E5QdV6OZBmeDuFn7Y/X+M0/hQWrb+HjJbcfOaftSwFOY3xBXt97ESeVYgZ1GtyJ - quXI5SYzNaEynA1ys+PY5THem1CtaRmIL7WP2CLHifIqd7uAZdY4Mnu8d8BRMoTlqP/otJrPAuTr - 2xn5qcrq7r7IiULbZkGXrsnzSY5TC77eCGDZMz5AOAv8oEwFvqEDFm+68O7DQAkelUd0u7yxWXzf - MNzO+QGFkucBahe3AS6NOyDzm59GIX66G1hwxo1c86vozv4namAucYg8JRrl9Or0pkLezROZqvIB - /eexVeFP783iq+hsMxwMoHJtgPyHFALibrMXWGbWEseTLbbI5rmD2vK8ksPa3+wtRJ4StZeS5Dvg - gHFzbwQlnYI9fncNyEezCGT4GS8dSfuxjISnoVCYqkWOtIshu8tH3FZQ+XYWKbwHi0juqxbkpMuX - pEXlRXP24AoovLkbSp/syxaEvA74m6wkjpSe62/lih0sAceIKi5eviBkDEp+T1sSX7tjvcs7a4Cn - qHmj5N0ngAuWc6XclrdCimdr6YukuwJ4Iu0ccKDgx+Xp6ZaS5lmM1lFLp+f+kcLx2SekqJmed23u - F/IoYRDw4Vljc+I/ZXiTdxwWDo6uTyg5q5Ads4L49vdZM3ehoVIFSRXwSavWHBf4Kgza2SCPMnT1 - xXQWVfnpo+pkb5d5j1KEFXEQyQXNH/kcm2elOldxAOTBzoVqPnOQ2JxInKu6j3Zp7Mzwennbgezu - x5weS3kD35egW/WwcCcJiiJUr9Yb+Z+tnfMSzDswbSsT6UGTrCew2kz5nI9L8Fp4Oi7ES+8wlpoL - ue9iI5rv23OjXBU4/q0vnvM/FHZ5pGKJK+8R3tT0v/5rAaue9vX5rlzP/gFvHRbmk5sdOCVNkoDc - uBJGnR2aMmwdzySqF2kj90gWTgGvy4xi9Zvq/M+P4AR0dOgzmJPE5FUoZ86RhF8/AXTeDTPM9/cv - enwaCwiRCjvYbOojOaX8c1wMpfaU0CgrdMsWEwjRcZrB+zhvSSySayS8XVWGwy7isACXPRDeRjoD - uuU6vPExB2Z3659hppKOeJK/B+xuhpriDOIL79JuPaSmNZpyDzQVXbB4c+ef3+n61kTWtaUj+633 - n/7cTiXJZ54Zd8VrD59gWNerkKiDBZXzhIj/RXudy7PIUcwX3gaAS9xxR9qiANr39EaqZX5yAr84 - BV7oc+Rc6qFLjepQKF0SyIGw9muXYkEDx7C7kuL0IjXzPy8OlOHRJ0ax3QNu6lSs6PkLBFDb1/Xc - m1WjrPkKcw4L6958Ly/wwk+KtGUowXxaVE1hapegv3oGBx/Dket4dPzuPvncTEcN6IW86rs31cuq - H2B93sR/8cM4JdttA9UxTkh4/qg5P3kDhJx0/SIVHAZ3Vm/5C27J5xpkV6HKGT2xVi77q4FUN2z1 - pQB3FQZvQ0OGw8Jx5k8RBy57y0XmYwf17/3wraCQyU90sF6dvpx2mfind4mZN+N092RZuUbSkcRC - fsux2FwDsOY9ZKShXnOCzhr4GJeCPNc8tYCmhAqiexawV2+A5VLNqrKNxyIA9raJKLqjAVS38x3p - TdDkH6EKHfiMRZP8rlnnnCyg9aBEOnxBtubtStECVw42r8Bw+UslatCL3iLm1K+o9/y1xcB1NJdo - aR7ni1xGDvTZvSP3DN9reuxGA4pFORMXaa96wdziwdG8PcmRXRp3WsMNXPuRmIe3Pi75q9XgIusP - pB5oH9HYcAsoWr2CHHc35UtmmwIEXFliwZXKnJaMswBxXzIJhQeO5hQmZ6hmrz5ol25ik0glCMze - tUnQViVb5G11V9hN9skTCd6442RbgHaQcwEswA5gbDWpomVMICodo3GtVwf5yzFG2qsjIz2lOAb+ - ng+CWeG6etCLTQdf+EGDXXcb6kVvPRWQd/sM5oU/16ufBPB0OH8wr4dhzvo4NcGaH4i9fVzAUt5C - GZanBaFL0qojD3EnQ1H8XIh+wryOJSBksAriChVbJDP6qK8NFHl0Io7EmItPxgWCbLNRMG9EE6Dh - HlnwxV1a5IlXlS3mqfy7JnZV72v2uhyb3/MKvqenHC3ly9ZAUKcqOhxFz+VWP5Z/eXLeO/647DK+ - kiRT3ZC4aPb5oj5xBuBHUIgePliNDeEiwNxtJGSHoc+ww4scePsDJQ53EXKyyPdYJtpJJ7muD/Ws - 32wNUEO5EhVuPiMNW7OBBU9ztC+ie0QrN08hvbkW0sbGdrnP20ihuJd44uyMuF7uou+ByfMGcjIu - JaODiCgIjPBCbqoBa2qNiwwdJUVEXfOEsCBNA9ImeyHT1dVcWKyTqLRWZSBHVKhOvv2WwvLTawFw - sgbMTYFnwA7TnbhUxOOiG1H6ly/NNT/OPtaxAsPIIo8xqcflwG4zeKr4QlzD/LjEyr4tPEwfPpAD - d8ixJZ0tZb1/lDzrwt29jp9kVa4tfp+sGizQLu/K009Ccqs2lT633XeCOtYGpCq3Tsf1kZoK9JeU - ZPGk1/xgHhJl1XukqsTX55ZcBAiVaSa5hSswudFjAg9eT/Cu+RDwW6/yhFqM9vePnFPiMk+Od3KE - W5633F0ezSF8J+KAmXOV2eR8WhM2wzUhTjzVI37PagJcAx6It3e1cd7KWIbH7/mMHCml45RnkQUh - VDHRk+UU0WEgFYStA4grbj75u/8oG8gP2yPmbmYzdsVrprD63v0A6IWrz3VMOMg/yoIYRVNG9FjS - DXQd1SXms97oVNnIGPaenBE72U76dB81DyA+3+GNy18j+lt/8KQ46FFf/Zopr2C9Hw0jddVrQcKJ - BSSg3rF0H5p86XFrwDQePGLOdATEHIkDp/15WdfjsV7OPR2UN74/yXnmVLAzi0CEQPBrglZ9ZW8h - 92RB6EzydIMZzB86DYr61VKC0rCOWKIaMkAgGpAqJiOgR8jusEs8GSWi3eq/9akEaWFieRg27hK7 - IacYvnVBRc3qfInkHkMpTmAgQXwd5+F66qADJh8FD9nUKTnXMSwcyf75IdttZiWAwePlEWf1H7Kp - 4w1Y64WC9/RidPJWv6t3FgZrnl2qeTaUHy/ZM8XJic/vDSjZN43sIS3HBWu3Cq7z2uqfe0C3n72q - mH5eBZvNTWK02jkbGN2+72DH5mb8hpxnyOZr2uItR3budBvvGBiWiTAvIJZ/wyfA8EQSFzdx0TEa - vTkTFA6w136HOVkusQxVKntYiZioz9r3loK9zx1IIR7uOiUvvVWaZT2hvN4/5zUelenY7YjW8A7b - Cb11hrIqh8iLswIMz3OiwRJFT+QyjOqfPsDpbBSr/pX1mocdGKs3c+Uph0iwr2UF7WM/Ied0NHT6 - yzPbL7hgFmyTeplw1CjEDyA6nKyaLfHOysCad5FVO5Y+7eJhgBx7ykiv5mPEX/qXpeRfL0JxXFiM - /+X7iiV7vH1g1eV8jnHKMNcTBkhTR1zIGwqmy71Cya4c9OX7yNTf/Ev2vGoCtrlPnHzbIBGPLCL6 - tzy/ZrjmAeSvPTSLRd9AZ6YxUgFwXa7jQSEHTy8g2ssnOsvsJVa25H0le23XgN3sXjZw7wsH4jit - wwiX0wpW1yYLBCk71it/8oDwTQN0y71j9P3l4/fF6/7mIzx9TA9sRLDD4Jst4Hu+ZxlYeQA6fhK8 - ntjEGeR3Hz1gOfetZ9YfKrjqDbFuBhpZ7luWYr+PFwz3ra5TcsvvsA+aO7kDqOorz4mhVM5vgjJb - rXmlVAclf3cF+ZunBSJmcBv3BY7y2/dv3gKTxbUo+ekZDnahPPVDhEz12AK6r54b4GivDXo8dMKY - F9oDMP1bhXxvr9R4PugGfI9GjvRW0cblMqeh5MxzjOIHVnMeXPfTj08EXBrW9Z9ekdQRkFsfH2zy - Gm+GQS44yI5FuxZe+/sd0O7MguXgamx3vU4mOJCtRFa+4M73RY5BLLUXYlmPB2A5nqlih1KDDv6m - Ydzj0W/gdvIjYpI4YKw+3gWoF+IW/a7piD6qcirVANmoOzGiHA7Djwchd94uOY2mJoBZenqsvNMG - O5/fm0CQNQfpV28BlLqvSV7riZKNpkfCIX57yttIAfEEvQJ0ehUmQJseY1zMH/07PJoKLsqtxJwb - LHp/FLoGvk1TQeY+7aLXLs9mqKbcB/m9aAJaGeMZrnwNj/a2ydf88FLW+QhlonJ2OQ05JixCqhBV - +vg5vQRaqvz833oJQ8QO0jGDgjtzKAQvp+Y99LQgSS3hb16m9eUcQ9ccjsHp8QD59MsHv/nmgFAb - fUvulgEnisf1/2iMV8+mo/z8IYo+54hY5sODppLIgbTV+GheeawSuOqDJFL2qVmevTK4/UqXQITy - Y1xGO3agjAIfOV9O0efthRTrG0wNsZ7MB7u2o1DZunlLdNcHbM1rKdwGy4CBbdc5W/kZvGSZhoJj - B3UcC7UjS3Aqgo0IFHdGw5CCn/+KV5e6f/ytPwox5jWegLnIrAFoHJ+Tm5sn4zxs5RRGj2NP1MO1 - Yy05jzF0tuaJuHZTs+Xkjgl8qLqIwQg5QNkuDP/0z/GMD/v+5ovwCFpkTk6os+x6a2E1UpdY8Y6L - Zj51N3950+WWSf8q3hzCY1s56CC135FOXHuHv3nM3dNep5q0c36/RzxBqgERhNcLpqQ9IHWiSU4q - R7BgPNwHcln/Dy4ZdGBhuHusVKbn/niyLLXCGCjfoayF5nkK4ZuLIxIpyYORhDYNmM3MQ44QsHGW - BJWCq9x+iH6KuHoBY2bC5kuyv/l6eguHDbwyjRI7yryRk3MrBD/+7S35x/3xbLjyrIC/7tcT7U5v - ADuOOmLfwIGxz1VKgV8N4Me7AU7efQNLq7yi4nqbQH++n1NgVvmeBMnGi/Cr9k0YlVGOiWHQ/8wz - x1jLiQmTMic0HSc4feAJubZdR/RRPxuQ6w5B3vD8RvPT6zLokE2LDiuvH3+8KDTqCv/xt+XIddC/ - kj1xScnli2SzDeDTR4oZfN2ZsLWWCY72UuPOtLfr5F+ewTs7Y3JMO4ktAsQO+PndZQiVfJE6NwS+ - 9UkC/gOf0RI9+wAKjXNFtpZm+XIk4h0et3SPHLoXI+YlEoXxx2yRVZ4ctmsKqIIDffZYoPo3X2bR - 0iDfnw7IOL/d/OvutqLUm+sbCvbWyPmz5Wi//EJCOrJxqdy5UxRBVNEzub7ZJDqnAC7z0hI/zWOX - /njomq/Q2T8/oxKMmQHGd+0QRzwuv7wkAJNokLjgNUTLyyoFZeW3JLDNmz6fPKuBuVkxFIgH6H7X - eQzqtbEg3fVzMKfZu1MQ1VlwOrga4A+yLUPmzBVJC2Vi1MgfCYhPAQtmXRoBtfyhhalmPdAz1C2X - vZ1x+PO/hU4dmyW9N6FCEgcZlkEjPA9OBXVx0yCTEBLRn7/89N+hdxHQRr7NkEhziH78kfvl9/Gx - jYl/embRstvCDWiGS0IQyif3az73KRTE2EJ6qDG37/jEU+KP0SKveO0iOt2+G/Crr6l4WvTnj9o3 - emMei5JLqzkTgF91YPXnwO0P8l6WtVc/oOByQO6yiw+ZYsenDnk3Mf/NZ6pcPFJxzR8tm99wY8Bp - CWxkrH4ogGXawCmiL2Rt2u04NdIn/Pl/sDmBWac/HoXtI0N7VtlglnP1rIwGeaBj8yFsKbhRlM17 - 7QfycVjc3zwvp19RJyGsP/qbRGoLDd+5IKOwVbbux1G48ldky3ymz1u5FaH9KJ94oScrp2/XkuVa - zjd/9aPFN3lBXVdM4r3vJ53RiTOhKL4vKEj3FhAwl2/ggPcQHQjO9GmXn2cYVcoDOYcNcb+GAEwg - HNSJhCs/nHW3m0HVpUf02y9aefgAVx4XyNb5qy+GMnpAGgJCVv6r/+l5M/tOsLWPQcQi9Zr96b/L - MKkJPYEWwr7aBry99OOkWTcIX7vvgOzyrLq7Y+dzcJjLCTnes4voZjkXShkefHTRx5kt6/3BRHRf - wTzRZH0DLxLgtHg2WvOTy4vj1ELzCU4BvasMzG8dh2Dd/8HYxzFrLBNtQCfi/o+HYHHZ3aHStVrA - RcLEyMoHoO3q89/+327e+RU45U5GbEti43zy1EZRL0WNWX9S69Wf/ng1umg8Yru8UwfoqAIMNtaY - jn96Z8STi45X586G336M8dhoyOEuSbTyoEDRbd0L5rlZ9DmzJwPKrmAHy3Cm0fJ57FTlisiFHL/7 - NqIr//z1d1B+NlzOfJxC+Lj3DB1LLczX/TEHlu+tiqxgAWN3EyGEHetqlD7Tb1QMIpqheF064j+K - OcLVWX/9eEsgfNVFn3JcVz8ehGl9/YLhYT0NmWy4BGXzLDBCmDDt3nQ4IWej4nzeD07ztz8WhNqZ - CayfNQUIqCZrf+ncrz6VnvbEFTS/5rS2DyFPX8pPLyOC91cNepTUxF39evnND/qht9Dtyekuu439 - Bg7DKAZKUgk1/g4PDb6PdIs36a5nzDIvnvJMGgv5n6YDeM2HQAqhiM7jtq/p9/BogVTO/wYAAP// - pF1Lt3M+uP9ABkVVYqgo6ha3qs5QVVTVJUE+/Vn2+z+zMzvDrrV3Szz53Z4kPuSsLXW0vRi3FSXZ - q5CqnUp782nHQyvN1p3f3fFf/+h+D2xPDKSPOvtdw8ILd5zR+XdcwZ7/llA+VAaJs7tuH8GKsCh5 - 1dNrz6bQ0H184F5vxFOsQz5LumxJf/3YCGVHmx7zSoERRLEHj6QFm27lC+RY74Dh5xg0S9FmGpSP - TOjdb2UGSK++ldP/Y0UB/3+vKChSQSXx8aXtHf8tkdxHU3msdSfqVug+ljxID+Sc5gud3lWUSMp5 - yJFd5Txd9HRNYKIlF6ItpxOdvcOxB/4Udd4aKknEr46awsq0f3jtDLnh3OPJBFEpVB4+vKyRRt5r - gufslBPlyu5rwk1ZgZG/SESZYrYZWfvUw08jb+h271mwRN+6g2vQH8nTd9h8GaUxhjObnoi5Sa+8 - x0XqQe31lpAcgm8+vy/uAF7jeiPXINdUVr3cLXg7WgTJ53Ecl04gIjzFi4f0YK3pdotaUcJPpvZA - YVX5rJ+LGl6V+URi/zCrk1MrGSQJ2ddcv7Rx9Q6HATyhWRNnUfdEeLxiiAxzRHF6Ce2j+fFD6bPc - YpJ9Lm20zBtXwjw+V+hsCK9ou5+fPmgnq0MKTfV8qUjCwyTovyROm5P6e0sjK70E+UniOFRG9p2e - WmAE3oCMST2rnFV1ujR/3YbcJFDYKxtwiWhapxe52sp15CdBy6SEOZgY4M88DqclzWB2evfogroz - PZ75Zwz8p8si7SjtCuu9mZLnpSrS1MCyeeNuh/D+eMrostHWXgatSiUSs5jsv2+zrbLq0vv3nsjZ - k5to7Y9rJsXU8Mg1cKrxeN6mAqaqMhBd0AabDYmTSCCsPvikuSCnkuCa8Cy4EnGYmQXUnDZZonO1 - Iq1DJj3+tnBfH+YjFAeeP/J0EDz4Nx53Pxvoyh1MQcRizCPPJnebc0IjgcqsVSS5UY6SffsMTCOb - J9bVtcZpOwoC5GN1wSALDmDlgC9LqezJ5O4LAZhfUZuC+7dt0W1r5ZFt9DeUYJYwxMaJZX/TKDKh - QE4DcRgo5IupCLykF46I7m78ofzZOsdSpPwe5AFe32ah7LTAW7Y7ui9n5JuNgQ7F0+mKnFuSA9Zi - MwXm+sSQXFNVm5eiUARUPEbEKxbSTPHD5v/mF9Ksbmuo+uU7eP74BnncPNoMARuF0nrLiUfm5Alm - A/c6bEcnJefhSpv5uXoDjJ5MiKnwc6K1kgUZbkUkovNNN/PtdNxMKZG0CRU1NVU++n0y6eh+H0Q+ - i0jl86rb91hKPtFe1wYsB8ODp+dr8ZDv3OA4i7meSSkuCIm0PKBboJYs/F5uKhaQr9qsW38tSCKx - QqaUPOhWLVCA5D6mHmMz2chXn9qHj6vIkfPtZtNVjKYeeOrFQtlUHMDaVucWamXRoPtyYe3t93h1 - sIrKBhn6U4u4wF4s6f6ZTJRyjZXTRbin0sqoFlKfMm9vxS324KNNZ2Tdj3nO+yD14SVO38T7Xq/7 - nsza2s8seHm8HKoN/cprKO31gexbXeckk8JWil4sJkhPHJtLgzQRfeEnelL28ZoFV2cNngUkEUOZ - DMDNd7JBYLgjybfradysM6lh4b3vCOksVddiWhWplfEbWfo1bbB1/lbgsY1PD5wQ0yzVODMQRCX1 - +FDW83UdOyzZz/GDrsBv7e0qBSxU722K9OKUgePYAAzdaxojQ8cZ5Z1vkEiaggqifKakobP+KCE3 - oQTdpbWhU1TJvPQq2jsKmuVhc5+CK2HrH2eiPsZ5XJ5SykIFOBq5f06DTdX2su+z9047PtjjFrC5 - D7PXeUHofaj3eloUKSfJAyl5/Ropz60atHs5+IeXy2WcJ4ifsCahiAy6HcyPLJ29t0ECdl3H+qrR - QWro28bsaj7zWa2/e73CkGQ8NVT2dLkPkFeF1vvteEIvt46BsGt18mLMrcF9VXTA9oovSQTNUuns - Ag+uzCFF2uuqUq6zrho4ZyDHA9+qIyVLKsC93rH8NtaIhHrgSdq4XsiDzTm6DGYQgpd895HhZwPg - zuwxlC7N50Vuvg5Vkp3tCm4fnCJ9HcmIU1yxUhq6PUqjkFenIuJNeNF9iURJKtjreRI2eEzFCJ3Z - et9zFcmDdDnwHfGKt2MvenqK4adRNi9Jey1al62ZpPBnEOLFkqwex8pi/54fZj2rbKj5u5r/+PK6 - 9F2+mmMOgXk3v8SvflbOKaueSa+BM7zVK62cc4+rCcXyqRL7/BTUhS9bH/bfX4yMPJhsDH5R9ocf - uAmTi8132+TApk1cTE3Hy9mNNWUoDtydmG0d2vzpEImSJH2fCPmVatPm49fgtJo10tKYa/AkaKnU - JN4PebJwAywufE96lryMzkGLoinPgAmONLNROcBT3rOzKkswDxykrs03osethdJXqR8kOHJqtLhZ - Nfybf+m5jXLOE54CNGkFUGE8bhH7XXVTehiajWIQxvZiL28TTu+gI49gMEa+ZPfEnlta5IdT83c9 - DtQLTyQB8zzn7OvrxxI3BibK9O9hXN9Sw8OCv62kYOOsYZvbXMIaTCsxE0MbWfa9ZFBHh807p7kP - 2MUyO7jjIfGjfgWUP/0Y6AU3EZ+Ky83+xRKjg6m6peiue/sejc0vJPGov5BScjd1uRpeCTeOa9Fd - x+K+Qo/XJOG1lEQu1VfE/9WXN/WUXMozibbgMIvQupMec+UryY+aW5VSW6oKMkp6sY/J7FQgEz4Z - ida4oWvNW7r0WGeE+b1nsApzIsPPfRbIHWRFtNLA70RRkGdk/LxtnF7bwMCJKQFS7sFPXfIoSKXm - oUb4NNM7oEpTwVNJZYzudceOM6esvHQY/AwFE1Pby2s8e9JPsR0Mb4aQL/MmlSAFa0Dsasii7XSI - BKn8HFNiUXKzt7jMFvDizTMp9r+fkP+qoHWZZJRHWZmvu16EY18zHuPdFPsoc0sH+5YdkDYdOzov - fNFCc+tmD/6cuqHtfJdhfgMQGbon0yllpxr6wih67BDkFP/xa9U2EkFThsCuHxOJbLZP1AM8j5xa - ZjLwCzMj+neBKu1ylIJUlQeSXnTnv+uxv5pKbqe8treoVnspFY+qxxzOTYMPGjRBVho1uqbhZ/w5 - 1qeDrhXXyMrAnNNQfzhQHI53ZFB+iagjybqUMIqI3Izd7B0fFfi8PBoin912pDpIF8imVw6vKxLG - 5bGAHpiPs0ry4mhR+gaBDE/XV0y8blToOkWZCa9PvSW6/P7a8+x7MlSvBUv0wVBH7lNIJQCSUiCv - ag4qDS1Lh9vj5ePTjicU6CMDza2dyaX0N7CdVI0HqG5XEqSnI93SEyj/8Nnrn+61GcOg4UGNU0SK - 66GKFnFrHOnNzwp5kqOUb+XxJMLf2WGQvHVLMwfoLUufNaqJnOAEHPlNdqQnOWXo2s0bWPqsF2EE - XcVb4/6d0/qAeyg8RQV3oKvpdnY3D6aKViD3PUbj/AgkHSiSO2Eh+dgjvQSMDnf9gqXo9RvXAL0V - qfb9CRW3BIBtOmALUFweiPWJDYqf3t2D7elpomfkd+qShDoLiv5yJq6CZjA9pZSHxPx0RJt6NSJf - SY4l7K8usu6dRo+OZGpixwcB+fMPJF7SHorPisNTcLlEx68ZyfAU8TJSd3xlX6IzgdAhOT7G9jdf - rEPbw6+WciT+tWu+vX8HEULX7olCrzXdSjAsoLxNjkcH8IvWrw4h1wq1uOvRwV5ty2TF5sreSLK6 - fUMQ8XQIWZYntnnhbXpdUQ8i4LMoWUfUUOZ37YHJ5XcPfCU0rtbDieHI4Rg5das1PP6wC5wimSf3 - aRkb6kotD7dropHrrjcXpPUmEIvui09YAfl2+YUi3PUqXv7wL9PYVLpboeLBwpXyed3PHCjcz9Vj - bidsb3mFeXDw2Jig3KzAYh2mAZSrGaKr4ITRuvBFB4/WRkh4WmVAT9UtAYqEJmLjjztu51vOwJN9 - W3D30MdomsW7A0kyW6ismpfNbamY/NOX22fix61krzJ0mEtG4tWUIrKPB4RdpyMNhKxK19nZ/vD6 - H79vecEx0u4/0f57DQ2VKoFN4vzIi+ke4/anN3SWT5A9C41N7nXSw7PbnZBareG4MJ4Kofasbsg9 - 12p+FF5CC+XKT5HGfK7NjM9KLWlFZxDvgD21n4nTQqXiHh4ru0KOA8Gx4LBMH2Lc50rtf/d4gzsf - YgY6t2i9FmkJh3U9/PNbf/oSUEgIMsdjlq8cSBXYnl4mbhubB7MLalna/bMHsuBFlzx67HimDJ4o - dnyzHJQ1AUkwfDEEVxNgMGsh3PmEKPrZjWhZtB4M75qBXkCZm3n8HVLYlmfFg9eDnHOfh9ZC/fcu - iM7Jgj0Pv6yHkVBaRNvrefkhsO+BXy+kPLyGsX98fgrssxdD5MiyACuXzQI5vlXIpXzx+czqNYYa - SldyuYihvVqF5En9NW09TmNalYZKH8Pj96oiRwc42vlVA4U9zShWyapu7FgwUDWrDrnnVQcrvLgC - 3P8fhbu/3sJprSVZ40Okp8cvWB96XYIkbDlkc67dLK+vn8Cyo51Xu0aYk/exsqRj53HkatHOJi+3 - x3/zEykZrdXl6cSZdOoov69r+uTTcoucf/70D29I+8MmeNzdO7IU1re3aDVMYF2wjKyjYo38CTSy - RMH6JpbJGCrbHmRL8nPPQTv/57SZxBCSmgmROjyVnAYVSQCVPR8febHe9RS1RDpMBnFq2qvrfmAj - YEzhjjQjktTZtkwe7vqIyJcSqJt16iE8Hfw/vXsDy56XSLaonna/3YOtBPUCFaX7EXsMBZXes9SC - P9lf/uUla2UGG+Q1dkPPtlLUVa3JAHf+9Eaozeoyj/kk7v4SMzYjNotyvtbQOag9cnnMgF+N+1jc - 9Sy5fIw4Op6+R03yhu3+50ejtVjeIqRnq/Qq6Vyq8yn7DUDPd/zWgRdx+WFOod5/PWQn5RLR0/eo - w+clb4h6uR/Aam7SBC9tM+Nl28+MSEZcAhDWH6TFb13d86oYvFm2I67o3+mmd94Cdr/uzeZDo3/+ - Eb7m34psgbTjElUmC+fH6Uyuc8vue34lEySXMCI7X6skvMUTDEh6QkbNpTY7Op0IDXGLiOYO7bgZ - 6F2DdBwu6AqwbW/fpsKANeEROa+AqKtx2rAEJLlAHmspOf2gVAG3som8owhvOZE5oYOmzc1IeZ++ - 9tIt70x6Sggiq/Y5Srp7UsEzf38T1H/lfLvM6iK51yzGPXp1zfpCoBaX4meQ5/W20ZU+Xtq/POb6 - 4tdoz98w3OUVuWIlj9a3882g7tomOTPAzFnTCjswMKyJypWKYBs8jfnzK8gkfULH5zfU4Nu9puhh - MMcGP3WugAK0NmSJd6puf/q1swfXO9TeByxxWxZgzydQjo1u3B4bX0iX9j17R2CnFLfOc8fT9uQ1 - w/ShSyd8Bfi5YB9drihrNhLJKdz1hbf5Yxnt/hIDR6mu5LrXc+Xd4xr8TvwDv22bBZsX+LzkoSRA - 2hGb6nHHcym/nSCSL2VuTwF/3lf4zxFyyfEZUZkTWmmvf6TJbpp/gVXwwPDDAl3gwbOPh1+XQe/H - Xglyi8Bm4ZeLwZ9+k+FijPTPTwYqZJEze7dxbJbGlI535oKh0aeg/rmVBzPhm2GOMFwzn8qwhvRs - luRqBK667fmWGJ3SFMVNluVkrzdYLfTiSc6pBHNbpBiqDZyI+kSP8XfbmkIyxCVCN18vbP4FjRqC - 2x0R43j9ABoft1Q6/ySXuIpsRYs4Xksgq4KGnhf6oMepYCa457v/1SOQBR8I+OwRpX9/7K1v3QIW - bHwlma85KpXUFwsmcVKJ/9h6uhY/05ISLb6gku1/0Vo5Vg/HIc88Ekdbsx2tKw8vYX0hl+TZRGOx - tfgPX/F2D372IooaA/vsySCTP6/R+qcXbKe1yU25ciNl7bWXHibzxQdlq2yiuX3xl396pw4mOZbK - uYZ7Ho3f8bGlc6VVwp8/8mrCz3/8I8PwrhvkegOlStDGQPBvPjk/O193vww+RjuRFL30cWlucwH2 - 6yFXIOFmTsg9hreT9Nj5LVWn9XaAwC6/9E9vAy64WxvwvYOG4S0BlDCfdPnT03seI1L8Hb8eDJhy - xEwsVeqSzd8aqg0zeYfr793QZeoVIDHNiM5HrsnptdZMkCtQJ8lPkNVj2boM/Pucj2FqD76y6fBP - f7jPXzUui2W2YHzEMglWlI70fbiK0EycB/HW9h5R4ZvF4NRIDjnH/Tv652/aIbSIvhaOOgNZCOEp - GQZk3GfZ5i5a1cP+swlEuTRdtJm+F0InIMa+omwGVE33M3latcDHwODyvZ4K4MvcGzdr880p1hkH - csfKR54xayr7+Pxk4LKHjji7v+unzZ/gjgdI5S5VTmmJZcC08RM9igtnk2G8lfCKrwwxXtFL3aLa - HqAzQ89b1/kwkr/89WTlNbFv6wes6BemsK5yAU/W8xtNXu6mINLeHLpk+EFXg7/y4Nx5t3/8M9+L - mvnveaX2kO/5Qi/t9Uj++gVUCx4xuJaSTM5Dele5dewmyK7zC5+KWBjXi2Ol4iliZZI00tSsz1Xv - wVe6FqS0GXFcn/NzA1LgWsTy60hd//JA1tBy5O5+hKyzs0D9OMro3ISbuibklUCYljWSR4JUTsJ9 - Cy+MmBJLYRd7RW9hgLu/8phCvdjbX17IPA4rUqtoHNfQdze4qqlBEtDtZxpovQCNV+Ii97T5lBL2 - ukGGexzJpf25+RSOUQmtt6mjFElus7y2GoLW52biNIo9bmdX9MBfnhHys0fpunCDtAiGS86HfqOT - nInCn19HhdEL4PecbwvshOuMZ8jXYIsj2ZQepvzAazrE+R/fQRoEA9nzDbp21lmD8FvFyHyjYqSc - FPIQXUKE3Ohc/10PA79q+SV7vqT+zW9AbgtEbmYY//w9AIwoeFTLV7q1zq3/x89udFbG4z1LTcg8 - zg06z+6l4dKngQGK4/MfP0bkojcl6L56RKz6HObYTbcOaPJNxkIUJvbud1kpAiGLuXsf03VY2Aw2 - 3OPrsb82iP78NfjLp3T7O4JFMRsfdD1DkLq9uHzio6AHu35HqfY4g+2ysIN0G05f4s11G62slXfQ - 7pkXUaF9ybeJjSporUG4X+9nXMdOr+Gej6DzKnP2FDCIF3d9gH8qbBsSuiIDLtF+BsXDBPmi4ZWX - 9vmILMX4NOv6DTToqOX9L2/L6b0ueyApMCOZaJLxx96wA8oWWGjPs1X2rtoJkCKjQ4YCxohIbOH/ - 41+o3uucerytwYZ9PkhsdeGIi1vhnV6KOxHrKnI2dsidB69n9UG6vFgjX4JhA+8CWcjjdD9a//KG - P/2mhp9+XI/T6sPb83shXmgGzeSWjxokRgCIvfdjJufXWpBNbc6ruEFpjtfV6GF8/5mYjcO6WdX6 - 28Orzz3Jhe+iZiuCJoanpB+8I1PiiP62rAR/9SATRRtX3WRCsPtXj3tuI1iyOhZguTgaZuTfFi0f - wCqSfvzJqBRygS7m7ePB6LSfybLzyT8/WvxUhDw5w2B9eJonXeLsjRRH2M/YEcoUTKAPMNz9M4uL - 1AH78yD6jd7oBn55Cv/4zFKMy0gNwFVwql0W8wtzUff7G2CLlpDkUcbkSzaT6s9f4KPuVXTXe6kk - iQtFKQZOxAdqwgMR5QG67P0p/u4QARTW9vLW5s2N+MMtiVRw4d07qfNZ5c3oYEFbuD6Rlx4idZKj - xz4/pBV5B99QuVP27qXMhQnJTOarYnYsIHS5E+tRQSC73nMg2Pt9yMqtBSxeNA4wb4wQ6QPrqfT5 - WmKp0gMXBet8aH7vi9vDK9oUojBVGS189Ojhb9i+O359bJpGuSUa0UXHkhRfAXUv+gCW5+eGcpTa - ETUOXQqL7DSi61CSfHm3mgMfTfEiTj6rKkcfL/2PX0hkOjhaPxWO4d7/2fsN7p9/MuGJKnvexMXq - 9pfvXMuDTNRq3cZtMw8+bNaeoEIc32B5fzkdRkTa9VtYjcsHQAWeO+eGgbbMDY2PYgrjxW+RkZNI - pWdRd/75D/nRXCLqyKsiOlgsPHbvN7Cm/pvgPt+RinFK8948LqJr+wu+tdkVcM/5tsFLWF3IJZOk - fBCFUoC/Z3lBbq5y42y1JQMN7RyT++4nlj9//F//4zSqiyVYAty6oSP6N7zZ29/z3vufJJKPHzou - U6UI+pgLSN7us71abQnBKx+umMffZ4Pzx0WHWxGIRE2ucUOfzc+D/48VBcf/e0XB7C8RSV4TH9G5 - bXi4wcMNC2y60mWezBKCNsmJHR21iF6iKpHEebl54jeQG05HXw+mxX46eBCjZrEfmg/d2jthOL4b - dVUL2MPomiJ0taR3RLZxUERJdG1iJYIYkY/LFdLkJysWi+vVns+Lt8CL++Sw1ISB3bAPrYK3h1mg - MHfWfJve4gBiznsQJBpqzrfzA8IrNnTiPWirLiLAOmR+8YTKJe8iehJaAX442SW+lPyi9dqSEuoK - Eb1Rb5p8BabCS8+fSrzaFZZoa5uulIIGUGTQMwuWQ5V0cHpzkDyV6yei0ytaoNIeVOLcvHdO3w6K - RWmMFYJWUxm3r5L58GM4M/Kl8Tqyr+5awgINTxLTjqG0bGQfaiMcyPUjqpT37LiC2sgMyEjHaly+ - UltKVnS3iREEwJ4/IXWkLmlS4pWmlnPz79fCa6xUxMXPPMLRojFSKGgeid4uBvSNnwp0HvJKEmFD - gEORXcLRLAev3vLzSJUiTCQVHq9Iz38NXS5BUEgqjztkODCjuCmTFN4T/4uMpZHAKn7MElwcm8XM - 1/jY23mzY8ithy9C5ifIuS9wOxhE0w/ZzvaKlq/hKxJ/l98kXskNHA9fWZB8snyQa9EPOPrPayi9 - LBpiln8k+WjnkSmpbISx+EjWcX7pEwTW7HyRoe2h+4/dE42JG9DDZIeR5rW+iXEjpKRMMENnQPoW - HC6sjZzSrW02a6MEnrITIOhb03EmBPtQP0UHoi7zKfq1Zq6dhJVxCErvvwZ/tlqWQoPjye04yw0v - o6sMQ35x0d3KA5tzTmSCHehCYnuAV7cVHjuR1z4/b/3cDvYqDzyWuEYr/40n96bZAoXozJF75uf0 - CMkmSFmrlshtwkClr/PXA64olciXf7PN5zrI4HiWdVIstgpYK4IejF4zJEUMo3HVrkP9V2/YxWau - UndUE7hZ5pXczrwXkUo/hFCwAoTMVjCijXffCdyye4RZQ87tFZqHXaFfdMxcM4lSs+9beFgd6jWA - 88Egfk89HKxCRuEY/ehqC0CHl3LlSXDj4obP8cOC6Pb64g+c3uM2hkcdrLG4vzUgvOSLqq+JpPJT - RwJF/tEFBboAR3TyyH2v51UVsSjKVRqj/Gl/GnpmYQUX/L5i1r7lOXd4HidwJJeVPPCpbZb7hbQg - mFeEHlPujvwYHnRgWe0HZVV5jWgXuTXMzIHH+2kDzSKjJpGaCbfIW/TOXjBSWCnVdAPpMmHtrXyr - WJrLe7YvYKoAZzYiC8ORnjxdBF2z9EOYSFnEx//wafu7X9X4XFCOWZ2y5FFDqX/WB6JUdphzDmOb - kPE7AxnvNGrWP7z6ZW7v/QT3AvgpNSyosgEmj7RTAHv96gkk3U9F5yJholU6+BtIXP2FVA03DW77 - qwg+7yD2lmvvUg6oJINW+3OQdXl3AP9UXZT+Pkffh6CORz+sgXCe2b2eHzk/PV8efA5mhxe6ZBEf - GFsnUY6JiNsHobqd568IEy+rUPKTj81WjQ0PtbbWkfwwhZwStpDhq7NFoimJOC6ZcukhIw4zXr25 - AFPbt57ESuwDxfyDzzfCbwJ8uY6FPKNSVH55SCHULl+CadS6OauHui7et0UnTgzpSA71pwCHu+UQ - 1Lze6taUZQqG6/1KtMtS2dOGG0+KvAf4wx9A9VIupSxiYxIFcxuxJA5FeLVsFRMYTWCZuiWT1rkf - SOK/nYi74pCBp3bfY/iqRbCK1UcGF5g5yFYXm7JfaSogg/wL+qvHTUivBdS7UfLmbQPjEqgODy/6 - OSCIszl1sXm+hlw1dshIHyHgJbiKkvoep7/7abp9fMVeLR7oPp6eKmFB6EhdlXHIvRyPzbjjLwAP - UyIPWOv2kY3YAqrj20TFiwx0TUNnf+vOyBL1fprH2c5lFpZDa5JgevJgPf5eMnR1PvAkTH7qggJP - gAM7r+T6MpyItfizIG08u+GTZg0qmZ1zD6V0kkkQKZK9Cj7LSyysU6Q6K28vBkMFKJSySe5Z/Wso - GAYWHvpyf4uEW+brUhQW1C/WRq7f1QH8dJgEsUyOCJ1nTQO8nlBWqi56gLf6d83/4ffXESNcj9GV - bnxnV6c59Y5e9XwK4/TM37E0RoOJ7NPgAzaFsigxztH1DtvBbubxdoPgFxsXdH7pjUrvBl9Ik9HZ - RNUeasQqR26S4jGGJEZDPtLrvAzSOXIrckXfMziS5L1IvMi0KORLVl0opb7EyoqItCKt8+Wl0g2u - iz+T14njxsXzQx823NNCvpN/8r96Ape3R0haXwhY2+4SS4bIpijzp8Dm1nfrwM2YAxKox5Dyl2Bc - wGgWAzEM6QRW73pn4EHNI+KePpHNPe5FAuT8/iDXDR7tRVhdBr6eeUzOQzDnKxuXG5ib7IZUV+ej - 3/OhFJJ8sk5I8Vx/3KIww1DIPz6KejUFKxsni1QzQMPs/XGO2Evyq6BWGx7RL4mi8mnMlfBr0wIZ - 6BKMvIbfmRQ8uwvS/BjmWG3PvvSHvw8cJs0qRfdOejlGhIVHjiJKvTaFuotdLIbeiVJ49yeJvLwS - r3dmzBc99HTgiocSXazKo5z5swT4PZAX0ouoHRe1bVr4UtwjQpofqJPk4wxWec4g49sWYH3ZciaF - wRwhO8DuyMJ5Uv7xi/K3ZxNFzwWqrd4Sq30DQI/YWSCQs5TI9isB6/Hm+9JBqy10fUccWPmSF+A6 - D4MnKddLzq66z/7DS4b/kqZ53HEP94OJvWNv2+r8szgePnCHEFLBMk7jHfKQ174/7/eObntCHaci - StqS5NPo2z+Wn0qJyhPCW8dcR85/Np0oKOxI3AVszcYoiiyd4+ROHqLmAVrCuwZ5Ner/9MNYGe9z - Lw04UzGjn1i7FS+PHrYB80NX92rRNb+GoeT1G4M0mQ3Acl1OLGj9oUeGGYrRBrfHBumk80iLDgAs - mjEzf+PtSR6f5ksNrhM4y58SXbXXkNPECURYSx+VxEqSNdjmCwfs+oLo6yCoay3UrNQEjUH0RzXb - q8IlLLCbbfZAyl9yGmc5hHs9o+zanqJNbtAGzx18kNfCsxSb75yBfvw1MbT7Lac25RJoKr5E7kdJ - VbfqWlrQ4SofGdKxpL9evk2Qe91bT5Taelw2X4AwMoOUJHZt56zcXypRPx4YLCaCmOOfs6WwybSa - 5O/qAthEflVwZZjRYx5OYq/Ti9fh0b5i4havB5jWVtWk04QgloIfon1ZnTIQ5t2evt0flAiSmYDi - xX09Qedf49JaKwT23F9RYUSbTS9r1kETbzlSk7sGWA9rlbjzP96qTY728eLBxbmyKFWcl7o48qCD - vb7Jrmfz9Y+PLyXl8WKv7bgwuV5B+ihYZJ0dvVnSuDElm9MxcRkUNUtTOhqk6HMnl+lzVY+1wFkQ - /5iMXFb0G6ePr8hQGS8dZudJzTk8lA74vuobuf4kPl+uXy+B39twwoLiHNRlFK0KtoaokPNbLsFk - 3EtTLEK/InGkne2Va8IECoJGSHbIx2alUW391ZMn7vNrGVixgOQpPFH4Mapmuh1qHs6vMUconX6A - Zt3gwD/972ITqIvY2AKgtZuR1N5cSrOCwTBYPJNc9f6sstsvMmEWkqfbHT6GzcYL8eHO3x5r30D+ - M+LdsWd6vdcHQ0dfpBW4L3cPfwrq07nt3ASwIH8Tcz729pTyAg/3mNoDBmUimrJTCLVfs5Eiz4bx - t01ODL17KREZvGSbDYw+BH4fn4n5q78UT2+xh3nAMsjvikfOfyXFgrof++iM9z1/5juHUIu8C3mm - hZFTvpl6eOl/kTd+uE5dNOec/eN3Y7Vsul7bbykdRf1CNOtS5cMZSQO8gaUkeaLOgAi3ugUaS3x0 - 3vXRqhbsIEoe26Bb0Ovj8gsWE3agDVF4IPVI93OPIU9WnVyE+zunAo15+IsvFyzs9z+zpeXD760/ - IZdBdJz8ZD8SUBJFfFRFI8LVOPKgmsiBmAp1VYqDB4bKY/2Qq6hhOu96FgjdgtGZw598CKp4AVl7 - Lsl115+cN356qFy7nCg/i6pbfDCrEzs6d3xKY05dLkTFcMcPry+8jg5i/etE24g8vFqHoVkcqHfg - 1NIUOZ7+G9c//+mt79Zjf3fV5r8ZW8BdbxFbMHt7/fu9/XrxQe2WfNWPGgRjy1yJpYkp2Cj1Gelv - PIvx3djbcKocaD6LjNio1mx2nuRSTE6r6bUDZPLfJwSe9Idf687Hxz9/y3qTgPQw5fIpmXwMVXT+ - /ekNemRB6Ekh90swYOIgP1pCukClcJ4oqfoWkP4bWFI0xyGKd76cNjx6oLwzOjJFvqezPPAT9G2+ - RuZzQc30eNkh/B3C+u/6KcaR0MH1WxkoeIdTs1LkYXg4bZ43mmYJqDzE/+HbM12/zfCAkQMlPTCR - lw+uioEFFZApVYSCXV+xj5fqS6cMALzIzByt43EcoMbOPj5Ecw+23FHLf/mHW7GZupYXIYFVdMi8 - IRGyaC6boRP/8c/+/cspbRa46wtv1dUiWsnpsYHugu7I3PFkK9/2JO7+k7z0hwE2OQ50qXxlA9L+ - 9D2Olla6PayCnHc/t25a18GnH1neOmsanXb+laLQ4vDEcm2+XZwcw+Oz65BJ8Grjr2RZQOAPAt7a - Kmw4VvZl4EJ9X2HDXtRVwESDbD1ALFlvQV3M7jHB6+FhoItue81+fzF4dVcRacoogtUo6wUK7Z1H - 6FtHzTFxHgIUTumCkstY/5fHlENnEo9d9PHIN20vOa2DiBEIV3sVhKyDMYIxenGbqq7m+WNBy+o+ - 5HJnEaD87FvS872fvsd7wbhUV5WFUw0psqNjmy9/+JZ1ReIVh8QFNP8mLHR/vye+BX3XbFv7TKFL - KxGdo/szP4pGVQMYqBq5Zs8RUId5lmDHH7zM7Wqvj6LNxLyxDkg3XFtdV3Vl4Ko8XST7e3t8hHSC - EDrvPf/4gr/6FLVH3O8rlI75nr8k8GH1DvLqpzauhOAQnDvmgY+fOQJLdE8Lyay0pyfezkG0/vld - yh4LIq+nNdqeydj+6ROy+0tK/82Xv/yoJMeRiEZfA/sBfHKNbqy9RvHaAcZlNKRJjguOpSx2sGUK - A0XOyqvr1rE+2PU6Su6x2uz4a8LAYlRMgzwZsdNY2j/8VN1UsxemFWWomfIX+Uah5Xxra74UhN/t - b/wbPGucDPd6Jheq7zsoOrEAux9CSv37ReufHm2vjkqCC4Yq/vhFAjV/dTG7uLNKpaLapGnI78iM - Vw/we14I518loHTHt1WtHQxxuZ8K7OVX0M/fjwKPpfBF3ioh9e/7Je51a4lFGTrOgFQdqIqKQ9fz - tAH66tMEvgt4Jy7UIaCJceXB+rkgpJB+sH/ikNZwFViTXN8fdaTAWVopDsIEryPl7Y1XORakcRQQ - c9T8ZmXq3AS7viWOLVbq9kyaFj4UbUXW4p8bChyhBe+CuWPC2Td12f01LFixJjLJM5vaVIphsb0G - dE2Bah/za9vBomnf5Nld+pG8HSOGWgIWoklvT91eDAdBw7dn5B/cMDp6njvBoxN+kFEer5QPlKGC - MtFCPOqNmh/ntmGlXi0fRMuNqNlY9tJCYYUOiV7nS8RqxoeRqpMco1cnD3RSJXeBRu8lxAo5g/J5 - Ow/w00wOuZXQGCdBXAtYpA8V84tJ8i1iq0z6VvwduVAvKKmDVwb2fIj4XXHKN3mivnQZddvj10Gw - p7AzTHgZNZt4SGDBNk/hJq3f2vCkG+PblHk+U1ieXgHmf/J9XNWuMqVe2zJ8VPtTM01PTYTyyTwh - tb91DSb8JopaIGqkDMcp+gVWUELp/r1i2OIR0O9z7kG3Yh6zRmhSEs1Yg6dC3oibWqeR7nwC/eI8 - Icufgp1P7Qn88cv5/MjpSkgXwrLCEzGCpqeLUWAP0rYP0Z9+mTTlJoLSWXrkP0iXL8+AY2GqaQbJ - vxNS2bW1dajWhxnT/taNK7lrOvyUY+Dxf3xWTFgAJ954EOfZxfZ6a38bqCxDIvbT/owrU0cmyAOe - wRu/vCl/DNcU3n1L85IX09prl9qh+Dq+AFJf50vOCVkfw2GUbIIsbKobdzcYGNX3AzJYgbEXB3rd - v7zyz1/R8mgoYmxkJ485BItNmffCwl651Vhs6bU5/qyogg2jfIg7ZCbF3s2UwV9eGiSF0izmmw9B - aMIZXSNY23u9dxCRu4qHP3/CSqsurWqn4TjH53FJLFsG+J0eSMF6fD6q7zqFOQAtMfl33/z5X/jt - fYDikMhgy9tPDziH1Mj0C9yQ49Z7UJFIT6yehyqJN96EfqFOxP3RDFDxEy5wa52FyPa6r7CNhHbf - 85njb+YDsK2vSyI2wdsg+nw7UlrKWwf2fJF4qv0Gy+McyPDJJZq3vABqVieXSrD7XaSPVy/fvJss - S4/pekc6+0TNrI2LJyXQ21cY5Gp0/Jvvf3mMbQXcuGT6T4f+fkYt62XySOvglQq9MD1RtvjvcffT - G4SDUJG4XW721l6gAx8xpsiss0VdEk1kwOnAcMRixVole94igoclIf2KWYrriAnFrNvD9H0+Db6U - ZfAuegNRiqesbv2+J3jIlzPx8tsEsNygBVqfRUU3rr3SbX25MXSCzsX0DBx7Hra8hc0nlP/lrf/0 - vJzfHmTXX3Td81Mx+1g3pO78sI1iwYK7b2oo66AcLSf53MGzqiRYNFJCF+X8DmGARR256e7oNMPu - T6Nxp8jl/JAuzH4KuH8aFGKcambcsOVqf/yGizyzmuP+vP/yZKIrgkU5sZplGET4h+HuD6cuGyxA - myZHejioNvuZWV/a8z8s7vjL+c+xA2Vv+8QNfoTOYW9PsOsZn+zPK+ec03cSbwab//lVuu16HYbD - S0C7Pgcb2e6lBLKSIbs+j1bgZiIsT8/AO6Hvmf7LIw+JevOEzenzLXSt4a//hnSp8PN/et950zuG - J0tr9vxVPF3utxBdix+Xb4+yZ+DO/2jvh0V//gJM7knY82W12f1kDBixn0mg1lszZt3gAeFWzkiR - WxQdCVso8P4u/jfvLpLAl2Zoin/9pXwRuHcGAyXtyeO+rxCpxpGF9FGyRGanHuBL0CyilqLVk0AB - 7e2zDcrJpbWILtn3G63Ph1WKf3jnnCxtPOb7W+G6l0axyLDPcbsEzQZHXd283Q+pS8ovvORh8Cb2 - sfxG26uLFvhM+Q857/km9l+9Cd1KEIhGhakZJ31t4QUEN++JTF09lo0cQnnFN2ISHNjr/PFMcBB+ - d6T77ymnk35qIethwZPWuvrz75b0168rHgGlK5OSHriXOcOrRt/5h9F6EV4ei4UeH7EBy95fkvb5 - 5HGs8rbpl0csVGXzjVf1uNF1eJxScJKa/a1a0kVdPr4lgz3Pwqdj81GH6flyYPQiELNKIjb/+O+1 - 3lKi7vnfdn78TPjoFJa4Hp9GC54kDTKpPKJgxwcW3YYE9IutINSXis3yalNBZ0g2j1HtM92mQyvC - VrJ0T5oEWeUKXtb/8nQUXdWGzu6wyrDYngPS8iKibP9RfGnPw8kZxweA9zwXiibJPTo4ZNzOMxGg - 3+9vSZrGxV7uXy6E1RE8kdqrAl2Nclj+fV/6XEizVF4kwLozX0iFr7UhUij5cM8Tve/wU9RpOcsL - XM+2g+SYCmrv+WEImGQaUBjWPqWOskD4geW657lKzp7ScfkfAAAA//+kXcvWqrwSfCAHck8y5C73 - IKjgDFAREJUAAfL0//Lb0zM7L8DK6q6qLtKdBPrtMlFfkxvGX76rAAfR+dLjsebS1XxXCzT4/EXP - sGDsh7cKRp9lh6M45FOGMzmCRNNtGlDMl+uLnHM0HjofZ9hayfq1pg0a+1jDmhmk6UqPnI7aV6zS - iFSvkh6uLxURW9si3vxabBmScQEziEt8vzLTmPRnbkKnD1V60Nal/fG9gvDaPebMxAcwCZ8m/6u/ - EbLvFhFrP5nhIHrfv35TuQ1ZUsN6nPbY+Onn1oubjayHNNGsUWDJpsD9QH28vufZXFC6Yck7weFm - 6vTYLi359dsTVF7ThlrOrven8m3W4Ld/Eokiav0lJVkAw0cnUPUBKJldwtnQdPQ3tf2GpPTXn1T+ - j4kC6X9PFIhxvqfFIZ9+iEQRXK9cHpVCem2X4/EUS0/p2lH9clSN7evLPVLommGzJHLLxGrRUSQ7 - Lg4SfSDr9WCcIH6ezpHQPpNWlFwJQslLpGgPnqHBP5pkhBe76WmgfBp/GsaggrL4DrD1/HzJpl51 - G85Fg2YZHwJ/zQsvh3YcWfi3PsACHQrQIo5Cb6cLLjn3fHIQ4t/xLJNvxZbmK19gye332Pt6Gvit - T1XQq04ozi6rsT4xbWB4vp1mrn7GbDoH8QZhXXjUB45m8A/+NaDiQAg131QtF2lwFXi6Rgk97xe9 - XVzq2TAvzhk277KbLiTZCugkmTMvs38gIhxQDT4pzPDNtWsiysIqwDj4PukxPu/ZJHueogDD42bm - TkW6WW/cAZs/bNhHTc62JLdsmIRxSh/X3CScq59NFJ2XiRaZkfsL9O0actxDxVFbe0zMgrmHQgU7 - mlytFozgHnbgnQ67iLN2lS+UBWcjeSB4fsD56rPgctjgrQUG/r1UlnJ7sg9kUXp70bbHALCapx2w - r8sL+2VeE6GBygWG+wRj9TF9jZmBZ4VOs6Fh57B7AGG/VR5Mp+lDTe59aTfyAgX04Z7S6Hz7+Pym - bR063aBBQ0O4lYJQFRFK6SmnlitLZBUXTUU/vFGrUvVS+MUDCsf5gvV3qxti+rJ3cHZEPItuahOx - E5ADXe4a0pvwENlinewOaHUqUeu0Gr64TFSAV3it5h36snaylKaGT9bx9N7TwWfdcDXBzd7bs9Cc - bMY2rf4gENATjbaryUThonkgzbCPXZ1o/iIv0gZm8a5QFQlH9r3Wa4zesxxQi5KxXEd5iBAdaxPn - URkxPlPVDpVR3v7yYTCxu+EANm8DRyR2tXYBnvIBINu3+KCcvz7/JVIPSf9YqQGA7G/7QPfgREBC - o2JgYJvDUgULOBLqvvArJVbY9lA5PV36CGeXbdWMdXAd6BwVEj8azPsoO+B+nd28B8/Jnyd79CA3 - nreou02hIVTPBiJo3MN5Ma8pmMqn5SBGLyLWGi1kwp06M0r1oqKOzHuGaNW7BtzpfZihrA3laE1c - DfdyZkRSO8Vgvj93AjybzwfNDV81eLldOPTVe53iP/wus7tAvrUJjvIjYeymEQg1qbSwucMIbFki - fsCiXTeaxHgE0/iAMRzPVoUfRuyX4k1rISDuk8Oln3rlYhbPHo3t4GLnbsvG1h8LBY4uH1Ljptbl - kvobRN2w6hjPQ0+23VG8oAbd53ksJI2I2zMbwGU3r3MjaCpZ2/gZoQddDKpJ7sHYUv+p//ENP07o - a2zxUT8hUgkJ9Xz2LZevtDTIu7hytEu+XbrcwMCBn778fFroc7HYJ+CM2UQP6cUtBUSuBfws0gln - xctpuWerV0Bz14Jix9bbtf66PTgkGsU+gh3btqju0StjCDssL/3Vqrke9bl1wbnDDi11waZCeVtu - NJDe95JfDqUCg4jcZiAfbulvr7iAegSaeR3Ejv34mEOb3FN8+1Rnn0GzbVBgmQifd64HOJtrc4iZ - atDqsD8QPp+0CoWZadKHLjzKtYanBrlBqEXfoVqNCbfiHXjsIER7+zG0FK5xAwVV/8xvLj3/7rNC - HdBFa8amqO7IfB6PI+IzrqJnb66NdT3QBa6k+d2adVF9Ibffd3RLwjM23bAsR14fLvAB1IRmu0ve - 8sElFGCeRRca7s8R2I6ZHIBoGwea5YHacmg35MD3sUE12OktU9DfGTJ1wtogmozX5NMOKuNwwa4C - dbCIsWqisWwA/cUb/EZ+dBjX+m6WdrHViiAbekDmWsV3sY+J2Lh1IT+4jtGT0XUpC4Wuh1OBO+zb - UsTm5X3iEH8sEA7XZDRGXzAFuO2NOz08CwDWsIIzUE2SYFvv+pSxsrkjkZCQ/un3dC9bAZ4EB2M3 - uZB07URsQv84tNEO82orjD2OYRfyFi3YnmOT+1lVyNEhpsYyFgbdno8PnKTlSB+XTgWcjCoBSufn - gXq63/i8dNMgOiQGnf/qLxfvtBjRsTGxXiSYzD98gzmGt5nb4RvYAiFagDZ5whxH5cEXk46ZMK7V - HdZ/fOG6zPb+vveP/6vaeyrAyp5hvTq1xrpwpwXZAej/4g8EWXgrsH8ZKY6qBKRMtd4OmMeA4IvC - f4xZvXomwgpi0W7HY8ap+1mFRF4wTUrzARjv7BPYhu1IXXEyCH87cRAxzn3jwguOrXjOdwv8TnKE - 8xft0nV5/maW2LGi5SRGhDschBz6scZwNiZ9KZ7yRkJfdZ/Q2H+35RZ8mxGR9rTggrQum9KXpSry - 6k70CLumFbJaa+D7dRmpjcsA8MW79mDyqh7Rii4JYI1b54AyvaEuPgTGO5UiHbjW2tAkLL/GSqdJ - gameV9jaXwSybdGnA5NBD1jnVAj+9AqGpcTwcTpCwgTcO6jzZpN6t/ptbEPyqdC1GBbqqwIg09vs - FPTd9xCHKUjJkoVqg6o8K3F0vYtk+eTpiO684tGk+Z3xmBVHR+wD8l+969l2i1oJCstJo3Fu7gwm - CyuHMsYp2JuziTCQNT18msjDpXSejEm+3Th4NIpi3rcqAVt97gfgV8qKA+Wj+9sTr3cU+HWFfSY/ - AZsmx4Z7zZEi8OPXcn7BDWp8bNLw3AKDxrflhHS2m6bXYn/88fvQe3DhvSPFVExS2t9VCV3UwwEH - xZ0nFJteDm+facPaNw59sXw3JriK3YTVWuGNrZoPvw4t/6Duzd+ReSmkDZznSxEtZRKSVfZ0CZax - mUSibr7T9RKMA2isuKXx6e4ZC4uUBq0v+/2Lb0k47VqpMJM6Eq0/f7Q58+Igzc89ersFdrnCnV3D - j558qduUT/KXf6T1NJt3VVKWsxO9Iyh8Aoa15RGkDK1dAUEn1/jwNd7+toSzBDj6iXHMRV7LroV5 - QXK+f2LnSm7kp48b/OlDBGvQkfnpXiooWt4yc+g2kYX01gUWp3qgaeSc2TLZQwDd5POcmVijdnE/ - 1IRHjQ3USmGZshjRC1h/70QqEh/4vH5dOWhHzQU7Mj8YWyQ6d6j0kUi9ijzIuvPHCxTa3KQ29afy - T0/BjKQrzbx6S5fwlvdwuTZ7HG5yC7bOiz34WZTTvK5jbiwkUXKoas2RlknzJUu8cxP4lMpuflac - brAr22KUCXtAXVdRy39+aVcdpnntK7ccL/JhBr98zKs9fsBycyL45yejLTXidLmfYwFhs8ix69an - cuHO3xze1Od73r1BX07i4qpyYNkIq8fBTmnbDhFkytvG1g9vy+HZnWStPkrz8852xrRpSq8cnDzG - 1xe20tVr7io83JWJ/vmHeUjqOyIf+xntlm36lz8YR8kRO+X4TMf3ceRgILVZyETFbYVyYg4CHajn - 9ivajLdq2P/VQ+rpKCl//t4E150eRM243ghzz7WKnHx50/R4eZEljKMZXJ29ToOIU9iakI/6V78i - mFsT6dQxNaEFsESNtsjA4uhahOKOn2jYcGG7xUfvovz4MtNyDYioviQT/v2vpXPSpIs51R+w3y8N - zu/1yxg/m/uBf3oSaL5pcCX3scF5PhX0rElfY+2y2Ib9l5nU/dX/7Q/fjN+y6CumIllHuQlgJHsu - DgdjBNR5/gcAAP//JJ3b1mswFIUfyEVVVZZL51IqflTbO/RI1TFBnn4P9guQUesw57fSZPTkya8c - fEi/dceW/EXqtKl8ftEDNH0xHwRZ/1D/aeKM5STg5Ch8pj6nlUQf3lR8rPFFo+xmR+Pdbzh0Mq8J - DZzqWcyXl/OBi5ylVO+7YzEeUBqi175HOL2ZbrfDF/Uh85L09OGwORQ9j/0r9PqN0ItIpqjeFZEi - P5SdglVdNF1y0gMi39t42eGS6QUPJf+QTRbY+PkX39n0E+0Zse14oUlFtYiVQWHDFF33RNaNwV30 - NoAdnm0cXb8G6saB8ugtu2dsaxPvsmFQDPDP8+DfmiBmS71I5W4/Y6p/tWc25jwDaDYl+GOgk2KO - 3EaD943tyYY7B2gbm28RbrtqoKqwTHR09pyhLfQL2YofprOPbqbr9yG3M2wyMvt1BbO1jfyNIYZL - /9wFkBSPgrrDjhSsn+UQdTkfUsNierGtppsEkpG75LD8PuQDUowWfebvuNHo+nDaW5DuRAs/7Ert - aDC5AWjZ+MZO/zA6nptTDwalssirGT2X19iZX/UbgfzVuGMWP20kZlxFrX42I7aoMeAuAUez5y1w - 5/EXC2hqUELVUn9nLHl+W2Sik0jYbR478qwHIr+eaEPEccP0aX8/84BTcYv17VvtZvIKDfjtlBRf - Ua4yPhaZIHs31BP+Y9udUJvHHsx3meL8sPl1XddbKfzsRX5vU78b+9s7RH7ueTjKqzgaDlkWQCSV - P3oyxQD1q55e4ov6v3eajf0GhfAjyMPqhWvZzFDzAIdW9//+gR0yJ4bvk/dotNRj6gnXHjq1NPBx - nh134jO47lUZ83jxyx2LOV5c9T1VTvjIdqQaLDCev5Ca2bcqpibUWuj+Gs2fhr+8m3YNUdC2eR2x - Lf2krF38AlxlwrChTGNE7ezuwVKvcSqpddavfukPlV9qmBzok9t4ErKii0WVftNlsxHxrUxxecC3 - cf9yJ3HMRTi0R/BblzlRZf5wBet6FaT57ugncgluoDO/YaGtT35QfmRbekfY930bseHJ+UjeeQqN - L+1RX/iCj+w2P2M94Xo2/67qKEe1tCcQNVq03X6KQD4m7h6b3pBHU+ReLMmh5Z1aQ3nuSNCKwfo8 - fI4ARWw/Pnio768dvlXqtpiS/a1FsXDEFDPzs/pnW17Xz5FxX7QrT+LLq4bXeCMLv0I/txhJvejF - hQ98oLKymmoWl7pscx1t+ZbWI7X0aXbZYdlNJ/V1gs1Pukez33ClZNfbLz3cb140V3+hKC/P8/f5 - zc94+Rqk//PVhbHuaC1q/3kXNs8FiwbFyw30RFqIXVv6RMyjMgHTqO/YTeSv3phx8IJy+nvhI+XM - jC1XNEN2kA3suV/UTYKuHNE+Po3YP37lbnTzRII4nBm1Kqpl//2qNN2tVT+606DcbCiPvUHDqfrq - swfvcY0HMn7PY8F01hAw3pFKj4ueHg/8JgTuEnLUxqeNO6287vCNTzj/HF5opEzgYNHri19D0fBJ - Hun/fAyeg6MLvHZN4SP2st+c4ZlNQpMe0TtpRf8bAcrmixWVMN2EKz4erickIMPs5X7wGqoG5w2i - LpxEmLbOF9tqWUdk4QPw7p4/fLJ7vpinw9RD9h4VemXXTJ/iDklgvSLdj+/DoE8eZSNsiMVjdyf9 - ZYy3uxQWfYFxoXRsMuPgIz9KL8eHPrSi+V7fBGhuxhOnnyFEtVP/jNUf4MM4vBHVB6uXdj4N8OHN - /9B/PVnbxKYOL+FOwKWTosfn1+Nj3AZuX02YA213IPgoqXVEcv+qrP7M35oP2Z10SxqhmQeOnniL - spa7hvHa7/yJl35suquyiK7mp8ZuVeNoNBRWg5o9bcJVtNUncrAV9NVkn4wv20D8ogfkYq522LiF - fx3vFea85ic1MzftqACRLR+5yCJbKzKK2Tr0D5i6l4Vj+X4qhL0+8Wt9xav+FtwPE9DRMhDF2dR3 - zG+Vl7zwwoWPvt1BjY6AlnpA13im32CnyS4GHkd3RYlYyDXKmm9Uo+VyVrx5JCA5gUQTV/nLpsnX - Rhhvrw3F/OB0k7Dl/FW/Y3fhF//5YV0ziy7/AnR36/uNP1+n9pcaUV8kUohi9WD6U85r7pZzy0T2 - f3lG7Z7TC7prcgP9XbFD8un6dGdlHhV58bN+07QvNCRe2YK4STSfew0JG39dOkth/XtRFdsM9cft - yMF5XzW+7JMfG65KnEMWnSJ/3CopG2URXxGXmwPFoWNmbJecNailnKeq8/I6YiE7AT/3PWxs2cxY - UiktLDyYKlNqsV0tHgnIaaBT/TYHxfBxpARuWlFgRxX2HZMvwgNdIyRSzVFOxc6Kwxyt/WxzNLmi - /95lDTJZ87Ep0wcbLyx8gZY6Ez2+kVZQtXI5EHKupO7+7URT+nvZqP0xm+JO+DBhTMURshKpdI2v - xY8fYXkf9q1N040P7TbCwoOpjpyazfG3FSB4XPDi/19dd19uWUqc7kd2hui600loW+islP3vp9tX - 45TgPy8nf68K+2LyKBpR8jdvybj4TfrEai+N7/qE77WgumNVdhJUDrGwcxB7RLnQCZApZA494IHT - CTlFmrzqbe/I3Gi4k1b8318u3Fh2C0/J4fjwbapk4VAQHsmJ9PKWM4AaCPVxYw4eLLwNnyRbdwdf - vn2gii7f//xnZm/2klMgW7r6y7nbNDFYqPijdml7haDtpRGYMpywKn2oOyl7x0Ja6KnUVks7Yt3+ - GiJ2Kl2aXHY7NKa/xANUvQJ6WuKHDU/BX3kCaZbvOyqWasjSO7Wx7qFWHxPrr5Xjw6fEuMWdzkyp - zEGO+NKvOGFX5FdGUnA+9OmjrXVirBEfJRKD/opTVCsun8Z6CH8Cz9P4sW/WemuD+Dz8sMovR0DM - MSsRMwug+DmD25jx9QMZL2+wYtwYG42hbhGmtCRj9YFoTIe9DxX5MapdHh2brkr8kE3fl6mnynE2 - J7e0RsKbxdhoLj82fstAkZd+9N+v0W+wUaAY5gM1r/PczXzdprDwUqoHsC+mg23HIIyJ6kuLXp8M - xlokglouPCRnTAn3CtJjtaPe6Uf1cdWfj0Z1qXq9VV0jHZMZWtxJ2DY3OpuMg1XDfvf16H8+9cqz - CjmS0GM3QBNb+JQG6zzH/6Fq8RNvgKOz7FDPZynryatXUBlfCmzrchDVaz3VP9XV3xw6i/3vL/ou - O5ABv0T3vx9f+DYZbZ5mVd3FPdpDImOdGmHBzBf3knI7vGMlFD4FyQZ2lIsnTqkGhl5MZQFHULeh - Qb1pdvXa/UwiWuP1JB422erfoFeCAV+Mx8Zt1Q3lpMjlgMyD6xTr/A1F4maHrb0fZGOwTLi//CBR - N5FNnT7skEdPjT9gqykqdyxkI0TFXO7oSXjuELvexhyW+omVxHwzlv1aC7DGamws/WAL7jVf9ZC/ - n0Vw5/PzwEO0uSU0N6ar2z/8LQ+beV/60trfFr0F6mALZG8+DbT9yjIgeJBsrefZlEnMAtWqb1jb - +d7if08GzK1PCZx+2GUnoa+QnaGawO2vQf39DRJyLveAat/PI5rW9XCbp0MV5VpmJNjFCbDSaKh7 - /fbFLMpHCWli/6DGloVM2LeuAJLxcKn3OShslksuRQZ+xThxbrdlOu2/YO2f3E3Uiy1M7QNAEH/4 - 2eUCG73kTSA+Ix4fnjMfER5tE1koUgObR6K4wtBzR8iEkuJTnW/daeVjpP0ifPR/g06k9tL/n+e4 - W2tg09909qVUiJLV70Wj5f9dZXFoYzLuMyEatRuVYOmfZP7CMZsK/RmDPM7v1X+wycG6Ia/6B2sG - 1nc6exOZ3DsOu19117Hz829c+QQ9Ckws1nkQ4BOI2JP8RJ9TZQ5lv/3lWNEmLyN6hXpwPKzi40UX - 9dlGsScfnRPQ6zM8R31sPgEKcC9ETE2n2y31VdbtKceWtdyKGn9fEuy/+wfWdy+563MecatfxNbf - /ZWRvvoHAAD//6RdS5eyPLf8QQzknmSIgICABAURZ4CKgohcEiC//izs58ze2Tfs1a0NyU5V7drJ - Dg4hfPsKtUO0zZdOXgqoDdKNsI9ReezSxCW4Xy63YIa13s/xaQchyUr0y8+qOX8eHCgq6hOv12T0 - U6m/2z//3zbvD0OiY87BWiIbGqRHjwnld1tDaBQHahquHM0MPEpwOD9iUknlrWJJ43RQbANGNUHQ - 8ylR7AFOYgFxyh9egGb24QRJyN1+fls1vap7AksoQWpLz3c079SuhBZyt9R5xi1b/c4CrvkzTqIC - Rl/r09/VK86e1PzV11Y/DeShdQrkpekAe1I1U9f6JMaaZOSC8OpD1ZtOwa/+mrfDhoXwt15/+mvc - nKIA7QIf0cSYT94EJVCCjvgfIpdqwab5Qya41g+pocu28aefrt1IaFhXZT5fNKcB8scXqG0cI09I - Fq77q6/5J93tGe7iDO4vj3DF51PFVr/kf+lRoPz3jgLxPCz08r541ShtQh9ePoUTKPmg5eS5uB3U - oh0KBC1CFeu+CYeOD9PH1vyxmPg5myKqw3NBD8VW96Y+/sbQqd4SNdSDbYiX/K1CfXfa4EBgViRG - 9vhSn4GYUI+7Wj170sVBQzF9sNEYQkXHYq7h/fINsFcIblXhz7FAun174ET6IsD0RXFVZXsC1E71 - XS8Zc6shNKlxIOSCni/cfVrgrPcY65s2AbP/yDh4ccxh7fJqMl7j1zNzYvAOlMZO+mVsXhBczXcf - 1E8isDnaizpShe5GHmufPdEJOw5MXFPS83dXG+QkZQ0U32kZCNR/VCP5dAnUyrdHnUU45/z0SRsQ - 8rsZH3OuzoXL2c/gliY6PZm5Gs3zUHLQOKu7QOSuNVtKeJig9bHkAJqJko+84XVQFdobzcuTXol9 - 6MRIMW45jTx4zPmz7PgAPeILtjuwMZahjAKEzkCmVmi4YL43rxJs7KmljvnIc4E3jA55tl5R/dZc - 8jHeqCI81WaEXUEymLi1RR6ameph09gHBgWXbwvJiVjY04mVC/FmEdGsFZQa1+Olny1DGaBReBHW - lu8ZiK9pfiHx89bo9mrG1TB9wgZtk4NAr9amAcKyG2WId6ZGzcO2yMWhzAN00zcZ1ji5B6O4z1P4 - FaIDPUiR2s+pUaboeS9N7F1vF4Pf2XkCWZIe8P1dh7nITfEEgy106MN79hG7d6wEanemFH+r3pvL - tUtk63YFEZdRBLOHNRPW/npmtZhHY3iKVae+BSOke6koKibnZw1txrGmGS52hiSaago337jG+Whq - /RJ0wIQULCq10abs55cfQHjVgh7v6k/Zs7HO7kg3rMeqWC2DN/aohlrBE5o4E2CLaLsyfKjui8y4 - +7JlUfsEpscHw7rHDr0YKrwIls/ZoEkpOIDH73Q9s/DF9Ord3Fz81lYtKVOyo9txv60W5nIniLaC - S6NocfLZycc7dLxpwNtNLVVka4siVN7GhUjPIvSYkhkqGob7mW4vpcbm/bGLoe/rX7wtBdugyvEo - o4PxfOFj9N5F/KW0VJRm8omm7xJHUjkbJ9i7xwgHL79g4+XGpcBnAk/Nzo4B2wijCxuIRaqZ8yYa - WmsaUGaPFcXvFvdi+74M8L4dTvg0HzQmvcIHhGcMRzJ9xwDwHKILeCvrHUtYLxlNcqmFu/hd4+QJ - UiD26RIjjTNTnJINlw+x7mUwdNMUuxslY2Q2yQtSpc5o8PIh6z/bMEaG+7kQNbpoPc/w5YQ6fcmp - Jbz0XLjLVqn+xlvXyjsbIrW7wxZNMd5LBawGnk0WDITshHdNKFckAZqJMnlzoIFKDzm/eTkdWM6F - gG1xv6/EsE15sNcmRs5p6vY81TcLhGjKafEsZjAb4cWCtivFWL/UmiHZuK+Bar0jbHjXfc4QPaew - MdOBepm2MaaPsxVRxbHHuv6aiMm1rf/mA5+pv+mJEQS1ur4PUcr20wu2Xja/+SUScM+GxJxxgYf3 - 7o7tmjeqQascHXqqm+NH9wXeIsfgBTf908R5OVmeON5kCLYJFujBP8ceL7mnCWlTqNCHsjy9+fE+ - Zuh26p1AVMAzZ1tul4JXVXr4MPJeNRl71KjTqNV4ayQ4F4uGDyGQ70YgHHegmnveHqD+NXLqF1yQ - s1mpIHoUm4YaODYB1Ub1hSLaK0Epbr/5hPJJBUk7hrRQNq9IinY3HcbLrST1YQvzUZdOLsIJDGhi - EbOS4AeWAO9vGr1sdlNPXtNcQrJ1XHwrs60hwGx+wS9uX9QJwxGMx0X3Ufnsc6KkuK2+ZIoTVM7n - PXZudhhJRa5MME/1C94/v5Enhdo9A/GLLdSC+s2bZHt6IS8ZjvgaxwdjfR4HwfBTBqpnNMbk1rtB - EQUzoMfQGfspVq4JUN+Q4tjYB96IrmWGypvKUdtjOF+6F7qrt41oUFv9jMZ01tQQjpq/4MPVF6ul - 75IYvsBGob/xZm788lGiT3HAr3g6XZt7qzwDPqHn+Ea9yU/nOwJX84Mdol4BL8Y3GaJJjoP2qfeM - Ha4tD+vScAnnX1nO4vNHhcIu5wgnkOd6z2ceqHsxv+FdCk49zweWA3cmEANgfbeAH2o+QNtHF2Or - PGlAmN6aCWqtJ9hO9Xe1xi+PJIsrMBakXc9Xy3SCv/UBsF4COhZKDb2JQmzOmcF4Jux91Oi+T/h0 - CZiAej/944vTuyD50FoygXJwi6l3vUneHJpSCr9bwcDRy8kMsQ+1GIX3m4UfrxB6A1ddJ4iEY0xD - 4f7M52A06h8+4Z9eECTHDeFhe9HpkbvbPT++rhBVh6eAY9vc93wgB5byFL9Peq/vZSTgXi1gXLwq - 7HX47PEbVnfwh8eeAgyPvwSdCkxmtFT/qkewPL9pB7P+sSFysdWioejPJ7S7awbOYsDnS9HAEG5y - KQmQtr32c/ZCuvq8v0x8IqmQT9K2nVCxP22pu9Q8m50TKCC3NyZqXubKI3nyzNDKb9j7DgePP3xY - ADWSdXjf6SdvYcrJR93+DOgFuIIxxNEnQ8/J3NL46Y2MLLMSw/snhjR3DueK/fC4fDAfO06WsEk4 - Uw2ezEHC52Xbe0TbUhMCKBmkC28TYN5ITdjx+o64jjkDmvmTjj5fqwx4zn+CJVSgCFl6yvGO+utt - U2tXRtHga6q7y8ugK37AMDyfqRcafs9aR/Fhdg5m6oUN3zPdPPvwL/6st8am7rwU6L7NNLzNOTMX - XoDX4MrfZNN9c4NM/beGK14S5MVOPm8VmKmHpRnWrnDMIEe5ItDfFCwAZJSraRiJrzy8XUpvM7UN - ZoWQoPK4GalJ3mkufHZKDKlbhcGsFZtqUQcawkRf4p8+7tnudOBhtnXWrtvLrmc//uzgjWFz1Y8L - rzEHRtesx1tVM71x6MwGKp+DhL206Pvv8Zv6f/rU++53hpBlho76S4FxutlTwOqEEijJvoWx0fNR - E3TMgnPo3AixNk7PetqQX/zT+Jzl0V88oUdyoW4qUjbN2q2EXJzc6T4S99HyDgRTdaPmSveuqueU - dp0PgVwYOItUpxJ0Op9+epRazgRAHzymBumG+aDY6ON8fvKRC8s9H9KbDr7RsovvyU+fYt0/HqPp - ML19WM6XPZnT51DN3fsA4aqnaYoX1xPgcH5BuW/DYHVFKppEuxbmg72jfqnsgZhPmxQu28wm/DkD - +XAy3x30vLwiskDf+QKui4MUbGbY9vnc68ruu8Dfegqj4rnuoCkG6A3fw6ofNCCicqtC5qMRa7qv - ebz/NRN4iA4htoVm24sfo+Qh4UOdSP6RefRrtybQd+EGZ+r5E02HD/ABG857+vt/qx7NUHHot9Sy - +bifNUOqIa/ZDXZa+cXmy01M4bQ/OTQQ+SMToq2v/X4O4hl5jA+azoKB82yDBXobbzTErQ61ENwI - C89TPz3FvoXnZ9nRC/Z0wK7B1oVRYCNq+Q6OhJUvwAWUYvB6Pk+MrPkUWPGFWtzVBPNY7u9w31BK - /fflEE1PS2xg7VqH1WHRAf92BxO6UX3Fhbjvo5XPU/DTO4l5KcA0RKEP9vomwn7N7yJewdcUvABS - 6APkirEs85zAfo+2dGuRF5vOp3uo7oLkjC2/v+Qi7A4aQJL3wnsxGfo54GgNM7QFeFtPej9FRlGC - lb9psFbgm6cKO1ncwoS6emd7LCdEBYfzdVrxbDTmPJ1VdMbcSK+FW3vsG54suOIL3Y/ZrvpidC/A - mg9QSw+33kw1GkLbKmL8yKdtLqx6GGzj7Zlqa3ws/H50oJeQI41VPogGTFwIR1of8OVZhMYs4d6F - J7280AK0asTYhhH005eXghv6pX1fCNxKbKY69p7GEqw9StZ8lgiKpHuzessn8HqT/k8P9qCMWijT - gcPG833yRF3Yd+rKdwH94dVGQh1U7otD940t9s9f/IxC5Qbr/d8RweDAAfmpPOiPP1l7rNYdycAj - CinP1fzkcxeGPYxpnmkPj7zRUYcm130CdEwe/SzvWh32uYDJkr57sOSTlEI1Np9kWr5rfitfORAS - hyerfogWubm8oBXpLvWs7xO0WIQcWL7PihqG5kb0sLwJhAE9YhOntUfI+8hD54B8okgF7Oe9IydA - tT5RgGZqe4vIhTzIpz2/8pXhCcdvGoDvIwmJ/OVOrFNPxxNCVW7R3covC+XuA8gu1yPdrvnysiWz - DF0pFrC9aeV+cPN8nY8sDRT/nhjCmfsOIC7KCh+ETu8nXQWLEuUfD+tXJ2KiH4d3tFG0UyCAUvWm - 9mkNUB5fd4w7KzJmXDl3KMmBhQ11ot7SetsBqtWC8aFlpTFdb3ULx4f8oKYxJN6SOvICW5LJhHta - A2srM47h5XN3AnalwJhjyRGh/t3mdNvFotdTVvgQFv6DhioK83GXNDy88Qdz1ct9P0ubMIBSW6b4 - ztkaEDZz7KAN8BdqkIKv2KOcMvjLnzyDF/PBfmALrL3tAuajm7f4op2oNSov2IiWNm/cZCGwSuAZ - mwMs87mWnfbPb9F9/hotHpUa+OCqF5m6UGTzoZCJGm+xSLc6++TzcJ0JMpQY//HTwt3lCT4skgbC - LZ6radbOr7/537/roB+hgEugvjkazP7h5LGXuFWRBn0XH99RVw3Fqz9Bokov8vbN0BBma6PDy2XJ - sO95dTWJOTf96TmHaz6MCZdPiZi/GUm62WPQe9HVghLHmoB/W/ucJdGhA59ruuDi9hSjP315Y+qC - Aw/QauJhZ8ELeIl06whcvzyNZkBcEwzUWlzChtfQluAaaC4+PlNszO7VC+TjuR9Xf+rjzQ0XZn/5 - Mh9eMaCxuXbps2lBeGNP/l/fbyDA7vUqASYe6hS+QP8h4soHJIAPE4J9qmBLve9zcadqJ/RIm4Y6 - TbOJxs9ujqHmLBo+XOnaUOJhQyjfPR1vsf1kAzd4C+S6z4B3q7/UviDLwBhoa9NBM+qnW+c7cPVL - gvJ6T8G06k9wnNp9wLzrN5/MjZqBVS9R3ZZP3iIlZov295MW8EC953R9XmX18+ixvBiV9CinFJnw - blNzzipGHv2iIvcrfKguDhhQ15N91bhyjO6+O9MTrPNtAGr6eP7zeyY+b6DAPxyMU/HYDz99r+y+ - AJ/KXGei+xgGwLzdAzvpd2CsVFAIA6dq6VYrHv38AlBTn2UWYT+0PcCfT0mIUtIRurOdFvT5G8pw - GxvndTxMg415WSCpIzLGpE6qeZzHApqEU9bxfeTkDK8BlFHg0yCURDCa6Kn+9Aq+YVyBqdFaDt7C - osX5j1+up86FiSZbAfT9MGKH557AV6IjvCMWrqZ3lJjI/fR2MJnzI5pMSUsB75F3oKx+jPDTc8bn - vqWOuZXAvHemBP78CYt56j/98uefFZeP8eW+LERX1gpUu135fDGCoAGY5pBI1+ZqrPg2gF0dnQh9 - jrwxH7+hD2OSzuud02VFQ6LGYPUvibLqOUnt7ou64jvdq8fWWxx7gn/5xS32PE9Y4xmdK+pSL91E - /XLSzBZJexcS5Wl5hvg9dQ384Ud+NlND+PHn+xjy+Kb2gSHJta3B8/PVkdHJ0p6d9WVCd5536bnz - eO+zGZgDDLJ90mDFwyX7Ji/EJSWHHxfhzcjHP/GI13YNPTRPWC1y83hBTjZVevcArqaQu/Hg3lNI - mPO49myhrQhd59hjr1uQMQqKdoeidCJ//CFdg62DVAB6ai9zYsyrPgfYPpWk8ejkjZF85FAmowMR - nsvZINJunv78Pt/3w1yw/H6A70OwC+pVj7HUak9w4fkj1jZ5xJYZ8AWoy61LIxzXjKX2nEEi3l9E - WGrmDeEw8394bPx+n7h5K+OEC6iWpg5Y3uiqQ9Pzb9SZ+T2bdnaUgJfUKNisq220+oUcfJ7td7B+ - nzHdFs6Fn82FBEK06SM2LZ718xcwvko3j3ikKf75afYUVWLJogWGfiMH5Pnl2fcT9AncL9oNe7li - RLTi5en3e+pYwK7GBzEymIj+nSa32zlfSrib4Fr0DUSL1BW/6dQWLue7QE0MfIMF7auBw8FUcLjq - ZbGfNwGoUrEik04dg/F23EEHEhPb1mKAGfV+9tNLOFj5izVedId5mnzxYV1vs6h9U0gMPQ3g3eSi - 6fF5qlC0DJOagmznxLnGHNDF8osL/HW8MTtedRikOAna5TRG0+f6Fn/+II0F+RP91hda/dhAKKLG - Y/DhlvDnT91K9PVGr6xNYICSUsO+j4wY8FVCMbMXbB62MBqreVLhzlTEAOqS61VuE/iqWq0nFmql - iQj+ftcKWrOhmtdjj73CC6d+PFDTFX+rX/z98u9AEZi57tB1SuCMevLLN6OF1g6BZ5bKq16+enNx - aRc1hUVAuoB7VsvKB3/xo2Py9cZa1lpUUXvGmorClc/2IkRU22Fs7teePf4VQiEcBqIu2wFML36b - oeCjmNR0tJANe8/hoRLCL842SgYm54Y4uKPqkdoGkvOOTaWPQCLb+CFmh0rYwXACPmU7uiuD7c+f - ESHL7iBQbiFv0L49iGD1pwO+cF85ua/3cP/8l+KtnvM2K6IXummLSg0f3YyXJiMNnti0Xlx67BhD - TZ9CYQlbfNjsj/0Soy6G9d0BBC4BBUzWUxdKFizoSb9WxnQCoFbR+93QIJqtfr4pl1bNtq6BfckV - 2Ug1eoLKB0vkY+aFwU4WcWE6TCW1nkAG4+HdDeDG5AWfr3e1mlZ/C/zyhd0Tj96SHrvpHx5z0Gf8 - LpUSuHwuBhE6jzem1+5dwI8KyS9fA0TkUvGnT7Ge6gaj/SPiUVbLD7J81ZlNtH0OKlP4C0HP8mhM - dnI2Ie8Nb8JmvvUmXWULEJq+Je+1/jR8H6gBJixs/JivujeOadv89ApRvSbw2DrecFHwkczHAPUD - ymUZxNg4BxUQN95kTHIHd1op4vOqvwZ403WoXy9jwNV3LWL2027gWfcCrDsKAovgxDpC0v4VCJ5h - RvO2DTP4ZeaI9VqxcnbL2haseL7qy7hf2uJN1FmOFGw5U85W/4eD2A5LbElfxFiKTyGyrXtM5JVf - p023tHD7LCsaYzB4k4VC55dvkvdzOXvs8AXDT+8Tbs3vJvwOOehNI8T6Ot8jr15rUB7RGIA1P2fZ - J9Cg9T2m1Bwfbf7nj6x6hYCzKXt/69XXrh0NVP8dEdDfHdAC2aIhs6/RjK5tCtf6UrDuq6pEOWZr - D5T5gY04HIyJ6tICPM8+EnV5hPnQPoMBFKN+wwfx0PXsJUXajx+Dqqybajk5YocOPrfF0fUoVbOJ - dxDawzVc9QCthp/f+z74O3ws1L5nl7DKYCA1C3Ztx+35kurT7/mxEzKBkXU7G4inSKM/P3nif/ts - HXVD9e6mgOHnxz4cVaZBsXtGK98HaCt8jgRuqjP7itozRb1jMhrcyDViMTjyKHelHJuicmJijK0M - tr4IsWm2KGdJvmkhdJMj3VPjVv3yC1R1ZMZG8e29uqT6glRJRNhY85UlMRbn57dgzQykaH47UY0a - 1UioAaKLN7dm6f/wlu606NbPm8fkQLJ13UBOvz6QjuM3APZ02+B9LAnGeDmbKbzOxRlnnXcGU9mE - 089PC2Bzu+U/PwVuPq5JZDAX+awcHFld+T3g9OfOaPNTmEKpG2SciYixYYQfCEctWIK39Ob6GYxN - CxszG6jf3G6RmLeH7M+PyorAiCT51Gs/f5rAr7C2aX4fXiBEZYjNlT+Wrqd3sPqb1LaWCjBT0WQk - RuUJn2yU9BMmLgcXG32wbuZZxCI2cUjYahw1nu/FmItLOaE8cfZr/rqwORi9Gkabywf7HpuN0dbL - Go7zvaD+MkuMZf6k/fgP/+qF7HPiE/Bbz6s/k4vC5sv99B22S6U1xm2bpnC4IxvfN1PKFk4JFthv - 3j2NnUoG48u3IPTS+hGMq/82obqTwbDQmoi58qymsTc49BuPU0h58A5NKYOqxCOqS/gCJLnGmtrV - yYU6TiaC9vY8OdB/bhp6uDuoWn58/z/sKFD/e0fBzXYAPXqHNlqIDVOoh9zapb1L2GJJrQrVZcdj - I+3tindM4Y7YLk7xw3GIsczd6YTE13CiW+Hk54SNR4jil64GGxRaFa/ncQ0vdzAHXPTdRcLzti+g - erQ1rH+fNuDxwyjA5daZwWzsVoXuxg5AnecSjsZ11e7Cgw6E83aPi4yFgGV321fIQdqTzrZ7MIAY - lrA4iind64VZ8feUJNBoFgPj5LTrl3k/vZDqsxBrZ+HFeLtMM8RZ8Y1GV37uKWN8gAqVUGzkolyx - 69XJkFqUF6zv1QQI+Q64sPTaGz2FqK/YrpU5GPskwVa7WExcwNeHlt7ZdJuUndFxzIYw0Lz7eoar - y4Wl515wHwsvmjsS9Zbv1GgQb/s4kDdXykZLPBRAqgMHbw/eNhKfswrBucQC9e5xmw/x1zLRXZsD - emmya768XV8FIVhaeigfO8bPrHohyRgtGu5uu1yolqiDJ1GdSb3BERC1Z+Kit7CP6eGR7QHbjdKC - hIP3xv4SVMbgXX0Tqprl451flLm0y9+hmoSDhLNNw+csh18e6GK0pYeD9s5ncUYyfNdO+Xtfj42y - JiI+OmXUQq4ZiTOsFzjo45Yeg9jL+TDkfURuTYi3fnLqhbr2M7U9btMACd69nxt7lCHaXzF98Phq - LLJZxggp8hF79P1iont3Ldi9DRnrB3E9c/J4WMpbLjDNmNtWc0CVAR7Xe4zMaX56TPHcBuohvFOr - 40dA53unqY9OfdHt4QH72UmrTDnLu5Rq73Tuh3vaxEgTm5Bex26p6G3zfUHSFC2+uq2R8zPrS6hM - lx22qKmy4dK7/lqx3FE99475stmkFtIC3JE0v1cVT6eeh+jqJvQuf8dqZF4xqcktBzgIxRtjJ06N - YfmAPk5mPQLSNfMg3Nf1l8b3h8yW9M37qDXfPr3ZG6kaIJ8RiI4nRv35yucMBPoJurukproxbbxl - PxwTKBn3A9ZDpwCzNgYpiB4lxqe+CsBUftEJBEMyEfF8iatlGziWcuWsiYDE2VfjdtB5tCNr1207 - IMZ8FT4ETWfnTbMpOjGReygygOVLDpLHc/CW9h69ID5YPd0fNyxf3FDMUGOqOfV8poEJH6UF2s8G - ETiVUk5TWymgcSpifGruzJhfT3NA549VB0q4VXuWoTGD8sF80GgbnvMx5SYX5r624Hv03eXidnmb - QPLqLz5/dKMSsIVfcJCmnhpK/qrm9GSd0C9+DZwNYL6MXx9Nz2YOVClbz7CQekFK0JypaXydXDR0 - uYSZRA0yeX3vMVGQE8RDd8a7u1iziXDMQk5lMpztz0fAOH0Z0NVPttRBFwcIk6iXQFgOId5bw1zR - 913MQOJfbXwAu70nas+7C6ARzzSjQQJodA9rqH95ATvuUYrm+HpqkGRrWxynB5MRR20byIHMppi6 - yJjkJ+5gvz14eKv7KqBgLO8o9KIGH+zNpZrOzhDDBRYavQ/PmycMk+GiRNAbusOfYyRF1SVDk7i/ - 0n0Tvap22qo8mJMuX+c38CTuvAvgW7M7bLjHF1gimCcQb78xfVAXeYz7XAZ4eiV76s7Otl/H64W6 - avLwubJmMK3tY0FYgR0O1E+Tz+S9ceH1y7bYi9SqmvSRWiC+ui9qP8+oetdzaKLuUeQUJ6d3v8BC - 9WEQ3w704HtJz2bOCeDp9NbxpXR2+bJrjhwSN86NRqjHldhhjSChUAe87z4ILOFn60KpfmyJHOfU - YM9dycEtVD7Uiq553xfa6EJYsjmoAxtES3x4JSjBg45/8cXOsVkjNe54uiePZ07ffqUifMxyuktl - 3xDZeOXgaT90+OgdnJy/0sMdFpvjhZ7DKe97y88K2NmNjA2de3qsdy4hvIr4gO3ltQdkAd8AqovN - U0cNd0D49DiGgXNUsbHzbY+6C2iAVe7sdf0+jEnZdCeVlhagrnW6GWTFa+CPyoAN93w0+Hq9t3vT - NgH1qV1684i/GfxY6ZFGU3Y2SDI5Plzq+kbPY/jtl2a/TVD9Wu9F/B62Hv9JbjwUpVDFWsL5TEhI - PSExeW0wfgtxNI/6mcDeLC3yVTd5RI8bJIPN+xjRoNxWHn+KgAx3C9AIS3u7F98d6f7i2bw/UjCe - VEkGT8l4YMs1T2Dcy9sabS+xR12YV7nIHjpB06g7eF+fv2xxQy5TnaT94OIhDxVLzHlCo6ZA7Iv3 - yWCiMMWoOuZrGPlePk+VrMNrOh2ou9uZYL4BagKOlTU+hiczkg6fWkN0mDc4wCPrCRG+IpS3vUOd - Y+x7vOlvJ7TkZIcvu+6eS4OdE7C9JB6NfbXyeOXhlOikSwAfDsorX+7IfcHqegypKz/mik2H3ITN - GWAi8junmrvRuCMrny/Yej52Hh9dNg68F7Ak1en2rNr2fDrB9urxAQd4rhr1kKp/fOK2UAf8S9Va - hPdswkETSUaZ+LQBP31VfHDRCw9PLdHR/Gr0EBxQNZunWkMgG09Uu/kLaEEMX/CIwjs+8fy3H81T - rYPrs3vj/XET5Qs5ayY6G1tCo6/x9r7GUxSRN43N2sX/WP3wBWR3sSdsja9BuG4dlNj8I+Auyikf - Nwh20IvCBkeH9M2W0ZgdtE3fcsDaIMmHjfNModF/ahz4Tcgmcj/GALbfGD/U2x0s/nO7ID5PRezN - 0sUYW3g/gVWv0cMtfTG6A0YBragDdJd0jceQwZ3A0H9SIlvbLmdhHsqIfTsbW9V89aZVP8PiULZk - 0T9nT+wjA0Le6ynhit0xYl49xNDK2QXbcFh70Bi29Tc+qSlr0XwSZgs+kprR+zof9O33Kvzp3fj5 - GvtFanYdMn2rISJV9j0fneT4b734l2FkC70aMap8waQxTO/eYhcHDojJySLSbTf1THJcGR6VTUUt - Y3s32J5tHLDiObZdqnojX1QEiJ+KYZsvX4B9tnEMmmvgkjiAnbHkWd2pxa7fYadXcc50BSdw7rHx - F88sBGoH5/5gBPMxD/K/9yuHNKUn+tAqAWV5AJZ82K14SSuWSESD6LaRqK8dDr1wWTPA+gQqasrO - 0RM72hVwnzo3qtfFYMxFuuvgFhcfGprt6idCxYIOKV1qNFqYT4fCWGAeJi/yGlqXTcrmdUIgcTvs - rPNFArMP4YGNJ3qMrrYxbzwtRdPyftOITxrAikKzkDnuMvIUTn5EY76JUZ4WHD2HwyEXH+sZkU2I - arzPNQ3wv3gaRzug2utI+mVbZjWcNFHEQbzgaqa7Wwabd5dgfDim3sy6BcJ1Pqh+3bv5kvLeHSg0 - DnF2+249yW8/DaQD22CdNFU/Tc/VARY2C+GgklXDq1dS2O9Ngu9Bve/n2PNi+Ht+58OhfhlOTgb3 - pbYNyGZ4eQQLaQqv7+BKKBG3bFaaJoGSPE3ULn1W0V0+hmCIRoH0y7Bjku/seMiIT+kPL6fvZ1iA - 9N1+qYEoYr3dTx2I93lOvcfhnbM2IuLf92nI3hjTPSWxKu8CiLc3STVo+C1daEiTRPPHYRdN6hzp - MOtDiZ6bA89mIS9diBc3ofh6dXoBDeMAf+PxEMz1DOS3qFVleq8nAzFfzbE8rXxwslb8Xtiys9gE - WGNGNHP8D5vdgrRgSofrP3367poO6im/xbe2N3tx11whWDyOw8a5NvslCt4ycrABsRVdQTUY2VUH - 1sM84XzjHHvJ2gQDyKjwoDp9lBX5IoXA4/i8UC1auy5T5W2huyDK9EHEJ5t+/FdcIU+x7HWsu3Wy - Dq3StqnX1OdoOSzpAP14KvDduzX9suov9dKwM1GaSO//9B7NZ48+XPdszDquY7S+P7Y/uDXYPMEX - zBVfobfsTIwZiDNBx6PtB/xvvq7w6kILn71g1v21reIES0g2bxn7DXhG7KcHrA/X4a2x/faLvP84 - cO9ck4BHhwFMn+r7gsbpHtM9p08GnQ65BVUUnih+vu186tIjRGaZJQTI24DNGzhbkKktCGbqxP1Q - ShoHVv6gxz3GFVOL4wut8URQL577vj93HNBv2Rev/MDoqmdhMtM+UIPs0E+1bvvQf/otPfrJUrFU - HQdwdpUj9b1QN5Y0+g5A3HIXunuaHRhe9zcBqW/L1E6krlqWA5vg+WPWeNWHBhXboQXKZWyww46b - fGpOpxoZ9lGgW3Pz6heJHWN01MyU+g956Nf46SDngh01y2tVUbB8TnDVH0TQExEsdyry0EWcTs0f - vj7nhYO+89kHX6hYXrtcDjqSat+hu+YQs/F20S044jag5iP+9F8jKCfYcHODc171mGTuswWq10NB - D9dyky8/faRtywan0akGxDDWZuWf/R0nm0LOJ5uRAiQhkbD5Ka18Vt/dBIVnUAQVWx3zVd+jz8f1 - qfM1lFV/1zEM5AliRz4dDPHkdh1EXKFQ42Xe+3Z6HjUUTaFHTydLqNg27mJo2T7FznP49vU+MVNQ - LIJPjbfNoq59PEyY7biYeiJc2IwUJQPGw0vpnrXeT4+UEGT0FICBjMb7JpYnVGmpQiQ4PKtRmpYE - jjW3EOV9Z2xp/a2GVrwPgGqc2XyGUAcnyxVxUKil8cMr8FgWa8XXoF/5Vv35Nzhu3698DhclhcUm - ulAs6VrFi+6soliRbzhC9MYWazf5aOVfrI31FK36bYHOYcLU3XJP8P3pB3eDKbV2/A6MQtlwMFWO - G3yIL+eo228+KeyqxQtkXbWj6fgyHLTyFcbfufFGg7ol/ICBD6opE7zpNfBQzaj0wFtDtSrpoIgl - PJw2F+z13+PqIHMp2G2dI81FeAJUtyITlUOWUjvVYjDZrLnDrHJzgiapiuZ9J8jwvlSILGWp5kNw - 4F+/+CCwNr2eZMk3A5PGi0QZnm0/Dfmxha+g2//wEcyzc7PAzo2eRHnEcO2pkkM4OZsT3gfw3kvx - VrUgfl0xddf8s/VT7QViQrkAvVsMmN1PLYSCXeJ9sfcM5s2TBt3NgWL3tXjeaMetA79ZOlPz4jhg - OW9yDa5+B1lOm1MvLge2qLDt40CaR+DRWr3GgPav0+pvFNVIzSuP3MZ80ONoHfpF5GMVagc404tX - t2zhtGiAWBw4MjVPUC3DO7nDNV+ga/4LyLyXX3CbfmR6kLjGW/T6xsNhA2wcSKWRz7N6WOCu8hTs - VN0rnzbg1oFs0zT4dvZrg3XVEqBl1r4B07mnQTiyd+DqtwSCdzFzoTu8Q+ANuYj9D4b98jQPJVzx - i9r7JjSm5MpCOKXkSpr7gKIZtl0BIZ4gfpySiPEg5l9QKpUD1VK4Y9Iu+5JfPBEFKpYh7h+HFkbF - IcbFd//Jp7l5iPDC8BM7zbyL+IdsvhBrrIhw3s2qpMEKXfTL53dev6mYezry6I2O04pneiU6Y6GC - sPY4cpMDls/Bi+hA31sqPegqAEsmeQ6059gkoj7qbI6vWQOHjWLjy3gjxsB/whQdW/jPn1oml7fg - 6t8RgQavaIFobb0xVzJ1NW1v8OgCfUDh08J2my45rcdD+9NX2BFj3pimrSoC8YS8n59WUWn4yOiQ - 2yyYpCXxpOtVy5BZcxzFh6PsLSQLZFg4UAuAvTBjedb0BCI5+2Cf6wLGzzvOhEztQIAUQzdWfzSF - qV2FVEu4ATDuFco/vUe302GqBnXONSBzQhBIbbpEhF6NBD1f44S91d+cD2FeQuGbyKs/Nkfzhr4z - GExdSWR9l1bLqUAubLqbhLHO3h5Ti+sLJqkQkd/zkXDtLnQKjqeAIYrA3PnPAV3EsqDR6s9OUoBc - 9ZdfifcB5UMY8gEymI8Jf82QQazu5sNDvmPY00cpZ8LO83/+dVBYnhUteTa0YObtPIBD6ObLJH9f - 8GA9Pfp4C3w09N1wB128NQkA/gPMdW2mCHF35ZdPrv4YHcC7HR16gO7b+MM/ZTrvSB67IVhyJ3Og - ZQc0EO5b0Rt/6+U91QPVSXdks/EUeUjd6YL9dN6zCbycdQcadvDhoO3yTw6/IuQ2HCWQ2pox5Mcq - g4/cy4k84zRnKt6JcPVjsXMtXTDnsZ7+4om6SqxFi8l/A7D6h3/rQfqeqhCGuTbS2/2dG1KSYh4M - slHToOCCaLk8J6gik/9g//ASc/b7+1++aP9uOS0fZQH3A90FnHH8MEaKtwrla9kHtbANPPaUb/7a - 6TLAuwqU/ZwUWQgmTofYbmPTWPV3rSZmVa96+taT9fnUH37YZ1L2SzxqL2iHzR5r9PX1yLWeOtSm - KqDbxftGI4mGP34N0HaewZQ9lhZy8s6nfjp/AXt7bQzX9UhWvcXGbjQKuOLjut623tzESg1XfxLr - qx86xOJ1gd+zuKeHmi6MndlrQt/tK6M//pnv7VVfTzw0ZMVrJjhWKKOr8nKC+XwpPT5Rzzr8+Qc7 - 9Vkb84ukGcxrxaXXXfXpmQLZCT7J3g42K38Ov3rEWh8I5IpMPRH0IQB8FGY/PcJYkF59+G1EFIjf - zcOY8k3K/+nt1Z+u/vw/8fNkZP18RZPU5pUi+SrYOl/4frbOFYFaf50pvuS4n3avRwmleFRJ61mW - sT6/Cqr1xgewceae3ucsBK2m13jbimXPNk3hQmX6cAT632u+dNghcHfa71b+GKtp+61EpAopxe7s - PKu5nC4lbMvpga0V73p42pXIuc9GoKb1O1o2x1cHeD4KqNd/Z4Pdw6kEpm829JJqPJsuN66EKz79 - 8oO+3bd3Wd0aWkHmrRGyOQHyHV41uaOZ65bVzNDuDt9G+qZux33Zn75OCjfAWnPQIn7r7loYyKVB - dxIp8uV8V07wY385jDVhiJb35Z//SgbpkxlTuc9fwPs0Z6pNPfvTrzA+ahb+q8/o9ZlHVUxaagbQ - NX78pvZf8UvNa0IYO45VBlOeXoJlzTd5/rPiZXhtf356v1z0bIGlMak4t75fb3FyPUOh4blUo6+9 - sfjP/fKr7xD+bD0Ze0nHFkqpJpDprMXecvOuIlQ+9xFvT3UaDV/b0JSf/+j04tOYgtzx4eNeJDgv - G9izkBAO6ntTDYaoKNmcWxxRV/7FRyl8e0323RQ/vUCkoX1E08UbNLD67bh4bXA17xY9gK3SioEg - ZVtD/OlHiBeIXb/gAWu1cPmrp/hLYBjL69U6sBdPAt35hRYtDRBqsP5/uh+7U8/uddBANN532Dnd - EmN5BMcXUD7FiE1bEfJZ3e01+L4MMTWu/FwxXbETZZbSZK0flGyBmLzAn/+24tlf/dA/Pp5UX/2u - JYJRDNlxqoO5Ek5sgtHEgcMWVtRYisLoToXg/tUTLEMcGKsPtIAr/wYyQbU3S8fC+fk7hK+/U7Vw - zOZgOaEtjq9JwHgYyZxaXDkeBxdl+eEjh3TOrfG2WW9v0tVlQSv/YK155v3U1G8TIuBxAfcMQPRN - 5/oOVz6i5/40VfN1amXwtXZ6wDN4MmaWlgXycmdPo5gKjNEg6aDGjucAHRQ9Eso+HeBGeesEIKvr - x4cvWj99gdeGK8ZsC3sVZpu6wcXKXxIY1+8LOJFMT6fuO65Ed8jcvYK92J3Y8hLEAB4+kYV1RXlH - K14MP3z41UcrPjpNMVr1HfWq2ylfIG5KqJa8jfftWwQT3+YQamf+jG3ps+snP3VKKC+ui53gsDe+ - 8vmSwn3dfPH+gnxDUNiiodXPo7966rRvExnaZ2fGVlNrYGoFfvWnCIe1cr2F9FdPWPEV39WPlc+/ - z2sHbiZKsfc8/qefJM07BbDT60jcbEILLbWUUJ/rCJvTuS7gKwD6n38uPLL9BPWZXPHqp0STsiSN - ygWBRuSXEPUUu1uIFDAV+HG9tv0UXaTVT1kkHORCbTAQuCeIFPVI9Yjf5tPP/67BrJIhFlLvp5fB - uTwI+HAtHzl76M8Qftx1h+uHu1WTe9Z4VG9lSjHZRgb1nZ0IzfnqEoFco375hmYIeTfTA2H1P7/y - +ZGpe+3JqBUGbrX83m/1T+lFx5rxPYT5C7BdkmLNbMWK1eoxhpvC1SlmFgEDby8Bsm/HA3VMWcul - tZ4NltY9ksl9bT1enXNd/UzHG/nKKM6n0/abwkMnqvTHH4Rklvy/9CgA/72j4P3QvjTaRmuXl9dJ - g9JZw/ioypaxCO05gDRV3UC6tvu81elBVoXmkBIgGYKxtB/tjuh9VqhX7ns23y2ygI9mhwRNScXE - tDAhlEDA463tvavFK3keHS99u7JA5wmLoIYwGKYCm/fC7ie5IxkEUD0GU09JRUIrPcGvfjjiSz9q - nqQdUAxL+Ttj3F2lfLqCYwFtTnqR5/Ei51OXKR3Uh/SMLzdF80RlE9bqsR0NrJ/klPF3PnBhXmc1 - NT/s0y/q3a7hc6PM2BaTE1jgQBsoZm5A+Nn9AHqLNQ6d9+cd9ffXFyCFGiboUpRPnEw1qKYp42I4 - T3c/mKzHpZdK1Zbhy6iW4CO+Ck88KNSFr3fXBtWrU6qBu1x52KumQzp83FaCPMgaVIM0xPkDez17 - vFoOaiYf09yTXwaj6myiw6nf0kPmb3JmUk8HJ9H7BgvLrsZUwu8LkWgfUx/dBUBeo5Op12oesVeY - 91w6B1WAlIsq0J35nPNlSN8u8o7Hd9DDYDDoUw45VJ/hl+5TkYG5kC0CR+jvcRR3YS4Fh9yFwL4f - ycZ8m9WSXvUWZvl2xHpSNnn3iHSC+q6Sgnz9PuJeaxHQO1OoldYOEKwxKOEIgz3d36II8O8lUpFr - fmrSRNcZMLOWB3jAW40+FPtSjeRCZJgNtoP1wsBMnBVVhDuDuwRCYGa9NL73JrzM8EWL990wpDyg - axfN+Yb3NHV7sR0PFvRuXEb9DnJs+iY9B+WyE3BQ40NFFKtskVV2KdVotYsE7tGaSPRzRLEk1GAm - cd2gq+aNGJddDea98JSRuz1lZHr5u7xzPJ+H3d0Z6PajYENcvHOANk9VwXs3LpiYdQMEZ/rsaBZe - emO6ctECzpxeYv/dDfnSPGkGKBe7+Lgkcj+7btCAJNlaNLb4T9Ttd1oArmFR0SIOOEaTFL3gOl7U - y5pXvuTQDuGmftyp7fgA9JJx9WF7qwQcHM+mITlhOsGQN1pqUXAylv6A73Cf6ddg3jxxPkbRokPY - EAm7Md974+54lWE1HwN8rApqzDp4iuilzgt97B9mJVpCMUFll4REZLQFkxYdQ/CS8SHQv33TEz6S - GzT2DqG5M/Jsth6Kqqx/Tw96vfFoHWx1+F7aDl93j/8j6cySVoWhILwgH2SSJI/MMgeZhDcmURRF - hgBZ/S3/u4Lk5Jzq/iqV6tgRHwznFX63fca4jLSRu7/WDNSZSojJF1G0vwJPE2utJNg4GZbDq+2o - AKExMA4ioKt89ox8+LiEKlFeBaFzCtwU8l8rXyANVYdZFE1CWCpq7CSTUnJ/+qUsUCQ5GmzA300C - YSp/7sTto2Qk4D34kDF8b+l/9e2jd25gMvktzm2TdrtY6gFiPxsibvl9AD7RMhHGsvolzmKxzvq3 - 3sspBmzjRR452w18qJh7QuKjfXEWOL17aB2bHDv73kTsxaxSeInjxj+tXxTtRxhCFLjbmxiBAqI9 - UXAAYtzeiKZczyVfakmPbm9R9tkBbeNkrTWENd9L/olpYofJD+UOJyou/twZx2inbtujcz5luK4Z - F2xsaxyg4NEc62FyKbkPUxvQ1l5PbOi52XGVVCu/FHSGnLdqB0t7zTI4JsW8CHCLujUYhwIhooRY - S8TCoURCO7jWkeHv+6sfv3CYQwhGhsMRvuXlktdZjH76TGKQFuVnoiOEJ4MtfFqvp267p6GJ3pIe - YJV1127K9ZGDysFNcMNbHKAxMjlYILsj5e+8mLsj9/CWMj3BkLzUDVqCBs1wz302PYjdf/9Z43Yl - xqkpR6pLrwINcjX5r6UvKKUPtUF77qX+LAMWbCfGlFB33FXifs9TObFqZcA9xyk5xw9JXbebK0EY - izw+b4E18g56L6jPjK/PIRJGU30bHqB1Whnn4Vd2Fp6KAqyjg4R9056jaRbBAa6GLBI97U2HQ7MI - oZ6f79gvsOasaXNpgH0Kjz+/dACv9O8v8O2hJ5IoLXQWkAEh7eIdu7GSdIxXchL8mw/pUPtgXq69 - KH7W8kLMI5TAMnq4guP1GBK51YWRNk0UogGqHJbYMHOYrgMCdNq7hP3bPe2YiesDNHDmGRfipwb7 - qEmPX1LIlThvpJTcO2srKI3SiZw7Jxy5E373kBm/HvFk4o58/vmI0AuMCuPw5QHi9L9UaH9MsYu+ - R5XmLNvCpLR0EqzDAtZ6ABxc5YODccv5Kn+4IFHMMjBhjauKssOXlkG88ME4u/RtR63rZ4VWyeVE - kj1dXRcrVv70+6+fgOafVgDGo8pw1vFSxzp+HMBv2Z+xbVzlbo+GWIDj1Z1JUuJLyV4FRUJK8Uzw - jfH4kb8R1ENbeBSk/tXD1Ld7i6oqt3xWKEqH1vVZRE99oOSSXrJyT05mAPujzPiMIwYjfy0kEXHz - 7uPzz993jswu5NN7TgolZFQq2UCCmbo/lmwovirtjE1BLPfwFxbePiP1XLxC9mL4WA8tU93xUXJR - qUQLdnnglrTdTIhQmYULDfJTt5ejECM1zgAxFUEq90Q5h0iNvuli+a9o5F+3MYWl3FnYs2u1ZJPp - GcOQswZS2IzcbSfeTpH5Sg8L0AIG7O37HUP+lDFEA+nvH2ajX5HRcNc/Pe6YY3PnRKtkcnz79Ys1 - hcmFERBu+E/vdn3DGlBPWood1R5LoqvvBbyqSsMZ92rBPmzFDp/Q8rE6NSe6tM+XArv5oBIlVPty - NkHfor/9+Udfi9bvxw+Bq7sl1n/9XQ91U8GG+j12bmUMmGMzMNDEIsLODY/dYruZD635k5MqMpOI - /HgLQH4NfUa8nwEfDHgVzeczJ9jir+ovlMeEYNQu2EnvDCURu3KoR+2XZJ7VdtNlPxWIVw8Mth/V - HfAL12XIr279Avb9EE3Pdnui/VgCYvSZSRdBrwN4B4xCsmfcgi0hywdWx+BJnLl1wCoXQwjvX0nB - Pz2lW5HYBqxDW8IOxzt0CzT/Cz6TMv5C7IxyFdrKANngPnAzJCnluys+wLgzW+LzVgq2KBIVKD6U - C3Z5ojo//jHARqWN1LHoUuZFLk/Y3mwZa90iOlROvBjaKnf0J7ltRjoYaALwBUOiccHFYdImryAr - SRQr+y9Ve/aqHp5L70H0dhcBNdMkE1+tpmOvP6rjntdBCj0ukpftwmQRM/PbFzXU7XEwGnpE1TeV - YPcDVTVaA3V/BboGqrl6kmvnhL8bZClEtay9Fv4dSM7G+NMHdk25E0Uw19+LklqA96Pl/vjtO+4D - mA34OQgG0brHCvbMX2P084//80fUN1DgrC8L9oKBc2jErsx/vvUOtgo27ll9YX+4YJyB8etsmi2L - qGple4F3uVP3ipopGFEZEge0ibo+ssYFQZt/sPx8CN0o2VQScXJ/E3/jtZJea8eHS9fYGAuh7sxO - 1vfABsYRy+/5Ral3oAa8nRYZn/OPFY2nymnBH3/4yT8AAAD//6R9Sc+DvNre/vyKT2cbHQUIYLs7 - phACxCZAJqmqgAwMIYTBBiz1v1fkeVupUnddR1GCsa/ptn2vhogjScdqLz0jsh9blE6nb2uAY3ds - GHnvKjCbBbOURR+wPcEQjLi0DXD9XAuyDyU3EuVZaNSAqneyPdfNP34gzLnFDt5j7dEFP+AhlzPm - JSuB99prkJczbGtmJd7IqWyEEP7+716pUMSgYhbAY3OCVeNxj4T4ji34W39BF+JyIxkHAXjPe0X0 - Ka9N/i6GBjSTJzHtKr6j0ei9Gsrug+DmcNqD+XHeaip2djqxR//rjXqSB0iDVszM4nAre1XyLYDm - 0x13+/PKm6/PXQB5eLzQwdbKaCxryQdrr4+YZhibcvCutIab0yywbMNKczqv7BaCkWTMHqu0ZD/8 - rdemQFmrL2cGQ1mFl7zrifXcBmC819ML1OlNIASC5RZyZxLQsYsapq9Ptjkdt4kGpznwf3xZThcj - S+Cib9hBFJ7l/NMzY30e8fya6nKgxzZQWdK7eN7nMphn7+7DYYwUcuiplI6/8b5+koJ4R9nrptvV - T+A0Zj6J7LeSzqvZaeAMYh+Pi35te5xI8KzeOios+mhe/BN8V57APCf3OrbwGyRv6U6VvGq7sSHp - CrKr7LLE+8YmN87XHvz8Ep7NoewjTG3Q2cqGOM85Mid0DSxkh16x+DM74pKxFdBN2w/snD8aPn4c - DwKKR0oCF27K8d3sK/ARDJtoHdjwpYWHAadAAexg+303ZmkqASE9XYhzOH3B1FwKDMv1aJLD5yJ5 - fPLTDFbbjuPVIVuDqdJqAe5s64pzs5S9mdG7Cx+er7F99OTldAtgAoLpbrLHurPS2TGhAOuvs2Vp - VXd88rzbChDteieL3uFD8swSZIwzYHt8ayP6XX0SKE3bGx1FG/Nx+j7O8JJ/e3K/eWW58YZgBvxY - zMTXDANI4dOt4Ky8fJKMSpdOBvhKSp4dJ/Iwm3U3yP1o/OOfoHkx+/RiO9DK6Jlsn8gxuSQo7h/e - 22/g8LkYjR5eYdkT//I+Rd1o6C4i4dogB2Uw+TS4SQB9tUnZo5wP6QyzawPn7fuMQWgJfFjwSfZ8 - JyGngzylszizFxwNOjLjFdsl7188Qe2LhczWYFO2d6ugsNwHDQVDXJTjhYEZZtPGxcoL5ClzH+OM - enII8RCrPZ9//DjNoc800cagdaMXhhYtzkwrrricJoHK4JIVOXFv2xzMX/C2UaqNFTkvfqDr75OF - VoocseAebiNR8IozcuKoJIQ+LLDo8QppUFIw3442H1zXrkF8jF9Y8OgnHb4PFYPVRZOJrb8nPjNx - cNTleTBE7dN8wgm8/vISeW9FJvUbS0BsTlWyhd2Rj+hsP0AsrF5siy2143ACBXSq+kbwbB66byXF - 2Y9vsIji1BznsT7DNKwq9pvP8/tgjOh4+TbMROouHV2ha+HzI+vklrVuOfk3ZQWqj2GxvfZuo8Xv - CvAOOkbhdjpFvC4nDckPizDcvKp0Ojpx/MNHZsc7B8y60Y8wfXt7urkrmsnfX7GBp5X2Isv76iTx - nchguFZ7FgJodzwUhgdcIXKnw1gst/LfFAg9X12TrZwAsw+RSVHXhA7RveZdjnZ0h5CWmYuV7JGn - s88yCcwn2SdH07560tnOIBRW/ozf55XazVa9dJWqbJVtq00T/fAY/fRIktio7C/99EA7W0LMe6YC - aJ7irVDytPTxCtRnPmWtuvRlD6zFr07m9DzZM3Te8YqFm43ifVHDNDhYcE+yFj74fNaCM2r38444 - w6b15mSCK/jV6I5celeLJC+RMSRv4Y+vvHEubz1Yvs/0m7IFm9sMfPV1iHWSqEGXTpumiFGDLJsF - xZWWVbPVGzTsdZlYgTDyhX996J7aNfHZQSznvJkpTNbMoZuA7qMJNR8DPqikMb1xa1P6raca9C2J - RH/seqc2GwhuwZEq246Xo2fbBtrCIcflCBQ+WjpowT6rVlg25Fe68ImMoptpYUUu9a4/zlPy0+cY - yl8z3QABSaD0Ngc8vxMGRj2fQrXsHZPsJkuLxsu8suE7e1iLX1ZN6i8V47/xulgw+uqbzP3xz5+/ - 4+b9tZyh97/sKr8g50J9dGDvyQ65znFrcpcRqvKiOrB9XrUl35g3DIddQSmsI7GbDrkO0YMKGsmY - 7qc//EM3wSrZ5aECwH3hNSJFnQO2e2y/0TiXRwoPYpFR8EFFJJoBfkFhN1yI+9l2Ji/34AyztKmY - 12yKjrn99gqt3IZsZ6t7b+zYIwAfErd4Hb4HPlvPakZlajW0t2DBp998KUKtYbjph5Lf2ChDNfCP - LCmPcdrJ7tcHEjteyY+/RjuIWoTH7Yf88p+pkuIH+s3HM3vM3izkW/+Hr8yXRb8cSXs24NaikDZn - sgdiOeMZauc2ZuYWjt4AE6LBRW+S4015/8PXiM0bon30C2fJRpOR//1oGDgnYzkj7q1+nzON6FtP - UKhgwfQNKNH9YgummmABnpz6RdVCiKPF34ywu6CQHaszS2sxdWo45Xhi5soOyl7PlVBd/Bo5VNq6 - q+hab9H30x1p+dkpJQef3P/pgz9+nS6yq8H7e9aI72Ix5a2j21D5UpEqox6b0uwkMkij0x7P8nRL - R7ret/B8njp8XfyggCLfBcvzE3elxOUvb4TvJ8dUPsvc/OEXWNYby6L4UYr2ckIocInBdCm1vXHr - HyAQpRf+0+Ozec1G+I3Cnlnrz9Ebb6t0VB9TUhPPWBVmEy9d7qS9zbGy6Mm+lvIKPki0HOn/cG8s - nrUPSIgM4gEkmmM/flv407u7YKt533bMXmCduzFxwpR7NAjaGpqKBHC5YgrvH5MnKL/8gFyeB87L - oxmD7faZU3nx+5M+6AIcv51KN00xAz40ZgK/TSphKW9njzKZZbAPQkpw87Iifq0iH86f3v/5MTDv - 1syGQ9bJlLdsZ44sqDQkuEfIvJ8+dbOPC5WLLBLrc/6Uf/WBh4c1LPTDBwihFc1QP75OzGahwzck - Ll2orVqFuapVeH/v9ze+27Sf0sVvCnDx/3/1BWm9/7Tglyesl7xbOpvhHz4xEt6qcjy/8wzuuuKK - UytbefXip8GiX9g+h8ic8fjCUMoox9cwjbxpN8U+4ML1wOLpswNTO8YveFTmB9tm+6kcj5ptQ/H9 - 4Euf7WM0vhu9hhJ9cWJ3euj9+BWM5vOEh4/n8zlTrzGsT8V28bca4Pv7O/nze940FR1T67sApHWF - SSZ/zWiyDr4PC/IwqEKBW87BPqlht9pvmLsaurTXzmkNStc8MD+Dp270bKzB85l3eFrWT/+91SFY - 8l66vhgH3tZpO8NmEDp2t/2+5OYHaHDJf5nraS3vn3IkAGGcrnhztFPOcXDP4GwJE22NMPZGLkDj - L4+zPnxXjiyiMeSiO2OFGLyczK2AoR6GKh2DHe5GdyVC0MWNQ5LunHN6+F59NH6/KtPXg9wN6BrY - 8K7bbwrfrR/NOnxZEHSSRKU3aMDkWnsI189zhNcsukQUIygBwYYzXS/vs9+c9zNcbdd3tvj96OcX - gcqlI+3DqOBLr5X+Vx9jhiWzbv6MJIEu2FQ//Cj56yuvYL3TCdvX0t2kEf/4ao4agzz33QHMYeaE - UGvDmFnT58P5dWIC0LWHSkz2/fzyM+33vllQ+O+0i937C6z9TCbRki/x/X1IwC8v2Wu33ByzNBLQ - 0CkR0+6AlvSuvxKUX32DPYZ6KGd7DjP1VoQlhbLYl2zlvmXY7scd0Tr7HfFyz2OYKiXFkmLvzZn7 - Tf3DS2Jww+42kjA5YIb2xKxemcxJ38QufCI+4u5IioiXl90K/PSwSaLRnDWQUfDsiPeXB9A9CQuI - h1Kn6m2r800Wogos9QSm6RKOpObjPGDrFj1zRbcwZ8V/B6BWV1ti2s4tbfU9HEElCxbOF//HI5tI - sBzgcsfP0SylclVQWIRGQ3YFf5nTt7clZIzGmZAIX8yhEwqMRttUiSleFTCp3kf7wy8kCutyPIqc - qheX5r/8tByr11TDtf+QiWuXLaeX3StGh+qpM6+qO7DkfSu45On08mnmX14Xw/PZtNmhXpcdHw3d - QZVSf8nhoZ1NoefdCia0G4h10BKPk7h0VGGFZ+b4vVZKxM9UeIV5T3/5+HgDtwcUrtqBGO1LMefh - VAjgUSgjbhzyKnm4+YxwqUexh8Olbtyc9RElF1oSd+cL3qALRwHdT08Xy+kp8v78F5FvBka+f/b6 - 72sVgKV+SE6prJYzGsUALvzA9GlgETsdhxdc9DjxJGx0i3/8h//wbjiYc94hA8bvYUuuz+TVUU0O - r2jBBzqKTPZGaZeP6JSClmn2bmvyXUgoPH9Em7lPDZXj6bLPoDKYa+aAzjWlIXwVv3ow8arVI5rM - wH5BfpAnhs3m2Y2Nf2zQ4seIdg+3qWjFGQZ+cnqzg9hI0TztSPGr3zLDnG2P7tbMgqv4Ef34wpus - OPOhGuAj0Zd8+Pt7/64SrHGVCIk5KhRa8kE66uxXXxs9Tb/+6gV4veRHwk9vLnkSretrw//yTqqf - YubzVuz4LtxR6E11yoznxo468Pn6YPHH5PSqp3Lct8uOgsBa6qvLVtTtgxbw52esEw2AdBtYAD7G - VyTbVFa7IUpdA4ZiamFwlLtyqQdBuEPxlQWjePWGpd75/3NHAfx/7yjYGQAzF339ctQZDtTXJCnE - WcOmm+io20Cp3oB4g7MzxwHfR6iCeGSHZK9zyZPkBzzo0ovhXS954xoGBrqs54jtjjfMhWOx6uEk - FhbR1XpjzmXRv1Bwrgj9Xo4xEEbwtGC2bzKib/w47c1ATtBDbS2m++aq67qYB1BcX30SlhlNp71t - 9aif8yNz9ushbVldaqp00C7sQUXfk07C3QebUFKpnO8vKctGM4DFONosaCVaDpPYYrC605by0+0L - xvXoOOiYXSdKjyzz5qjoMYzu9y/Z09YsZ+G2fSn6UTiy04qU5XiooQov1NsQY/vNoykzbAGZT9Fl - 2nc/pOOJTjVMZ+lCcJzlQHgXrQGfs5QxXxdGj8UNrkA2SQLR3qrPJewlNlglTU8OyneMvs9LrqF9 - HSy3XoQ7b5Iav4d0yh7s8JRMb/4K2kvtL8gitigi0MdRHiKlqX26mgkFcyC2EtwWryfJyOYEpF75 - tnBVFSPZpkcCuLGLJWSLR5NZceOVY461BwKhtGa7z6x4g8Z2yy21Q0LucHeMpvv3Y8BOaUZ2st82 - 5xlMnaWvUUN2ySsEU7XcctikGmJWCFxzip7zFdbbY8dsoFmp1EqKBcU8l9leHvbdHBUVRklBHlTp - /Ec66frk/32eDM6Wi+UugfDmNALzWP3h8zI+sC4UiWxD1wMb9s0E+EzjAwszw+PiLCcz7B9bjSxb - 0zwp4bsatsZhZkZqf/n42tEeVvLXJMfUbsppEgsfWYLvsZMJzVLKhmOG3pvvC3fA88FGu1sGOj4f - jK7Pqsn5uQgydLG7Lzl0qWNSJ7xLoFZbzLzwdCrHQUQreMJjRy5A25Ybt80lOIB0YruYTyUd7x6F - h1P1IWQmo7c8jwUZfynkdMCtN7lwiAHsbw27W3ehG9s8vsLYrmMW3xnreCKpGngguii21uw2xfmk - wmX9sCtN7+X8uptXqN77N8n6S8PFMFAgIEFZ4ra/2BE/qzKGRa0aZL/LDx1PQjdRhdT0ifN9As5v - 5h5CAg2D+MCjJgWhPqNTlkjsigExN348NzBZiw3bSWIZsav/DeAWWR1d4dMt7RszLSCUjj0z3bcV - zWHUjVDpGpeiQ3XtuKqUPiQytMg5W83ljC9fCIn9fRJ/fQ7SzRS9GxgLaoOnzGg7pllVAJPd90jS - m5p0Yhm1y3jeJpJtNI9LDy9M4HRe15TermbJdmctRP5mlTHb9LapcNuYAYr9PiDPuvU8IZ9HC3ah - W9C109xSqUKCAKd1siMmKEA05wEvoOjqX2JV1xCMjpBcYVEKFom+cRR9l8YPiI93jR3sMfLE0zsq - YGZXD3a9sbjjzlvr0bC/nDHI43s6drCngKxtjcVx03XDatQTlNfrEMsxD7nYixsMdQ2siLXWdD6e - 32EA+8vaomskiOboDl2jXjvji6W9+fWms1aq6PL2d8w6rJ2UuhcJomvu1GQnFBWY3Yu0+uEDRtTq - opkkSQHIBESqXI4C5++d84DbPimI9XgRTxK19wx5ZjCyWyVOJDqtcEYTPO9peAa7cgLJx0eBTz12 - uF5X5lQEaojECAwYcW0wp279KMBDbSyyz2OVD8MQFMhI6ES0g6l30gBDH5lg+2Kn0UmjMU+CMxTm - MqXrVNqWovjOHEhvOsRzzgZz+pC+hXKsnEjEdKHjhxLY4Pf712W8Nr/5+cPDxApxKfBeb+B7f/LJ - 1vDWfD4tdxJ8ntZjOVNfg1mYhhAWarPH611/iZb5EKIsxl+2S49hOk+970BPQJTtPzPopvfU9bAM - 3Suz4GBGvA1dCpzotSPPvsBAGgMuoRzxC8Fco9EskhqjVfUaySNkWiTstDaGJ65s2RkOW8Ck897f - dFf4IYftmpdN58AGkiAvl/nb8X7SZRXOn+rJrrQ1UmmDXg6kpRDh+bU6dPzz1mNYreY1hqW+5uN6 - 1BwEDFein4YCb3mfGbyVRGe7rdmmzUquBQT1fcQuXu55AqgOIVA/1Zc8iueu2+CZv+BTyzfMP7Fb - SWW/adAPD5PI5oCq5v78m+/MvRzPnnS6WTLMVXaj62RvpRtZcC2ITRmwH9+MAXg84M0oZGKQfLkz - QGlsaLbphXhwh/lMZjWAwQbm7Hk6GJ2wf3QYoLQlFIoPn0syZC/w4S9G0sONgwnlGoSA2BviFvEr - mr7D+wWrTpPZXU0I/37ze6Mal5YT+7BuAb99lAr4t8Yhj5du87l6n2zkZL1OQik9m3O2VQI4Lcy/ - 7em1E8moZaiKMSc26XtPPOJ9Ah1d6di506/pRk6LGD2ZHTN8OuhRczUjjFxNkNm9v7gelz+qD8+K - /qCb3dkBwvOFHz89w7wGWXyS7UMD1WCsWHzZeFzsamlEXk4+/8wXsjZcNEw5pWHdUd7363v4Gy8M - 3rRKO+t96tEbTRmVakkyZ5PyEbVTdmMLf3Opa3Mffet1hj/arulGU/gkqOy1FIMi1jspooKNov1Q - YLlBuTcW+lghdpIpuwpF7c2D07ooWm0AnZf3Jfah46t7PGc48frJnDWxfUE47kdCvo5vTjB0CiB2 - 7ZYQGJ7Nzdhvryg1PyPZHm9+OoPHrVUn8/BkbtLJgOdPhOHZZS45LHg/5ueP8+M3pp9dG0xpdVBB - qJg3CuAOA050qMF+/QypcLH9tHiUqgudc7Ulhp1oQFrwDbantYJFoJne5v1sfMi2/ptlX2XHh5Pr - +eo9SzVmPl4HPkFj9wDpLFwot+TSZLLg2jCvUciuXt5H/a0FV4BMsKFdgyrAV9iS1E/o3NmTT8wc - U1glSIwFDZd7851umo8iwzB+71l4XG9Bd0tzCpUiC/BmaDP+bS+rGNS1vGNJXmpgTGxzBOc0jfA6 - mz/lkBm2BO+t7jF3ihxT8kNHgjWiPi6O6zfnbVtkqm2qEwZWSDj3nXaGRndVSQQf90iEBnnAVTB+ - 8W3tuemEhqsE95Z6w1PGXW/cvW8QHr8xZxf7vY+kH/7vbkOE1fppdBPunwGMo9eZTgezAAN2mgDq - o/umcu/nnXDg/gy/q2dL21XSm72NFKqKtDfJCY9fjwdWVYFCYRIzU9vzGKr7Gtr3UWDuRttH0hp7 - NXiktzvZD+06mvrH1YaWgD12yMuqnGar6OEwZzq7LPpn/Nq+DYvw0DB9/753E3xcr3BI9CfZkf6Q - bkb5HsA4yK50aR1oLvh3Rnb+gsT5Uq0TZzkckW1DhjfjxejEIpgDlRnOh1hGUqZjdhFGZBzsmhjO - rkh5rNVXVbJxiNOFj7vmM6kobE4BO2yx4PVW1MAfH7Ht1CYR3ynrEKxfd0LfUe9y/kG5DYnPUzpm - xj4a948SK6ynnJDbtQITW+sv6HicEi3jDp9Sjs/gt57Wp0PR8SpdG9CN6zWWOwRL/pmOPkre6E3I - oQLeXLr+DC+95jP3o6N0tOSghgvf0/x5bMEIxrsD6PR4YOA3ccQ312lECowjepflgM+HWLlC4f25 - MUu1NJPCIbThc1fbeG27ctSOYplB5oQllpfx41M0NGCZP8Q44jrq417X0MJvVAmaIxdMgV3hfdwc - yT7ahKYA4Nv4449rVSZe99N78Se4s6s8OJF4ffQz/LzlCsuRPS1nmicNmnt9S378O2brGCKxa7a0 - HhXoMdfUW3jb1hIjUhKmPenPGtCv7ZnC+zU35yR9ZZCe1SOzbtWQspCiGITvY8jSPEZpU4H9DL4t - KxlO1KKcT75ooECQS2Kih14ufBfAZT2SeNFHEynlWo7e2CDGSAXw4xP0ktsVs2op6gY0XAUoXNaY - AuONUy5G1xYaB6sm7vu5ikboDS+I2PVMTumRcP6Zbj7kxB6Y3pfHVBzyZYdA29zJaUXMThysTQ+V - pvLJY4gO5uZRlyukPHqHpI2/TfsmBQE8Om2L58WvzH6oSbAOLIM5m8le8GZSkRJsBLY1veqnwGeY - 9NucXZLXDMZDMhhQfBd75q4S35yr29aH73p9ofm12XHJ/cgGfGmPD0W6OfHZ/ciaeobTmyz8V/ZL - W1bkkU3M9g29d2zsDwlw30eHhGe3MeeboGHYPGxAsJpM5mQESgumpjJYcqxiPnDKH3D1qSkxuX8G - PFXaCho2bIn+yQaPS+h7hWl6dsjhEVueoJ58AaLj6BN7b+496TtezuBNng5e6SY3x3xFzvCoapA9 - kXkqJ/nxvsLsEnxY1G5DMB/grYZ9N3iUtu132ZF1eKDC+tyZ/yqf5hRaxXLnSbJl+yU/mAC3V0Cx - 0UiOjvwqhXGthcrCf39+YZRJ2aP7sN2zvVB80lGgqg/Tx9UmP//XotFo4TIseFX5e2/+6R8t5RWx - EHksfKgVKIheR3a9rk4eo3UfAj/8KATrywnzLZpXqKd8v/Bx3nHBCny4LssDs2ggRVM7ni2wvC+y - +N9y0ztlos5TVhJzRXJzMAKlgbhL98y7qWr501MIGI5EHOXxARzNQqbm0cEj2xUxy0koyhq5XWpR - NbIj8LKnrwSPYUFpz6cu4s74XC1nnjiz663UDYs/B5c0NomO83U0oXiyIJJwtlQU5JT//K4gHQvy - BOfYG61e61EpVQ07pUcGBlvralgXYIkqP3onehexR80Hrpk/OE3aqw5cQeWw7X/+yOSiKkvw+VYT - 4mDAzHnRlype7wOcbddhuoznC2yeM2DL+uSUI9H58zOHRW+xKt0YEKUNYRHT446H4IDhLpuOJIHy - YA7oSGQ470OPhu7bNMXPxpLUYN8c2CPOcj7S+9mH1077kr1Q7KIpv7suXPQCsZzwY/7Nr58/cwHs - zH6jleFy52Tw4ztzlN6hjJ7r+c6M59H3JCe8C3DJB9gOg4G/2SWPUQlpRfbabtmR77oqfMbHFTMy - 7qXiq86zvzzk8dxWi/7ZqjDerGqmrzSFj+9z4qizQSDBz63FJ/Op2rA9IYXgZG+arHw/Yogfzx2F - SZdGY3qyMGoqoBN8/VSeGO3CBv3w+ZSrkScmcGihejAyQu7lzpSat2OhtUDOzJkcM/3j30VfMEeT - vyl1lJXx0y/ktpl2HU0+cgyRC212p5aX8uc3LqDb3SwWL354hrnh/83Hk7WzvXcYKCs17XarRd/i - dIyapwT1nakSvL9dSqEC+gxPqq5TGB2vHb8K+wd0hqvLtFbCS5/2Qw1ih62Jv13Pf/MBoZ3q4G92 - Vbw50OcR5epwY9u46fmST2XI9Lfm33jMyUc+Qx9vJqraI4nG4r7s8I2kO4VPKeebOrz0qIR9RVIX - hHyuU3H86VtmBvfGnK4bq0B+Gx+XfKqKpkWvwwV/iWuFuGOjfArQfRSPZDcpedfK17FAP71i7AWz - 27Rt+4BS2xZUOp9IKkkP2sKpek8/PI+GPtQwym+nHdmJ1vMfvvGHhBDzG/N0Cq3WhlUurQiRdxGf - jO8xgd/X5pefnk1eh8/+l1eSm3n2omnuXVX1wK7FgpGU0XTfjgbarfYlrUN38KaLmWWQVFFLP0se - J4Dx7sLq6IrMq3TTE+p3WKE1sSDDl87w5iY/qzDURIPor4PPN452s6B3fz6JdcAN56sk7uHiv4nD - a8tb8jDt53+Iw9IczB6+tCpWzJjWQrTiY35mLuxz0cHz5rKckJO1ALRMhMTYea90KJ9qD5G7shnh - Guu+ufhVQdSEGrGu/MCZuNNmhA/jijj+HUd9jeYz+mTGi+k584C4rLdfXoIn6wv40OOnBha8IU7y - ir3NLr8U6pI/MOfRSVGfjtkL/vDZ8MOqm17JrQLX3K0xT05dOZXA1BCKJZcdbioyJwIuFB5UY8O2 - kpWn/KrdrnD+1E86cn3k9OElV6infcycj57w8b105VlMHAuXfojzyUcG7AqpY06+v0SN9tpKkJK9 - TGcpPXvc8tEKipEysKWNUjnR/iLD3/N5U/RNO2e8rODG1wT8metVN92Hq69yNkt487qeS2HTeBl8 - WoWzdNlVUu4+jisYfMLrD1+94TVP57/566y9vTn59cpVBz+NGd7fNuXAd1EI/SDDzH0/H9GmoKgB - 4YGbVLOh4jGhk6/wWvkVM1spjIZ8OUE1RKBkmuyI3kDHvQVP22rC4t788GHItfaXtxEn5jNnXxK6 - aLM5upTj0e4G4RxRMA99Stz7/san4f7MwOPpVVSkwaXjw9rSQC0U4cKH365d8AiinewQb4+nck44 - qeB5VgpGgtWdT81jsICoeTXuX7oVbW7kKPz8DDn4zRmM2NgGv7yC7GerTrnQ7P74l7gdepr0eDct - uDHciZAhoul8cPICUuLJbAsfT875DY/w579XC15zMQpauF263jr8so7mr+AU4LoaR7Kjre4J4bRx - 1Be1JzwLysWc3gjUYEnWKXSB4PHLdJXh/BoTph/XVjcpwyCAJY8g2zW0gHQm7xrO4Fwzv9Sf/Ff/ - UK023v3VU8S5d2U4BF5GDvEqTad9mFXw2y5dkBZ9Mcb9XgO/9bDR5MabVuVUwEVfse35hM15VXcx - GGF6ozfj3XGato0Kf/WLbdS3nN5H14BUeGXkvOQ5A/GuAtxsfbbkfQ+TqpJyBalVmOwEpqKbBxhi - xK6rniy/X3aHElgqvH5z8ljwf3juGIT+rXXYNle5yS9ToKrZB47MjQ0RsN0UVAAV2YlFTRnx8fnC - GZQ0+0y8PL5H/ZLnIb14JcwqumPEV0lG4XTYFYwo35kv9YUAROa7JiRkesmFbkzg+3zEZM/rbbSJ - 3G0Pvy/xRde3qwX4Ny0MpN7pm+xCF0fzGpsVItWxxeKtouWsVGQGv/VoldkI2IIfUGdZSpLN1ICF - HzVo0wfCCvp20VRR7iLkmpwsfJGOVZRXf/72GazugK8DdIUPPU+Iht+vbsxn2VYWPczsKx+jfr8x - Vj99yfz6uTV5rWxUyLb4jfWRRuX8TEIBwd36unTNtjqhHfoQNpRbxKg7zIXw9kxgM+c+fVUHr+PD - pZQgvWsHYt/Z7I2eIxjAqk8SXfAhavLVLlaX+UA8KOOIh+dTBV+rWWUafmslz3enDKaifCJR3LiR - WMjhAy55BNHnYC75wp+wVhvMTk7opOxtPkN4clpGxWw1RXNJYfKHX1trV3t1r+Ttb34RU7GScmCX - bwyjPSvwWxJfvHskmQ1/+dw9fCUpv6Xf/q9+kd2vuinhUpHgp5BqsmN+yudWmuwfnrGFPyLmha4F - Q3fY4kknitenjnBG1nXosOKO1KThEzRg4T9meGejFNdLl8Fnej5QTnLbHG+P8gWvZpGSwyXLvB9f - gdJ6HchtvLxNagKz/dULcLF8f8aj3sMjnmsKs09ezoNTuChoz2ey/eUDPYweyKovEjnE2avjt89U - ofMMCsrP7jdidy9qf3zMdmJkRX2hjzWa71rItpfNwZy/WpAhvz0fmbnWtE7a3vwaolhwySFR82je - O9ILLfkmuy71V/7Tg7/5KAqXlynuw7iGjM0ThvVz600KrANo0wzhUa1FPvXb8Yys4bp0sQz3ZdsP - t1lttGxi+E0db1IrXAAimxMG5bwy52471apwFEN26qnc/fwYcu2VgGWg2WAQp0pDtilPdKt/t+b0 - qr8POOp7i/gVss1hycNg4PceeS7rtbPDpwW3Kb1h+caEjp02RIPvVj8Ra2/WZWN+Axc2h2SmvTAM - 0dxtlRqqBy1bNlS5gDf+t4IXut9QERT3pb6xgjAW5IYdc+bx6f1OJViZr/ZXTzPn3/htdzBkkenp - 5Sa4W2ew6CssVNeQj1uqhOoyXswZlAJMFp1fSKk+4Fc/AL/8Bq7VS0jnqN+ZIhmdB0BKmCzzqSiZ - Mz4hnOSTvNR79aiXSddDa3IOJN5c1IjWKRrBA/XGn9/pWSTJYKl/EW+pX/FOKWQINkeCxyEmnSgJ - cQXuaAqxNF70aHO/lK/fesLK4s/HM5oNKF/WEV6dO8R7MxgT6JqqhLl3Lvmon+sX7JR2ZHbyOpbj - +tIkQNsEFvHGYv/LpzP1XJAbVe97BUzL86DtY7wwEhw1LhR3TUZAIEdGqHiIsjbPrv97R8G//uu/ - /vuyQeDfdXN/vJeNAcNjGv7zf7YK/Gfzn75O3u/fvoJ/0z55Pf793/7ZgfDvb9fU3+F/DE31+PTL - 5QVo88/9Bf8emiF5/9+f/Gv5uf/5r/8FAAD//wMA/S8EZHQjAQA= - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99afd945fa8a-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:52 GMT - Server: - - cloudflare - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-allow-origin: - - '*' - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-model: - - text-embedding-3-small - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '145' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '10000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '9993623' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 38ms - x-request-id: - - req_4e89305535ce42cb47774650cea16db5 - status: - code: 200 - message: OK -- request: - body: '{"input": ["Are counterfactuals actionable? [yes/no]"], "model": "text-embedding-3-small", - "encoding_format": "base64"}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '119' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/embeddings - response: - body: - string: !!binary | - H4sIAAAAAAAAA1RZyRKqPJve91Wc+rZ0lQxKwr9jEpmDoIi9YlAURWRIAvlvvgtPV3f1xipMFELe - PO8z/Ps//vz5pyuaWzn9868//7yf4/TPf67fVfmU//OvP//1H3/+/Pnz79/n/5t5a4tbVT0/9W/6 - b/D5qW7zP//6w//vN/836V9//iFSxZOT1RYJVfbWDfZHM0QXaTuBBXdzrBT97CO7971m8e0ugyIc - I6Kp3NOYeVdY4HjgW2JZ9wcYO94R5XuV5aQ4t1Iz5sksKx6Xmvj9Sq8NG0Zcwh03+cg/23ZBTKeO - lbd0TzF71BETloOXA6ySMpytl8TGnStDSOPNFwUULWBx86+l9PClI9t1ybAs2w0PG66tkfPQ6YB3 - cW0rprG1ibtvE0aVdn4qSn6okYFC5GGy42XFDdQKmQ9da8TJEzqYtk+D+O322TB3Y8ZKkOUzBuf2 - 0kzbW1wrRI9PGNy/ZBjX96O8quxLnMP22MzwcRAhaykmx004FJRc3j4wM0dHkQgcIB1UJ4Op/q6I - rQi7ZlazJoToOmYoJ9y7wZzPOiiw1kV3sTkZ4u6+y6F/j3ri0OHWSJd0KRVTmE8otKU+wc5HTmHd - dRzal1Dwvs8tNAHGlwn39UsrlsN8DZX7VkbkwNHBk7yZ6UqTWCMyOvmcSHuacsBdEIefXLJjy6d5 - isAusBhuSt0xpEZ9pHDf1QY5DjbwllGeOaV7BxmKDsvHG9On6iuo4AWUCK3mLW2hqUpGz3dkyD0a - +KaUMXSbq0Aivb4D0TpVrTQ/woBo/t4GQnTLeqXkbg+CWtkyaGBOLXSRKyNXM/2BpYlqKmdTf+ON - d0gatnwTE/72O5ItEwj7T/aEwdusSLx5a4Zw47oXRFecEXuXxImUS/VNSfVPhVD/6Yplb1wh/LRd - TY7yRmO0l2AI5363JXpb6Ak/iZIt01ZG5AxkPRH0tWZ2IazQpbrojNpubELLe1vIdQQleRPWvuD1 - HQ4YksNojF2cLlCnIUNulpeMpcVDhmnB8WQ/nRtvfiRpBw/ClxGXcO9hqZQ2BWJz2oWSDmAzf97c - CdaB55CQmloh7roihPfetNBFv54LkTufS3hsHjw5gWYcluAZiPAS8RqpuNhlFPLuAsUgEcLdJqqY - 8L4+bXiTLhc8V4LhSZLaxzAbzRd+WlbXMFMIc/hqqUbSZNkB+tESW9G8l4TKQzaxiR9rUfm2NCK5 - B51GyNwzr5CjVSF9Vh+AJtvtCV5m+iFJOB4YURINKi/LI8j4EjlhelFDqAvDCzmSeC3EzXK0YK/H - e4S8QzJIxVnIYSLLN+Sjj1Swd24t0DPVA4nDc2wI3vuWw6qjd3SOdyXoZ5qqyhHuPJRtvrYnaXWW - w/X9ocR11YbuszpX9L0sEI0bTW8u2LuHdaHEobyIOqPfffAE0DziUDGIMfBh7+lwXNwXsb/9K5nJ - +YihNj4qLPu32Zu7bDLhrdpsiKrXG8Ac//xS2Kk/ksNFnAHF5amDlnP7YoWihZFimXX44G4Us3eQ - eMv1dHvCuvEDkp83GhNzrsmV0PZqZD8U3VtAsB2V11a+EfN5DcD8ljYqhIQ9MSfYiLFvHUdKrX8A - Vq740Lz3+/dN2eb7I0HGhAyxtT6+cnuHMgluS+oJlZFZcD1fKHY7BKiVbDkoheIUbl+Z4jHe1Wz4 - 7BYcxsqNB3jXFT60digj6vZxB0yrs0zh+9IgmnEtwR2UHa+kXVeTIn9rhrQktQ6U4Mojtf52CS3d - RFWkqhNJNe0gw27+MBX6YAtyH2Bq2JRPCzx23REZeZc0IuQ/ORjTjqAzLB0mDktkw27qHJSqju+J - NchU6KvlE9lVfQeM9JkLxYNkkrtl7A1BL8+tEqdxg/wBzgZ747pVlJb5xL7vX2Ay3dMWpsLJJ7HY - SsOUcs5L6fVoT+wrqwY8iXoNx+vGIPZRH9m8aWqoBAXZI+Opb73Fb4UYbuOnQbxRejLW7yQM89Mm - J2qXugPfFkoH7Sx7EnXFP+pDA8MVP0NxBs9ksQ5+Civh5iHN0mnDHnGv/84z/oy37zCq3YGD+ecg - Ee9o0gLPseYqfeB8SIzci0fnWKEycDaUuMVbahaa0pdic2yHXGNSjWnKHRueAh2QwCv3QDJ2Aq9s - DxyPzNLYem0p+M9djQ+E6BYIhnmzO+kw2VKNaJXQeEt23UM4eraMTu9XBxYtd83feUMrfgAeDDWn - JDrmwt0e3QymyosMSyvRiB1fr57YHF6x8njvr+hqPfYJrj4sVmY9M9f1RoPw4s0ekqNZoWJ3PjP+ - RmgJWQc8YsNbUPBG/zWhj01GzvVXKoi9bE1FvbtlyFvmyRgD8/2C792b4R7Ix0RCapdBulNH4meJ - kUjUUDrY7vQQ3TWzGCjVFlc5qLcEc9JsJF876nIopt8A2cHZbUh32ZYK3Lky2X+7ms1WeIgBCRyR - aO5R9qZTpJ8UFlxTUrQPHyzVh0XK/u4Of/GbSgf1pQwhviNLdXxD7D5nH5qW+kbXkGIwRe1SKkVa - x0jl47SQ8sbqYVFsDHI2IgwmPQErHxJbgmjAFV0x3VN4ZvOJ3AN6KIRmeIow6JuQqLZGwTJsvhl8 - xBCiQNxayRL4MVVQmpnovDuGbJljzd6O+9HBW2vDD/NE+BDalZQgI4qV//m9sudb4pro0IznXOrB - VnhbJD3urGR+hp/wt59YvmyFZAw3nAnHjvlEfdymBAcG5ZREfT1QUjvmIGbRaMPjVbTxcg5IQj9D - rSrvfu7X+UHClk1ow8ipDn/7J9NE7wbjD3JQuPE+gHJP3YTaLnCI8Q6YwdTp0UErtc/4ewRfNmXx - pgeG5Jd4qcoM0JPDq4pXkQWZ22xm9K10I7xzFxWZ6eabzNKZjr/+FMKofRjsNjNOsRj9kBTWDzAH - yrMFWUcjchP6iyF8vUKUMxxoKIgCETxxt4ug/FoiYotKmTBeIBGEfTqTw+e+YfjOoxu8eKQhh0ml - YBG/WQS4RaPE5Yk4jOod25BftJi4epExNsUehKHaxsibwbFg149owgwjDTk4MhpJ7xYfPq/ce2VB - HaCBdIrhQRhYKFbC4uHWIqGM6gMlqN8/DGq+515pzLghXpPObHlN2qhYWbwgxxnrZB52iaWs6wnx - 1zeN5cj5GewXnSeepjjNwlS7BBGXJsS5cw82qzq14fuAP6G0ro++Mi8CJTyZJDdnzD7lMYDQcF4u - yVd84bH1eP3Wh6rez4flWd1TmFe5F3azaQMqh1P64z8hd3c8QE4yECHo2A6DSi0NwXp+LWj1V0S0 - 74zZbPJfCp9tdyRW7XXe/FCGGsJQhMTDNE2I4nY2fHmxjw5mYBoLFfblzkaai+741A9M6G4qdOSt - StLxwBi7yxWFMmuv5K4Ux6Sd8x7CgX5GFFSFAybnMapg7T8YLnxY0A34PKELZ584/qEoZrff83CU - wTfcHKvvsP7fAouuI8h6qIFHto5vwi+9Q6KPzWYYNRLVMBJ2T6Q9d8eG7jcXCx4L7k0OXNgwliyx - CgFVeXRf+Qxd8U7RPuZIvBPyEp4NGgTCTgfE4E87xlrJzSCA+I72sPyC0d0eU2XDXh1+p+YF0PX9 - K2pTXAh6WJeh7xw7hfN1ExP7RJCxmG0aw44sbbi7cxoQBGW8QZj5J+J5zaZgQpfqsJEuLbIOIvZG - d3tNZaXbfZH77hD4rP0crPWIvLiADculroTW+BgxVT9jIW69SwviSpiJW12tRJTkYgvUlAwYBBds - zHZU54pjO0dk762HN18/x1IRVKEIuWEvDd3JgSrE7OaFyzZvE5YJKAXe2ADk3b9ooA16lJCT7IAc - BS9JiGVSEcrCKCJnVt7g9/xgD6mCVn7CiKHfMVzvh0dyGL0lfY68LKkbDz98OWBUeAwmTAvIk2rz - lQw23isfpvfERtbjeUpwZuQy+CiGFPKfqizGgfkLjHfnllw278igGxZgeDWjPbrcMPKwfY+ecCJU - IFqqvBmp7yxWbo1ZEXflb0tLeVNZ9Roplmrb0C/JM7BLm5EY6heB+ac/VrwNRdUZjfEWBhkcMVJC - pgj9MHsHrVYkSQ+R30yH5sfvoOxlW3I4B7fkBYYO7n54H+FRTf7yt/V5kYXpp5iN3TEEB3mJkDGV - ZUMj1NXQD+yAnNvqDWilaCPUbS0OxcsdenP+fKrwVikbZJ69nOEsGl3gSqcCaXV085ah4k5gXQ9W - kNN5H4FuRKBV5BTK6nOfSFsWcZDTjxqyAdwDsd1VMkyP1xaVn/zL5vG2fYGHdYyw4N+eHj4TzgKo - 7zhkwv3JIH3i6LJzEhUSFlwEhNspGcF6PpBJPiqbR/e2wP0xcfCgFc9ifny2pfzyIp/E6asYGAu3 - LzC90QfPfZsA5oj9TbmHryO65qPcsNm/3ICi1zEJ8XJloqqEGRSp/kYqo5jNUbvc4PQACXLRd2D0 - +vBcOJcHQkLZ55Kxl3gfandpJvsdS71lK7xKZejDE1E3Yz4w+9qKMA9bnVRRkDJeKkEP8+1WRcU1 - uQL80SSqyLgCyAt2QkIUMYygSdOGmNKpMWjYeyr0aXBEyAlVj99vLqbc0ndFtDriPNbcRB9SVfBR - mMp6MaJyR6Fb8C3K+/GWrHhyU/LH11rrcQ+k5iaGv/NNDgN8evPn8CiVQ+5x4WLQeJjjp6eDpahd - YrYW9RbzGqjAe/RXFPr6vhDqLLR2lecGxHpNKBG/ZKPDKDtGRJ3ynTc1Qy/CVU8gV8pcg+6zLgPW - o3dD6SLO7HvKXjYoSZch/5AFbOF80AFwtwJkl9x2eHWw6ZUqiPu/49Jm/OTw68UFMWTbThZ5l4pw - K5EGrfqfCeEWuOBM6iuxBy0aGHi73V889TAVk2loal1pZlkg6nX2DPbrFwrWLeIvPE4mhqYQbHIU - EdQEDWPSTXJhR2hL9OWCPYKDsoa3edsSe3OoGRO3dg5vhNrhRvA/jN5TEsLGjBqkHbmyWM7Hh68I - 4xkR86E/GhaB7RbOQdIiR7BVwKzhUyuNdG6RMTscWPh248LLJ8hC4Ksam3m94KCRHEViyD1p8Exv - OtwlKMPJoD2GZasrC/jVT+7fjsbiH11f/vlrZe6nbJZ7jEFxFVtkak8fjNb3pUNZwCI6fBM8zPv9 - u4Rcgj4h3T7ubHaurg9sJl/Cnx+1aMcugsojLYj5fnWMkWXHydf+WxInYweDP9YvF97D9ojC6DU2 - bDN+MuV5fPAkiMU+YQd7CGH53rvEA6VuSEBWVn0TcGTVHx4e5RnClnQiUtWPX0hbx7SULuJumKcD - GsStMJZQcCIfpXyqJLPYaBBeK4WiwCBNM9OracKfn3DcprigyrWLf/iNTpVkNaJiHnkFvKs32X/4 - xevE3tzCR/V00eVzv4Pu7lQQHu5xgSx1SAoxn3MbLj7WiP46pcZyyBAHr/P2Qo7Rh3qC+uZLIC6e - /rf/fOWNWv/4HbK9RwEWo4IlFKU3I5VmUePXr+BiFVP4wyPhxzc/QvlFp3YdR2qXK5vMYiQU+tVN - Q28fFoEth9z1rBTMwWqvrM8XPlY/lLaW+4L1VRSQc+58gzh+9YK1/gbho8v1gT/v/Rf84Vnc9wMb - f/xC81oJ7fXiy1gJchG87g+LBKP6TSauizJlR3YGlsHeL9izMEQoZOZ53Y+xoNxpy4PNnktRGFzn - hux8roR1zwosAcNmfN8/oeL2HSF6b7+axWDkCbQXh5AaRXIyLmTO4fswflAFIA9mxTtROO3cAHkb - x0wWzme9Ik7bw4qXooF5EHHK48H2SNNqpxAuu9yF7nV00Gnl61SQbjY8SmVOEv50Zaz9zi7YZCYj - xlTCYa3fEOq2ERPrjcRmvh/cEp4g65Gvi1bDK+3uCXN6RyhIBlIs7AyeYC+dfJJP0jXhf/h6XM5K - yDpZSJarx2PFinmNxBaYhvHEqzJ85dYZi65vegLbyRYwnocNcsLw04wrnoOPPsnk4HYcG3Eii0Da - uQPSnhc3mcWtosplzE3EyUd5mO+72YZcf3uQcCvEhnjbijJcx0MOLztGrrGhw5Wv4+8naxMW+/wL - BFk2o1x97otZ3qhP5XMQL0Sbb0djzvgnhl/L47A8Sde/+6n8+K4qWy9AvnUeQZtsKPJfCBmLvLvx - UG8eBxRZiu6t/a0EJ2f8kLLx1ETE3RyBXB3rn143lm9ucPBabeiqb+5g5gW3h476MtB+ePCJqOWu - Bc9juEeavNEA+/lX4OgPyFr9DCpcnRiaMT5isHFeBSW7KFZ+eoCDT76gGnxQyKdNhbRb3xlzJ2cu - UOu9gexTsBL2UuPB+dFJmHCRWBDNE1vods8Yv8r3PfleCYjBtbFCpA6Pd7Fk14D764+iTAoNERe1 - qqz9i1yb4eWxCFAZTlWXIa24B95Y1papLMXTJVXdvgDNcJn98Boz0tiAbUaSQ1TvKQr0CibYO+96 - eb0/Cno/b5b+OS1g9f+x2HzVQai8WYbn08YLl8374U1FJb9g6XzO4W7Yd80SF6oJToeXi/LVXx// - 8q+u18lhC8th4ST7Kf/qrxrVb0HE/RXDRzhqSPXu34TZV8zD7/1xxqws9USkKW3heL/ySG+LZ7Kk - 5+L1O3/kcg64ZFTjciuDg5TgRSrgwD4yXMDwYA0xuiw0xMumLpUKBxHST0NnrPgfQwo8B6HjJgKt - uDFO0IBD8JcPEfd7DIHMXlfkrn4un4TNDabmJBPzlAoAr/wHsnyvY3A0aTLE360Pr7N8wbNWf5PR - Z18ZvrbbG3LLw8agD6kK4ZrvYH7lG/NOkjgoJYcOf81JBGSHv09ocKxHh3JfNMPpbVPYnvfdr794 - NLacDg6mcySn53Vi35+fsetvMVn1dDIfBleGb3zgws89Ele9WNvK2m+RQz6EzbdTgv/iI6pfNOmc - osIwtDSDaK+pTqa2nyg4HVoXGbXzGkYroVCphNIj9zXfoHMsUNBNvUNcJxG9ZS/aHfzLZ5ZCMHAj - xxCeuifFkrQN2I+PQhpijDT69b3Zd4EsRwJ4ksP8zIypXx08/YQ5rCQDKvgGoBew2PLB1DzhZpbO - Wwx+flF6OFlAVPmxhSYX7om1Xjc4KJ9w9d/++gvUG1RL8dgiYnDQMmPZ3ef85y8S8zYOA3sdR05e - 9X64eVhSs0jzYinv9EmxsIAMUOs76nC9DrcYL2D1p1W4GS8TlolTN2SZoQwi7pSQVR+yNS8p5X5R - eWIdTa2Qjma+gK86Bcj8Hj1DktJzCp/3axruwvAwSC/HcOGryr/Iux+6ZLbbpAS/+4+PNT/86bNL - zN1CZd9dErYktar8+Lij7OaGnaa+BSqkmIRTMTXzmgfBjXc9kP11lo1ZpEkIwPv+DqXk5DZ0qK7j - Xz4XiyZNxpMDdbnMrjdkbLXS+OEv2MS9Hsrpw/Fw9QExCJvhvOZfnicc2AtD1D0faPWvCv4qir7S - FMKMDPVL2PQ5fEvQ6e4ZGWyJCzpUxxFOsHyQS9HwKx4srVxmxS0UZv1djFoujvBqxvtQSNuNR+MC - YiA4sR8uuzke+IXMGVjzMBLZGmXTD+/XPAEdghp6LLjLKVz9U3LUpfOAK3q2oN3XG6LOB5CQLVio - ElkuIzqEU0NKEPPKZg/TtV5T42+e85HeFvK7WzawgEM+5KflEsqr3sGf9Nru6HhWV785KtjprVI4 - LzZDdiAd2ZJuLzJ8ZNcJy9GUN/Tx6F15w9puzU+jYXj7Qg1ZUKTksOrl2XhJOTzigx5KznEZWHYv - I7hViYzph28Mxs2pCC9vCxDLa6JC4qeghKueC+nKF6fMPYswTqMmFFc/Z2kptMDqx4UyOpgN9ZxK - l8WO9X/zk/l1u7aAqdPw4/eAKInDwbrrOTz4T8eQfnhwv9cSMvbexVhswWiVjmtJKDVLNeBbkC1w - 1Z/IHLkpmc9evoUqXDDZn9f8WKaKDs7eh4Vba3MaaBVDEX5HpOFJocYgcHTxgeG0LvK0Qk8E4yVl - sGQgCZs17+LPudQp2T3fEjRw/DAkS6wrPz61d5LPmj9NnXx6+yWq1FcH5ueWN+FAvvs13xyb0T9y - InSPSUYCAHm2mNEQAvqYFxSkjd/8zTfcev8g/w0AAP//Sg6y9Esy1nfyCjRPzChyDwD1p6vKjUKD - DCxB4Rng4hoe7GJkXFHpag4eL3QNyHE1NzYOTwm3TDIMTQrw0M8p96kq8o/KtojQTg4uC/dxKoCO - l5hWumSUQdp34PJPCbwqoJZLQSEWvMIgNz8lNQe0MKAktaJEF75UQNdYtzg3MScHsgyhtDgxPVXJ - CroCQamgKD+3oCS+JD87Na9YyUrB0BCy1kCpJL8kMQdZnAtkVS0XAAAA//8DAODA7KPgIAAA - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99b82fe6fa72-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:53 GMT - Server: - - cloudflare - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-allow-origin: - - '*' - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-model: - - text-embedding-3-small - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '32' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '10000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '9999989' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 0s - x-request-id: - - req_7de2b85bdc15a7d05b7fedcfb8159af4 - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 14-15: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew - D. White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\nlanations - are adversarial training and matched\nmolecular pairs. Adversarial perturbations - are used during training to deceive the model\nto expose the vulnerabilities - of a model109,110 whereas counterfactuals are applied post-hoc.\nTherefore, - the main difference between adversarial and counterfactual examples are in the\napplication, - although both are derived from the same optimization problem.100 Grabocka\net - al. 111 have developed a method named Adversarial Training on EXplanations (ATEX)\nwhich - improves model robustness via exposure to adversarial examples. While there - are\nconceptual disparities, we note that the counterfactual and adversarial - explanations are\nequivalent mathematical objects.\nMatched molecular pairs - (MMPs) are pairs of molecules that differ structurally at only\none site by - a known transformation.112,113 MMPs are widely used in drug discovery and\nmedicinal - chemistry as these facilitate fast and easy understanding of structure-activity - re-\nlationships.114\u2013116 Counterfactuals and MMP examples intersect if - the structural change is\nassociated with a significant change in the properties. - In the case the associated changes in\nthe properties are non-significant, the - two molecules are known as bioisosteres.117,118 The con-\nnection between MMPs - and adversarial training examples has been explored by van Tilborg\net al. 119. - MMPs which belong to the counterfactual category are commonly used in outlier\nand - activity cliff detection.113 This approach is analogous to counterfactual explanations,\nas - the common objective is to uncover learned knowledge pertaining to structure-property\nrelationships.70\nApplications\nModel - interpretation is certainly not new and a common step in ML in chemistry, but - XAI for\nDL models is becoming more important60,66\u201369,73,88,104,105 Here - we illustrate some practical\nexamples drawn from our published work on how - model-agnostic XAI can be utilized to\n14\ninterpret black-box models and connect - the explanations to structure-property relationships.\nThe methods are \u201cMolecular - Model Agnostic Counterfactual Explanations\u201d (MMACE)9\nand \u201cExplaining - molecular properties with natural language\u201d.10 Then we demonstrate how\ncounterfactuals - and descriptor explanations can propose structure-property relationships in\nthe - domain of molecular scent.31\nBlood-brain barrier permeation prediction\nThe - passive diffusion of drugs from the blood stream to the brain is a critical - aspect in drug\ndevelopment and discovery.120 Small molecule blood-brain barrier - (BBB) permeation is a\nclassification problem routinely assessed with DL models.121,122 - To explain why DL models\nwork, we trained two models a random forest (RF) model123 - and a Gated Recurrent Unit\nRecurrent Neural Network (GRU-RNN). Then we explained - the RF model with generated\ncounterfactuals explanations using the MMACE9 and - the GRU-RNN with descriptor expla-\nnations.10 Both the models were trained - on the dataset developed by Martins et al. 124. The\nRF model was implemented - in Scikit-learn125 usi\n\n----\n\nQuestion: Are counterfactuals actionable? - [yes/no]\n\nDo not directly answer the question, instead summarize to give evidence - to help answer the question. Stay detailed; report specific numbers, equations, - or direct quotes (marked with quotation marks). Reply \"Not applicable\" if - the excerpt is irrelevant. At the end of your response, provide an integer score - from 1-10 on a newline indicating relevance to question. Do not explain your - score.\n\nRelevant Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", - "stream": false, "temperature": 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4257' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAAwAAAP//hFTBbhtHDL3rK4g9aw1JtmPFtzTHBmhSuKcmELgz1C7j2eF0yJWlBv73 - YEaK5DZIctGBb9+bx0dSX2YADfvmHho3oLkxhfbN27s/3l2v4+9P9OG3f/9699B9+HP/9H7f37rP - t828MKT7TM6+sa6cjCmQscQj7DKhUVFd3q3ubm6Xd8vrCoziKRRan6y9kXa1WN20i3W7eHUiDsKO - tLmHv2cAAF/qb7EYPe2be1jMv1VGUsWemvvzRwBNllAqDaqyGkZr5hfQSTSK1fXDQEB7RzkZeFY3 - qZKCDQSTEsgWnEzRKG/R2YRBgSOMEshNATOkTJ5daRdqQzqHgfshcD8Yx77ocIbi5cjzFICLXMpk - WHkYPajlydmUqU1ZEmU7QKZQcR04AUYMB2W9grf/c4OZAFMKTB6SqLWDuCpZAPpn4h0GigYmgH5H - WTEzBqA9ljkdu0EbaERjhwGM8qhX8DDQoUpMSr6Qiy9R+pVTnUPCbFzTCYciX6L0MiLHkuYlOnUU - bX72yrFIjxQNS0RA+xSQYwnxGNslap2DTm4AVOiCiG+7XNQ7zJkpQ6I8UnVU+/jJHF+88TQcwBMl - CIT58qrCk+RH0KnvSa3uxQEcxpLHjj0BVkfY1QFrGXtRNjmZ7mjAHUuubXpyrCyxHfGxvJCyOCrr - 9n1oPk99XUfZUT6c2DsKkkpCUPd3b3r1MX6M65ebnWk7KZbDilMIp/rz+VSC9ClLpyf8XN9yZB02 - mVAllrNQk9RU9HkG8Kme5PSfK2tSljHZxuSRYhFcv14d9ZrLn8AFXS7XJ9TEMLwAFsvFj3gbT4Yc - 9MVpN0ePHPuLxOJstHba6EGNxs2WY18ujY+Xvk2b1fZm8ap7vVxeN7Pn2VcAAAD//wMAO9coUvIE - AAA= - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99b9dc83176d-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:54 GMT - Server: - - cloudflare - Set-Cookie: - - __cf_bm=YSN79BcRAveJFGvOqV0R_nweoIuBpBSKTY3paJSWtWU-1727451714-1.0.1.1-6qdk2Jk4a0ZuH2ztuLubayTecZmIRiBA3RAWNl_o1zd7RytTwvzZ0sv4XI4iMgLk7fgDsBQSfmzLY1qs.CD0QQ; - path=/; expires=Fri, 27-Sep-24 16:11:54 GMT; domain=.api.openai.com; HttpOnly; - Secure; SameSite=None - - _cfuvid=nxb3prQChp_2F.4JnUZS09dFvJKTYYCKn_T8JvX.EMM-1727451714959-0.0.1.1-604800000; - path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '1165' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9998' - x-ratelimit-remaining-tokens: - - '29998973' - x-ratelimit-reset-requests: - - 9ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_8c4faba466f21bb9b8a0617e29284dab - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 35-36: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew - D. White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\n, - S.; An, J.; G\u00b4omez-Bombarelli, R. Chemistry-informed macromolecule\ngraph - representation for similarity computation, unsupervised and supervised learn-\ning. - Machine Learning: Science and Technology 2022, 3, 015028.\n(99) Doshi-Velez, - F.; Kortz, M.; Budish, R.; Bavitz, C.; Gershman, S.; O\u2019Brien, D.;\nScott, - K.; Schieber, S.; Waldo, J.; Weinberger, D.; Weller, A.; Wood, A. Account-\nability - of AI Under the Law: The Role of Explanation. SSRN Electronic Journal\n2017,\n(100) - Wachter, S.; Mittelstadt, B.; Russell, C. Counterfactual explanations without - opening\nthe black box: Automated decisions and the GDPR. Harv. JL & Tech. 2017, - 31, 841.\n(101) Jim\u00b4enez-Luna, J.; Grisoni, F.; Schneider, G. Drug discovery - with explainable artificial\nintelligence. Nature Machine Intelligence 2020 - 2:10 2020, 2, 573\u2013584.\n(102) Fu, T.; Gao, W.; Xiao, C.; Yasonik, J.; Coley, - C. W.; Sun, J. Differentiable Scaffold-\ning Tree for Molecule Optimization. - International Conference on Learning Represen-\ntations. 2022.\n(103) Shen, - C.; Krenn, M.; Eppel, S.; Aspuru-Guzik, A. Deep molecular dreaming: inverse\nmachine - learning for de-novo molecular design and interpretability with surjective\nrepresentations. - Machine Learning: Science and Technology 2021, 2, 03LT02.\n(104) Lucic,\nA.;\nter\nHoeve,\nM.;\nTolomei,\nG.;\nRijke,\nM.;\nSilvestri,\nF.\nCF-\nGNNExplainer:\nCounterfactual - Explanations for Graph Neural Networks. arXiv\npreprint arXiv:2102.03322 2021,\n(105) - Numeroso, D.; Bacciu, D. Explaining Deep Graph Networks with Molecular Counter-\nfactuals. - arXiv preprint arXiv:2011.05134 2020,\n35\n(106) Krenn, M.; H\u00a8ase, F.; - Nigam, A.; Friederich, P.; Aspuru-Guzik, A. Self-Referencing\nEmbedded Strings - (SELFIES): A 100% robust molecular string representation. Ma-\nchine Learning: - Science and Technology 2020, 1, 045024.\n(107) Nigam, A.; Pollice, R.; Krenn, - M.; dos Passos Gomes, G.; Aspuru-Guzik, A. Beyond\ngenerative models: superfast - traversal, optimization, novelty, exploration and discov-\nery (STONED) algorithm - for molecules using SELFIES. Chemical science 2021, 12,\n7079\u20137090.\n(108) - Kim, S.; Chen, J.; Cheng, T.; Gindulyte, A.; He, J.; He, S.; Li, Q.; Shoemaker, - B. A.;\nThiessen, P. A.; Yu, B.; Zaslavsky, L.; Zhang, J.; Bolton, E. E. PubChem - in 2021:\nnew data content and improved web interfaces. Nucleic Acids Research - 2020, 49,\nD1388\u2013D1395.\n(109) Tolomei, G.; Silvestri, F.; Haines, A.; - Lalmas, M. Interpretable predictions of tree-\nbased ensembles via actionable - feature tweaking. Proceedings of the 23rd ACM\nSIGKDD international conference - on knowledge discovery and data mining. 2017;\npp 465\u2013474.\n(110) Freiesleben, - T. The intriguing relation between counterfactual explanations and ad-\nversarial - examples. Minds and Machines 2022, 32, 77\u2013109.\n(111) Grabocka, J.; Schilling, - N.; Wistuba, M.; Schmidt-Thieme, L. Learning time-series\nshapelets. Proceedings - of the 20th ACM SIGKDD international conference on Knowl-\nedge discovery and - data mining. 2014; pp 392\u2013401.\n(112) Kenny, P. W.; Sadowski, J. Structure - modification in ch\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nDo - not directly answer the question, instead summarize to give evidence to help - answer the question. Stay detailed; report specific numbers, equations, or direct - quotes (marked with quotation marks). Reply \"Not applicable\" if the excerpt - is irrelevant. At the end of your response, provide an integer score from 1-10 - on a newline indicating relevance to question. Do not explain your score.\n\nRelevant - Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": - false, "temperature": 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4294' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAAwAAAP//fFRdaxsxEHz3r1juocRgG9txa8dv+aAh0KYmLbTQFLPW7d1topNUaS+x - CfnvRTrHuVKal3vY0Y5mZrX31APIOM+WkKkKRdVOD0/P518+Vxer4mz14/eNvTu/2u5OZpvxttjt - zrJB7LCbO1Ly0jVStnaahK1pYeUJhSLrZD6dz95P5pNZAmqbk45tpZPhzA6n4+lsOF4Mxx/2jZVl - RSFbws8eAMBT+kaJJqdttoTx4KVSUwhYUrY8HALIvNWxkmEIHASNZINXUFkjZJLqbxUBbRV5J+Cp - IE9GUYBAD+RRw6P19wE86egCxIKyjRHyBSppUANtnUaD0XAYwGPFqgL0BAg1SWVzKKwH5+0D52xK - 4NjrPAluWLPsgA2cXkEKI4zgqyPFBSvUejeA76gqIQ8kgHoER9PxZN6HnINqQnhLSGSViiDZ3ArY - ArARWycPOSkO6RSaPB27JJO8XqAgrLwVUpEGbqhsdGKEo8uL1U1/AKEpSwoSrUhF7MHZGCSjhhh4 - vFhTiRpQJX17myP41ChWHSfTST/df93U5G2w8A7OUCluEjjuJz/W05s2Y7aXHl0F19REB9ckaWAD - YJOzwo5QdE6z6sTevtRtm/0/gxnBR09MQdOGTNI0jZqwZkMhpZbeRJRRsYMNySOReVNttIv5A/mA - nhOIUcHh2dS4A66d3gGm/DtahXwd4hhbsd5umiCGQkv6MtFhjfdsytGtuTWL7nP3VDQB47aZRut9 - /fmwP9qWzttN2OOHesGGQ7X2hMGauCtBrMsS+twD+JX2tPlr9TLnbe1kLfaeTCScTBYnLWH2+mvo - wMeLPSpWUHeB6fx/feucBFmHzsJnrUg25SvF+KA0Wc3CLgjV64JNGUfN7f4Xbk3vaXYyOabFcdZ7 - 7v0BAAD//wMAqvgEMwgFAAA= - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99b9dc8b252d-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:55 GMT - Server: - - cloudflare - Set-Cookie: - - __cf_bm=oRXIDnhMIwrzVuWU43TNLxwOtps1lCgWkFEvurOX1ec-1727451715-1.0.1.1-mYe5OriqH2nTJCHOfNKVGgnv_Lr814JMGAyRoBJTE1d0ErDbM6dRXFJP1DQGt4Jyj5X.gCKiPYGQka4gk.fuEw; - path=/; expires=Fri, 27-Sep-24 16:11:55 GMT; domain=.api.openai.com; HttpOnly; - Secure; SameSite=None - - _cfuvid=UZKF9DkEQdqSzH0GSGBBSxSSOZ6dijLj3FNJS3PYWj4-1727451715847-0.0.1.1-604800000; - path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '1746' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29998970' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_cca139d54c9ab4e71b0d78555087d24a - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 11-12: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew - D. White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\n) - between f and g as the \u201cfidelity\u201d.\nGraphLIME87 and LIMEtree88 are - modifications to LIME as applicable to graph neural\nnetworks and regression - trees, respectively. LIME has been used in chemistry previously,\nsuch as Whitmore - et al. 89 who used LIME to explain octane number predictions of molecules\nfrom - a random forest classifier. Mehdi and Tiwary 90 used LIME to explain thermodynamic\ncontributions - of features. Gandhi and White 10 use an approach similar to GraphLIME,\nbut - use chemistry specific fragmentation and descriptors to explain molecular property - pre-\ndiction.\nSome examples are highlighted in the Applications section.\nIn - recent work by\nMehdi and Tiwary 90, a thermodynamic-based surrogate model approach - was used to inter-\npret black-box models. The authors define an \u201cinterpretation - free energy\u201d which can be\nachieved by minimizing the surrogate model\u2019s - uncertainty and maximizing simplicity.\nCounterfactual explanations\nCounterfactual - explanations can be found in many fields such as statistics, mathematics and\nphilosophy.91\u201394 - According to Woodward and Hitchcock 92, a counterfactual is an example\nwith - minimum deviation from the initial instance but with a contrasting outcome. - They\ncan be used to answer the question, \u201cwhich smallest change could - alter the outcome of an\ninstance of interest?\u201d While the difference between - the two instances is based on the exis-\ntence of similar worlds in philosophy,95 - a distance metric based on molecular similarity is\nemployed in XAI for chemistry. - For example, in the work by Wellawatte et al. 9 distance\nbetween two molecules - is defined as the Tanimoto distance96 between ECFP4 fingerprints.97\nAdditionally, - Mohapatra et al. 98 introduced a chemistry-informed graph representation for\ncomputing - macromolecular similarity. Contrastive explanations are peripheral to counterfac-\n11\ntual - explanations. Unlike the counterfactual approach, contrastive approach employ - a dual\noptimization method, which works by generating a similar and a dissimilar - (counterfactuals)\nexample. Contrastive explanations can interpret the model - by identifying contribution of\npresence and absence of subsets of features - towards a certain prediction.36,99\nA counterfactual x\u2032 of an instance - x is one with a dissimilar prediction \u02c6f(x) in classi-\nfication tasks. - As shown in equation 5, counterfactual generation can be thought of as a\nconstrained - optimization problem which minimizes the vector distance d(x, x\u2032) between - the\nfeatures.9,100\nminimize\nd(x, x\u2032)\nsuch that\n\u02c6f(x) \u0338= - \u02c6f(x\u2032)\n(5)\nFor regression tasks, equation 6 adapted from equation - 5 can be used. Here, a counter-\nfactual is one with a defined increase or decrease - in the prediction.\nminimize\nd(x, x\u2032)\nCounterfactuals explanations have - become a useful tool for XAI in chemistry, as they\nsuch that\n\f\f\f \u02c6f(x) - \u2212\u02c6f(x\u2032)\n\f\f\f \u2265\u2206\n(6)\nprovide intuitive understanding - of predictions and are able to uncover spurious relationships\nin training data.101 - Counterfactuals create local (instance-level), actionable explan\n\n----\n\nQuestion: - Are counterfactuals actionable? [yes/no]\n\nDo not directly answer the question, - instead summarize to give evidence to help answer the question. Stay detailed; - report specific numbers, equations, or direct quotes (marked with quotation - marks). Reply \"Not applicable\" if the excerpt is irrelevant. At the end of - your response, provide an integer score from 1-10 on a newline indicating relevance - to question. Do not explain your score.\n\nRelevant Information Summary (about - 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, "temperature": - 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4358' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAAwAAAP//dFRNbxs5DL37VxC6NAHswHbSfN2CNNgWWKAosIdd7BSGLHFm2Gioqchx - nE3z3xfSOLZ7yGWA4RMfHx9FvUwADHlzC8a1Vl3Xh9nd/dXXP+t/muj/uP72dfPz20Pnrp/uPz8t - P2+8meaMuP6BTt+yzlzs+oBKkUfYJbSKmXVxtby6+Li4WpwXoIseQ05rep1dxNlyvryYza9n88td - YhvJoZhb+HcCAPBSvlkie9yaW5hP3yIditgGze3+EIBJMeSIsSIkalnN9AC6yIpcVN/HgRVTbZ0O - NgBu+2DZ5gYEbELwKC7RGj1YAesyYNcBgRi0RShMW4VYw993X+DkIefTeOQuKdXkyAb4woohUIPs - 8BTqmMC12JFoej6Dv1p8LrUGQQ8awaNi6ogRKvPUkmtBOhsCioJrLTe57BA82KCYiow4qIsdZhmW - gTh37Mov5e5QdFoZIPbkrBI3OYnScT9sdUhYxABuHaZeRzOIBbS1mmseOTW60yBjygOG9TN0xNTR - fzt62KDTmMDTTkxVnfiT7RS2H06r6hTWqE+IDDWWyjIFGcpdyg7Yt0aJoU/oqQjNFFXVWn2pX0+2 - p1BVjD/hEBmJi7shz70u3UYGtfIoU7DsM8WvY47Zb/m/oKqawvkJg9o3uoRNQpE9VbaJBGRoGhR9 - x58+xQ35PENiHUhpgzCwx5T98NmlWB81J5UpAp1lGNjFDSaQfkgUB4GEYbyTLfUyhc4+7lzujmdo - tRi/H3/ADYaziiu+Ob79CetBbF4+HkLYxV/36xRi06e4lh2+j9fEJO0qoZXIeXVEY28K+joB+F7W - dvhtE02fYtfrSuMjcia8Of848pnDQ3FAF5fnO1Sj2nAEzG+u38tbeVRLQY7W34waiZsDxXwvtHRq - 5FkUu1VN3GDqE42vQd2vlvXF/HJ9s1icm8nr5H8AAAD//wMAxVwQjRYFAAA= - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99b9ca525c1c-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:56 GMT - Server: - - cloudflare - Set-Cookie: - - __cf_bm=IILXIMFwKOCQyKIe4t968EFupyUvRYsQBF.htt_DAKY-1727451716-1.0.1.1-F1yqwTS4CuwVLYANhXnSyGZFgAr0L.AzGv5eOsnzBFQ9NRL2_1YMX04RRJxatmgDNjoIdL1XQYOm7V.JI7BKPA; - path=/; expires=Fri, 27-Sep-24 16:11:56 GMT; domain=.api.openai.com; HttpOnly; - Secure; SameSite=None - - _cfuvid=b8zaetoDNOecUNc4DfrZLUwwHg_5euumayxjJ2pW7tA-1727451716043-0.0.1.1-604800000; - path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '2349' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9998' - x-ratelimit-remaining-tokens: - - '29998965' - x-ratelimit-reset-requests: - - 9ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_db32ef9a454b07da72281b0c98abd634 - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 12-14: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew - D. White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\n - relationships\nin training data.101 Counterfactuals create local (instance-level), - actionable explanations.\nActionability of an explanation suggest which features - can be altered to change the outcome.\nFor example, changing a hydrophobic functional - group in a molecule to a hydrophilic group\nto increase solubility.\nCounterfactual - generation is a demanding task as it requires gradient optimization over\ndiscrete - features that represents a molecule. Recent work by Fu et al. 102 and Shen et - al. 103\npresent two techniques which allow continuous gradient-based optimization. - Although, these\nmethodologies are shown to circumvent the issue of discrete - molecular optimization, counter-\nfactual explanation based model interpretation - still remains unexplored compared to other\n12\npost-hoc methods.\nCF-GNNExplainer104 - is a counterfactual explanation generating method based on GN-\nNExplainer69 - for graph data. This method generate counterfactuals by perturbing the input\ndata - (removing edges in the graph), and keeping account of perturbations which lead - to\nchanges in the output.\nHowever, this method is only applicable to graph-based - models\nand can generate infeasible molecular structures. Another related work - by Numeroso and\nBacciu 105 focus on generating counterfactual explanations - for deep graph networks. Their\nmethod MEG (Molecular counterfactual Explanation - Generator) uses a reinforcement learn-\ning based generator to create molecular - counterfactuals (molecular graphs).\nWhile this\nmethod is able to generate - counterfactuals through a multi-objective reinforcement learner,\nthis is not - a universal approach and requires training the generator for each task.\nWork - by Wellawatte et al. 9 present a model agnostic counterfactual generator MMACE\n(Molecular - Model Agnostic Counterfactual Explanations) which does not require training\nor - computing gradients. This method firstly populates a local chemical space through - ran-\ndom string mutations of SELFIES106 molecular representations using the - STONED algo-\nrithm.107 Next, the labels (predictions) of the molecules in the - local space are generated\nusing the model that needs to be explained. Finally, - the counterfactuals are identified and\nsorted by their similarities \u2013 - Tanimoto distance96 between ECFP4 fingerprints.97 Unlike the\nCF-GNNExplainer104 - and MEG105 methods, the MMACE algorithm ensures that generated\nmolecules are - valid, owing to the surjective property of SELFIES. Additionally, the MMACE\nmethod - can be applied to both regression and classification models. However, like most - XAI\nmethods for molecular prediction, MMACE does not account for the chemical - stability of\npredicted counterfactuals.\nTo circumvent this drawback, Wellawatte - et al. 9 propose an-\nother approach, which identift counterfactuals through - a similarity search on the PubChem\ndatabase.108\n13\nSimilarity to adjacent - fields\nTangential examples to counterfactual explanations are adversarial training - and matched\nmolecular pairs. Adversarial perturbations are used d\n\n----\n\nQuestion: - Are counterfactuals actionable? [yes/no]\n\nDo not directly answer the question, - instead summarize to give evidence to help answer the question. Stay detailed; - report specific numbers, equations, or direct quotes (marked with quotation - marks). Reply \"Not applicable\" if the excerpt is irrelevant. At the end of - your response, provide an integer score from 1-10 on a newline indicating relevance - to question. Do not explain your score.\n\nRelevant Information Summary (about - 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, "temperature": - 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4236' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAA3xTTW/cOAy9z68gdOoCdjCZmWSS3IJsUvSQ9rBb9NAsBrRMW9qVRUGk0wyK/PdC - nq/ksL0YMB/59PhI/pwBGN+aGzDWodohhfr2bv3l8ba7vfz6bbVcLj7h8hv5h68RLxr3l6lKBTf/ - ktVD1ZnlIQVSz3EH20yoVFjP14v16uJ8fb6agIFbCqWsT1qvuF7MF6t6flXPL/eFjr0lMTfwfQYA - 8HP6FomxpRdzA/PqEBlIBHsyN8ckAJM5lIhBES+KUU11Ai1HpTipvuMxKuUOrY4YBDATtCQ2+4Za - QIGU+dm3PvbwZAJbDPDBx8JoqQ70TOGPCtCWjrEJBPSSAkYs/3L2ZOBvR0fYB69b4A4Q7Ltn31aB - F2hp4Ciai3XQbEHGvifRIuKH89ZBR6hjJgGLERoCDEqZWlAG6zD2BOoIeFTLA1Ugo3Wll4Fb320L - DYLbtpmT48Zb6Ma4kxigzzwm8BEQBg5kx0CF9Zjvg7fAcQr6WMYrBMJh3HV3NjVML5ZyUmi92FGE - BJ4xex4FBlLHrUDHGXqKlHHq6r0dUhXqME6u3z3UHz9/vi8G+Ui5gsf7jxVgbOHx8fbuvgJC6+CH - Vwet7zrKFBUwpcxoHcmUGfzgdT8T+JMkeZ0cEjoIqibDDrojKwmoQ/3NoOoGhVqYFhl8SUqZ9DjD - QCJTOpfBlLvA/YRYHWVILFo7tgcFZ0/xKV693dJM3ShYjiSOIezjr8e1D9ynzI3s8WO889GL25TJ - cCwrLsrJTOjrDOCf6bzGdxdjUuYh6Ub5P4qF8Gq93PGZ00Gf0PPFAVVWDCfg+vry/8o2LSn6IG+u - 1Owk+tifGOZHnVOjRraiNGw6H/vir98dbZc2F/OL+WJ5aZvWzF5nvwAAAP//AwCk3gJjvQQAAA== - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99b9dc2b67f9-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:56 GMT - Server: - - cloudflare - Set-Cookie: - - __cf_bm=wK.6ukfHg1BLQhOmEKuQGt_bKH23H_IyHb9et6bcrRk-1727451716-1.0.1.1-gF3JQa3K_8aTqBqVfjUjhCerwSA.mWieSmxg11iaMMZpQ6.vcJjTDvp7uDWw83y0BpYZLj3VW0EOX1H4aY8Ibg; - path=/; expires=Fri, 27-Sep-24 16:11:56 GMT; domain=.api.openai.com; HttpOnly; - Secure; SameSite=None - - _cfuvid=ssBRY_76Fejx4hGYkLB9Fim3aLOPd2XHAFZCyLFemj8-1727451716089-0.0.1.1-604800000; - path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '1940' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29998974' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_32ac99c82d71b5fd14ec43ff24a3ff1d - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 26-28: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew - D. White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\nability - and Transparency. 2018; pp 77\u201391.\n(20) Lapuschkin, S.; W\u00a8aldchen, - S.; Binder, A.; Montavon, G.; Samek, W.; M\u00a8uller, K.-R.\nUnmasking Clever - Hans predictors and assessing what machines really learn. Nature\ncommunications - 2019, 10, 1\u20138.\n(21) DeGrave, A. J.; Janizek, J. D.; Lee, S.-I. AI for - radiographic COVID-19 detection\nselects shortcuts over signal. Nature Machine - Intelligence 2021, 3, 610\u2013619.\n(22) Goodman, B.; Flaxman, S. European - Union regulations on algorithmic decision-\nmaking and a \u201cright to explanation\u201d. - AI Magazine 2017, 38, 50\u201357.\n(23) ACT, A. I. European Commission. On Artificial - Intelligence: A European Approach\nto Excellence and Trust. 2021, COM/2021/206.\n(24) - Blueprint for an AI Bill of Rights, The White House. 2022; https://www.whitehouse.\ngov/ostp/ai-bill-of-rights/.\n(25) - Miller, T. Explanation in artificial intelligence: Insights from the social - sciences. Ar-\ntificial intelligence 2019, 267, 1\u201338.\n26\n(26) Murdoch, - W. J.; Singh, C.; Kumbier, K.; Abbasi-Asl, R.; Yu, B. Definitions, meth-\nods, - and applications in interpretable machine learning. Proceedings of the National\nAcademy - of Sciences of the United States of America 2019, 116, 22071\u201322080.\n(27) - Gunning, D.; Aha, D. DARPA\u2019s Explainable Artificial Intelligence (XAI) - Program.\nAI Magazine 2019, 40, 44\u201358.\n(28) Biran, O.; Cotton, C. Explanation - and justification in machine learning: A survey.\nIJCAI-17 workshop on explainable - AI (XAI). 2017; pp 8\u201313.\n(29) Palacio, S.; Lucieri, A.; Munir, M.; Ahmed, - S.; Hees, J.; Dengel, A. Xai handbook:\nTowards a unified framework for explainable - ai. Proceedings of the IEEE/CVF Inter-\nnational Conference on Computer Vision. - 2021; pp 3766\u20133775.\n(30) Kuhn, D. R.; Kacker, R. N.; Lei, Y.; Simos, D. - E. Combinatorial Methods for Ex-\nplainable AI. 2020 IEEE International Conference - on Software Testing, Verification\nand Validation Workshops (ICSTW) 2020, 167\u2013170.\n(31) - Seshadri, A.; Gandhi, H. A.; Wellawatte, G. P.; White, A. D. Why does that molecule\nsmell? - ChemRxiv 2022,\n(32) Das, A.; Rad, P. Opportunities and challenges in explainable - artificial intelligence\n(xai): A survey. arXiv preprint arXiv:2006.11371 2020,\n(33) - Machlev, R.; Heistrene, L.; Perl, M.; Levy, K. Y.; Belikov, J.; Mannor, S.; - Levron, Y.\nExplainable Artificial Intelligence (XAI) techniques for energy - and power systems:\nReview, challenges and opportunities. Energy and AI 2022, - 9, 100169.\n(34) Koh, P. W.; Liang, P. Understanding black-box predictions via - influence functions.\nInternational Conference on Machine Learning. 2017; pp - 1885\u20131894.\n(35) Ribeiro, M. T.; Singh, S.; Guestrin, C. \u201d Why should - i trust you?\u201d Explaining the\npredictions of any classifier. Proceedings - of the 22nd ACM SIGKDD international\n27\nconference on knowledge discovery - and data mining. San Diego, CA, USA, 2016; pp\n1135\u20131144.\n(36) Dhurandhar, - A.; Chen, P.-Y.; Luss, R.; Tu, C.-C.; Ting, P.; Shanmugam, K.; Das, P.\nExplanations - based on the missing: Towards contrastive explanations with pertinent\nnegatives. - Advances i\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nDo - not directly answer the question, instead summarize to give evidence to help - answer the question. Stay detailed; report specific numbers, equations, or direct - quotes (marked with quotation marks). Reply \"Not applicable\" if the excerpt - is irrelevant. At the end of your response, provide an integer score from 1-10 - on a newline indicating relevance to question. Do not explain your score.\n\nRelevant - Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": - false, "temperature": 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4328' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAA3SQzW6DMBCE7zyF5XOoMCFAuFVVT62I1FbqoamQYxbixtiubZRWUd694ieQHHrx - Yb6d8eyePIQwL3GGMNtTxxot/PuHZLN5Waft8ZC3/Pn4neePhLC39zh4fcKLzqF2X8DcxXXHVKMF - OK7kgJkB6qBLJUmYRCuSkLgHjSpBdLZaOz9SfhiEkR+kfhCPxr3iDCzO0IeHEEKn/u0qyhJ+cIaC - xUVpwFpaA86mIYSwUaJTMLWWW0elw4sZMiUdyL51rhyiWgvO6E7AVm4luZ40ULWWdkVlK8Son6ev - haq1UTs78kmvuOR2XxigVsnuG+uUxj09ewh99iu2N62xNqrRrnDqALILJCQNh0A8X3XG0cicclTc - uOL/XEUJjnJhry6Fh4pc1nNEMPXsF8X21zpoiorLGow2fDhcpQtYQbQmS0iX2Dt7fwAAAP//AwBR - H080QQIAAA== - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99c9cbc767f9-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:56 GMT - Server: - - cloudflare - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '132' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29998967' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_d59432c0fd036ecd248c7ef42ea11f55 - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 4-5: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew D. - White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\ncation, - interpretations can be categorized as global or local\ninterpretations on the - basis of \u201cwhat is being explained?\u201d. For example, counterfactuals - are\nlocal interpretations, as these can explain only a given instance. The - second classification is\nbased on the relation between the model and the interpretation - \u2013 is interpretability post-hoc\n(extrinsic) or intrinsic to the model?.32,33 - An intrinsic XAI method is part of the model\nand is self-explanatory32 These - are also referred to as white-box models to contrast them\nwith non-interpretable - black box models.28 An extrinsic method is one that can be applied\npost-training - to any model.33 Post-hoc methods found in the literature focus on interpreting\nmodels - through 1) training data34 and feature attribution,35 2) surrogate models10 - and, 3)\ncounterfactual9 or contrastive explanations.36\nOften, what is a \u201cgood\u201d - explanation and what are the required components of an ex-\nplanation are debated.32,37,38 - Palacio et al. 29 state that the lack of a standard framework\nhas caused the - inability to evaluate the interpretability of a model. In physical sciences,\nwe - may instead consider if the explanations somehow reflect and expand our understanding\nof - physical phenomena. For example, Oviedo et al. 39 propose that a model explanation\ncan - be evaluated by considering its agreement with physical observations, which - they term\n\u201ccorrectness.\u201d For example, if an explanation suggests - that polarity affects solubility of a\nmolecule, and the experimental evidence - strengthen the hypothesis, then the explanation\nis assumed \u201ccorrect\u201d. - In instances where such mechanistic knowledge is sparse, expert bi-\nases and - subjectivity can be used to measure the correctness.40 Other similar metrics - of\ncorrectness such as \u201cexplanation satisfaction scale\u201d can be found - in the literature.41,42 In a\nrecent study, Humer et al. 43 introduced CIME - an interactive web-based tool that allows the\nusers to inspect model explanations. - The aim of this study is to bridge the gap between\nanalysis of XAI methods. - Based on the above discussion, we identify that an agreed upon\n4\nevaluation - metric is necessary in XAI. We suggest the following attributes can be used - to\nevaluate explanations. However, the relative importance of each attribute - may depend on\nthe application - actionability may not be as important as faithfulness - when evaluating the\ninterpretability of a static physics based model. Therefore, - one can select relative importance\nof each attribute based on the application.\n\u2022 - Actionable. Is it clear how we could change the input features to modify the - output?\n\u2022 Complete. Does the explanation completely account for the prediction? - Did features\nnot included in the explanation really contribute zero effect - to the prediction?44\n\u2022 Correct. Does the explanation agree with hypothesized - or known underlying physical\nmechanism?39\n\u2022 Domain Applicable. Does the - explanation use language and concepts of domain ex-\nperts?\n\u2022 Fidelity/Faithful. - Does the explanation ag\n\n----\n\nQuestion: Are counterfactuals actionable? - [yes/no]\n\nDo not directly answer the question, instead summarize to give evidence - to help answer the question. Stay detailed; report specific numbers, equations, - or direct quotes (marked with quotation marks). Reply \"Not applicable\" if - the excerpt is irrelevant. At the end of your response, provide an integer score - from 1-10 on a newline indicating relevance to question. Do not explain your - score.\n\nRelevant Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", - "stream": false, "temperature": 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4304' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAAwAAAP//dFNdbxMxEHzPr1j5OamS0HyQF1SKEAiJSgjxUYKijb13t+Dznuy9JqHq - f0e+fLVUvNzDjmduZuy97wEYdmYBxlaotm784Op6dvPx9S2/5dp/233Rdzff30yvP9TXn8a3N6af - GbL+RVaPrAsrdeNJWcIetpFQKauOZuPZ5WQ0G006oBZHPtPKRgeXMhgPx5eD4XwwnB6IlbClZBbw - owcAcN99s8XgaGsWMOwfJzWlhCWZxekQgIni88RgSpwUg5r+GbQSlELn+nNFQFtLsVGwqFRK5D+U - gINSbCIp5jAJMEHpZY0eJIIXi74PG9YKrLT5aIFWW/QJ2FFQLphc5nQn/xXrQ00YOJSgFe2Ato1H - DiDB7wAhNWS5YAscsnFLF5BNKm0VHCfbpkQpMwFVI69bpQSFRKA79C1q1u0kw/FvHKxvXZ4vzZXN - Q1x76i8NbCq2FWD6nfqwNO8TsIL1hBEq2cCGcjrvwFYYSur+yaFpFQpCbWO2IVCL42LXgdJq0+qr - pYGvFfs94ViuE0oQRDtvbFn9DpKiEmwq0orisyYxEuDZbqfmqODAeQZSPImzNMB147nrBvWZ2j7J - OkcKiR3FfEEnNnCxv4wmyh07OrRQtuzyFYCEQ87c4vMO0FZMdwQIjhJn6UNl+Vq7Ui6WYRnmjx9h - pKJNmHcgtN4f5g+nV+2lbKKs0wE/zXP+VK0iYZKQX3BSaUyHPvQAfnbb0z5ZCNNEqRtdqfymkAXn - s+lez5z39YyORqMDqqLoz8DL+ex/tJUjRfbp0RKavUUO5VlhePLZBTVpl5TqVcGhzAvC+50smtVk - OBmOX0zt2pneQ+8vAAAA//8DAMNSLCScBAAA - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99c2b896176d-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:56 GMT - Server: - - cloudflare - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '1470' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29998968' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_75fc6debc124b86df16f89e61e53f2fe - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 5-7: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew D. - White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\nlanation - use language and concepts of domain ex-\nperts?\n\u2022 Fidelity/Faithful. Does - the explanation agree with the black box model?\n\u2022 Robust. Does the explanation - change significantly with small changes to the model or\ninstance being explained?\n\u2022 - Sparse/Succinct. Is the explanation succinct?\nWe present an example evaluation - of the SHAP explanation method based on the above\nattributes.44 Shapley values - were proposed as a local explanation method based on feature\nattribution, as - they offer a complete explanation - each feature is assigned a fraction of\nthe - prediction value.44,45 Completeness is a clearly measurable and well-defined - metric, but\nyields explanations with many components. Yet Shapley values are - not actionable nor sparse.\nThey are non-sparse as every feature has a non-zero - attribution and not-actionable because\nthey do not provide a set of features - which changes the outcome.46 Ribeiro et al. 35 proposed\na surrogate model method - that aims to provide sparse/succinct explanations that have high\n5\nfidelity - to the original model. In Wellawatte et al. 9 we argue that counterfactuals - are \u201cbet-\nter\u201d explanations because they are actionable and sparse. - We highlight that, evaluation of\nexplanations is a difficult task because explanations - are fundamentally for and by humans.\nTherefore, these evaluations are subjective, - as they depend on \u201ccomplex human factors and\napplication scenarios.\u201d37\nSelf-explaining - models\nA self-explanatory model is one that is intrinsically interpretable - to an expert.47 Two com-\nmon examples found in the literature are linear regression - models and decision trees (DT).\nIntrinsic models can be found in other XAI - applications acting as surrogate models (proxy\nmodels) due to their transparent - nature.48,49 A linear model is described by the equation\n1 where, W\u2019s - are the weight parameters and x\u2019s are the input features associated with - the\nprediction \u02c6y. Therefore, we observe that the weights can be used - to derive a complete expla-\nnation of the model - trained weights quantify - the importance of each feature.47 DT models\nare another type of self-explaining - models which have been used in classification and high-\nthroughput screening - tasks. Gajewicz et al. 50 used DT models to classify nanomaterials\nthat identify - structural features responsible for surface activity. In another study by Han\net - al. 51, a DT model was developed to filter compounds by their bioactivity based - on the\nchemical fingerprints.\n\u02c6y = \u03a3iWixi\n(1)\nRegularization techniques - such as EXPO52 and RRR53 are designed to enhance the black-\nbox model interpretability.54 - Although one can argue that \u201csimplicity\u201d of models are posi-\ntively - correlated with interpretability, this is based on how the interpretability - is evaluated.\nFor example, Lipton 55 argue that, from the notion of \u201csimulatability\u201d - (the degree to which a\nhuman can predict the outcome based on inputs), self-explanatory - linear models, rule-based\n6\nsystems, and DT\u2019s can be claimed uninterpretable. - A human can pred\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nDo - not directly answer the question, instead summarize to give evidence to help - answer the question. Stay detailed; report specific numbers, equations, or direct - quotes (marked with quotation marks). Reply \"Not applicable\" if the excerpt - is irrelevant. At the end of your response, provide an integer score from 1-10 - on a newline indicating relevance to question. Do not explain your score.\n\nRelevant - Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": - false, "temperature": 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4313' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAAwAAAP//dFNRb9MwEH7vrzj5Oa2aUtZtbwgJBEJCgkk8MFRdnEti5viM7wwraP8d - Oc26ToKXyLnvvvN35/v+LACMa801GDug2jH65avXu48fby43TLvf77//qLf28sXbLzefPry59L9M - VRjcfCerj6yV5TF6UsfhCNtEqFSq1rvNbvuy3tUXEzByS77Q+qjLLS836812ub5cri9m4sDOkphr - +LoAAPgzfYvE0NK9uYZ19RgZSQR7MtenJACT2JeIQREnikFN9QRaDkphUn0zENC9pRQVWic2i5CA - luhP9BlLK8Ad0H30GI6/I+nArUDHCUb2ZLPHBDFR6+wxoTQnFXRss7jQAwdA1eSarCTg3R1B51ry - Tg8VJG6yaCCRCjC0IBGTOD2s4J3CSKGULJJQ4fOA0dMBijSSCn4NzhPExD9dW+5BmB+AzgVXgIkg - sAJO+rDxBA1ZzEKl1QO0PMHcdZQAQaicoSPUnGi+2w4Y+ikfOKvlkVbwLkCZZULRakIw68BJAFOf - aeZxDkqpQ6sZvUxabk1DqpRuzblOeS6qJDasw7nq03xoBTeDE5Dc9yQq/77rOBk6r+CCuH5QgeYA - ri3T7Q5ldBLJus7Zuc/HghgAvVKaujt74nkEFYx4V+g60AhZqMt+2ouWrBPHYTnjMbGlslur23Ab - rs63MVGXBYsZQvZ+jj+c1ttzHxM3MuOneOeCk2GfCIVDWWVRjmZCHxYA3yYb5WfOMDHxGHWvfEeh - FLxazzYyT8Z9Quv6EVVW9GfAuv4vb9+SovNyZkdz1OhC/1RifRI6dWrkIErjvnOhpxSTO7qzi/tN - t11fNFd1/cIsHhZ/AQAA//8DACYReYOmBAAA - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99cc7f8867f9-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:57 GMT - Server: - - cloudflare - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '988' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29998968' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_2ef8dc3cb05442e7f16a209a9e6146cc - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 21-22: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew - D. White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\nscussion\nWe - have shown two post-hoc XAI applications based on molecular counterfactual expla-\nnations9 - and descriptor explanations.10 These methods can be used to explain black-box\nmodels - whose input is a molecule. These two methods can be applied for both classification\nand - regression tasks. Note that the \u201ccorrectness\u201d of the explanations - strongly depends on\nthe accuracy of the black-box model.\nA molecular counterfactual - is one with a minimal distance from a base molecular, but\nwith contrasting - chemical properties. In the above examples, we used Tanimoto similar-\nity96 - of ECFP4 fingreprints97 as distance, although this should be explored in the - future.\nCounterfactual explanations are useful because they are represented - as chemical structures\n(familiar to domain experts), sparse, and are actionable. - A few other popular examples of\ncounterfactual on graph methods are GNNExplainer, - MEG and CF-GNNExplainer.69,104,105\nThe descriptor explanation method developed - by Gandhi and White 10 fits a self-explaining\n21\nsurrogate model to explain - the black-box model. This is similar to the GraphLIME87 method,\nalthough we - have the flexibility to use explanation features other than subgraphs. Futher-\nmore, - we show that natural language combined with chemical descriptor attributions - can\ncreate explanations useful for chemists, thus enhancing the accessibility - of DL in chemistry.\nLastly, we examined if XAI can be used beyond interpretation. - Work by Seshadri et al. 31 use\nMMACE and surrogate model explanations to analyze - the structure-property relationships\nof scent. They recovered known structure-property - relationships for molecular scent purely\nfrom explanations, demonstrating the - usefulness of a two step process: fit an accurate model\nand then explain it.\nChoosing - among the plethora of XAI methods described here is still an open question.\nIt - remains to be seen if there will ever be a consensus benchmark, since this field - sits on\nthe intersection of human-machine interaction, machine learning, and - philosophy (i.e., what\nconstitutes an explanation?). Our current advice is - to consider first the audience \u2013 domain\nexperts or ML experts or non-experts - \u2013 and what the explanations should accomplish. Are\nthey meant to inform - data selection or model building, how a prediction is used, or how the\nfeatures - can be changed to affect the outcome. The second consideration is what access - you\nhave to the underlying model. The ability to have model derivatives or - propagate gradients\nto the input to models informs the XAI method.\nConclusion - and outlook\nWe should seek to explain molecular property prediction models - because users are more\nlikely to trust explained predictions, and explanations - can help assess if the model is learning\nthe correct underlying chemical principles. - We also showed that black-box modeling first,\nfollowed by XAI, is a path to - structure-property relationships without needing to trade\nbetween accuracy - and interpretability. However, XAI in chemistry has some maj\n\n----\n\nQuestion: - Are counterfactuals actionable? [yes/no]\n\nDo not directly answer the question, - instead summarize to give evidence to help answer the question. Stay detailed; - report specific numbers, equations, or direct quotes (marked with quotation - marks). Reply \"Not applicable\" if the excerpt is irrelevant. At the end of - your response, provide an integer score from 1-10 on a newline indicating relevance - to question. Do not explain your score.\n\nRelevant Information Summary (about - 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, "temperature": - 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4246' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAAwAAAP//fJNNaxtBDIbv/hViLm3BNrbjxIlvJQk0tCEU0kOpi5Fn5N2p54uRtrUJ - +e9ldh3bKW0vC6t3JD3ilZ56AMoaNQelaxTtkxu8v549PHwy+PH+Qxp9/nU3+RK+Xk9ubmez+9sb - 1S8ZcfWDtLxkDXX0yZHYGDpZZ0KhUnU8m8ym5+PZ+KIVfDTkSlqVZDCNg8loMh2MLgeji31iHa0m - VnP41gMAeGq/BTEY2qo5jPovEU/MWJGaHx4BqBxdiShktiwYRPWPoo5BKLTUjzUBbTXlJGAs64aZ - GKQmaJggrsFHR7pxmEHHJgjlNWpp0AFtk8OAZVYGZEDwJHU0ILHTbICVQ70ZrOIW2nkZbDgpmDIZ - q0sBEOQND+FOgAWlJUD5f8dMsFAN07pxsCKNhVdq2rVKppSJKQiZwqZr8lajA5bcaGkyMbxdo7fO - Yi7AJvrCS9tEWfhdHzhhZuoDBtMWxJYTV46GCwWPtWXgpqqI5a+sDBoDrAiMzaTF7QBTcpYMxAwe - g02NK4tReqOuLf0kMMQ2lxeN6OiJ++BxY0NVpvKnAHBqGjqOUNuqdraqpbPO+hSzYNCtgyWCWjcZ - 9e7l/w9jii+GhLK3Yd8RFkrHXOADMS/UPpPplQ390svtuhyUE0qwgTsiQ4mCgeJyTV2/NwyZnMWV - dVZ2w0VYhMvTBc20bhjLfYTGuX38+bDxLlYpxxXv9UN8bYPlepkJOYay3SwxqVZ97gF8by+reXUs - KuXokywlbiiUgpfnF109dbzlozoeT/eqREF3FK5mo3+lLQ0JWscnB6o6RBuqY4XRgbMdVPGOhfxy - bUNFOWXb3es6LemcplfjM7o8U73n3m8AAAD//wMAaAurb7gEAAA= - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99cd1de4176d-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:57 GMT - Server: - - cloudflare - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '987' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29998973' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_25abef45e5a8ca421f8900cf5d52a9e5 - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 15-17: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew - D. White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\n - on the dataset developed by Martins et al. 124. The\nRF model was implemented - in Scikit-learn125 using Mordred molecular descriptors126 as the\ninput features. - The GRU-RNN model was implemented in Keras.127 See Wellawatte et al. 9\nand - Gandhi and White 10 for more details.\nAccording to the counterfactuals of the - instance molecule in figure 1, we observe that the\nmodifications to the carboxylic - acid group enable the negative example molecule to permeate\nthe BBB. Experimental - findings by Fischer et al. 120 show that the BBB permeation of\nmolecules are - governed by hydrophobic interactions and surface area. The carboxylic group - is\na hydrophilic functional group which hinders hydrophobic interactions and - addition of atoms\nenhances the surface area. This proves the advantage of using - counterfactual explanations,\nas they suggest actionable modification to the - molecule to make it cross the BBB.\nIn Figure 2 we show descriptor explanations - generated for Alprozolam, a molecule that\npermeates the BBB, using the method - described by Gandhi and White 10.\nWe see that\npredicted permeability is positively - correlated with the aromaticity of the molecule, while\n15\nnegatively correlated - with the number of hydrogen bonds donors and acceptors. A similar\nstructure-property - relationship for BBB permeability is proposed in more mechanistic stud-\nies.128\u2013130 - The substructure attributions indicates a reduction in hydrogen bond donors - and\nacceptors. These descriptor explanations are quantitative and interpretable - by chemists.\nFinally, we can use a natural language model to summarize the - findings into a written\nexplanation, as shown in the printed text in Figure - 2.\nFigure 1: Counterfactuals of a molecule which cannot permeate the blood-brain - barrier.\nSimilarity is the Tanimoto similarity of ECFP4 fingerprints.131 Red - indicates deletions and\ngreen indicates substitutions and addition of atoms. - Republished from Ref.9 with permission\nfrom the Royal Society of Chemistry.\nSolubility - prediction\nSmall molecule solubility prediction is a classic cheminformatics - regression challenge and is\nimportant for chemical process design, drug design - and crystallization.133\u2013136 In our previous\nworks,9,10 we implemented - and trained an RNN model in Keras to predict solubilities (log\nmolarity) of - small molecules.127 The AqSolDB curated database137 was used to train the\nRNN - model.\nIn this task, counterfactuals are based on equation 6. Figure 3 illustrates - the generated\nlocal chemical space and the top four counterfactuals. Based - on the counterfactuals, we ob-\nserve that the modifications to the ester group - and other heteroatoms play an important role\nin solubility. These findings - align with known experimental and basic chemical intuition.134\nFigure 4 shows - a quantitative measurement of how substructures are contributing to the pre-\n16\nFigure - 2: Descriptor explanations along with natural language explanation obtained - for BBB\npermeability of Alprozolam molecule. The green and red bars show descriptors - t\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nDo not directly - answer the question, instead summarize to give evidence to help answer the question. - Stay detailed; report specific numbers, equations, or direct quotes (marked - with quotation marks). Reply \"Not applicable\" if the excerpt is irrelevant. - At the end of your response, provide an integer score from 1-10 on a newline - indicating relevance to question. Do not explain your score.\n\nRelevant Information - Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, - "temperature": 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4238' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAAwAAAP//fFRNb9tIDL37VxA6dQHZsBM3X7e4e2kuRbE97KJdGNSI0rAdDQczVBqj - yH9fULKTNMX2IgHDeeR7j+T8WABU3FY3UDmP6oYUlrfvLj98+Ha1u/vz/p/zvwM21I3jx+2d3NKB - qtoQ0nwlpyfUysmQAilLnMMuEypZ1s3l2eX27eZyczEFBmkpGKxPutzK8mx9tl2ur5briyPQCzsq - 1Q18XgAA/Ji+RjG29FDdwLo+nQxUCvZU3TxdAqiyBDupsBQuilGr+jnoJCrFifUnT0APjnJSSFnu - uaUCZL/oCNSjgpMxKuUOnY4YCmAmQGcasQm0gvcKRVGpzNcHabljh3ahgAqoJ3CYG3k4BHaAjlvo - s4wJOALCIIHcGOxOBJpyAqsBE+XB7JsyNEGkXTYZOUKDOTNleLPb7f6oAcsrjkAPKWA8Uihj31PR - F5zBeYw9TewoejSpc7GGA+thBX8lcpOKEA71VB/blg0P0gGqDAU4WncLWQWrTOYM1iD3lJ0MHPsJ - 6A9tluTZtEfUMZOleGXK5EcN3z07D956nMsJKQ07YJM3KygQyVnP8wE6ybDb7U5WscQV3B6Zztw5 - QpEwzsIgZWp5ylL/0leOrbWNftNGKkr52DyMLXhSyjLbYWPh8ugYQw0YuI/mwHdWb+2gzANFxQCd - 1Yl9mRI4T4O5bPpGnvl/8lRsKNFWqUBLg8Si+TQIabLBMJhSYHfsmZn6q6LTdGH+WdDqS/wSr1/u - RKZuLGgrGccQjuePT0sWpE9ZmnKMP513HLn4vQ2CRFuoopKqKfq4APh3Wubxp/2sUpYh6V7lG0VL - eHV9Meernp+P5+jm/PoYVVEMLwLr87f/h9u3pMihvHgUqpkjx/45xfqJ6KS0KoeiNOw7jj3llHl+ - I7q0P+u264vmerM5rxaPi/8AAAD//wMAbVHCmSwFAAA= - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99c969105c1c-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:58 GMT - Server: - - cloudflare - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '1889' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29998974' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_e2638adada7d4bd806097f0f99594bf7 - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Summarize - the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon - pages 20-21: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew - D. White. A perspective on explanations of molecular prediction models. Unknown - journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\ntics - Hamburg) de-\nveloped by Schomburg et al. 132.\n19\nthe \u2018fruity\u2019 scent. - There are some exceptions though, like tert-amyl acetate which has a\n\u2018camphoraceous\u2019 - rather than \u2018fruity\u2019 scent.140,141\nIn Seshadri et al. 31, we trained - a GNN model to predict the scent of molecules and utilized\ncounterfactuals9 - and descriptor explanations10 to quantify scent-structure relationships. The\nMMACE - method was modified to account for the multi-label aspect of scent prediction. - This\nmodification defines molecules that differed from the instance molecule - by only the selected\nscent as counterfactuals. For instance, counterfactuals - of the jasmone molecule would be false\nfor the \u2018jasmine\u2019 scent but - would still be positive for \u2018woody,\u2019 \u2018floral\u2019 and \u2018herbal\u2019 - scents.\nFigure 5: Counterfactual for the 2,4 decadienal molecule.\nThe counterfactual - indicates\nstructural changes to ethyl benzoate that would result in the model - predicting the molecule\nto not contain the \u2018fruity\u2019 scent. The Tanimoto96 - similarity between the counterfactual and\n2,4 decadienal is also provided. - Republished with permission from authors.31\nThe molecule 2,4-decadienal, which - is known to have a \u2018fatty\u2019 scent, is analyzed in Fig-\nure 5.142,143 - The resulting counterfactual, which has a shorter carbon chain and no carbonyl\ngroups, - highlights the influence of these structural features on the \u2018fatty\u2019 - scent of 2,4 deca-\ndienal. To generalize to other molecules, Seshadri et al. - 31 applied the descriptor attribution\nmethod to obtain global explanations - for the scents. The global explanation for the \u2018fatty\u2019\nscent was - generated by gathering chemical spaces around many \u2018fatty\u2019 scented - molecules.\nThe resulting natural language explanation is: \u201cThe molecular - property \u201cfatty scent\u201d can\nbe explained by the presence of a heptanyl - fragment, two CH2 groups separated by four\n20\nbonds, and a C=O double bond, - as well as the lack of more than one or two O atoms.\u201d31\nThe importance - of a heptanyl fragment aligns with that reported in the literature, as \u2018fatty\u2019\nmolecules - often have a long carbon chain.144 Furthermore, the importance of a C=O dou-\nble - bond is supported by the findings reported by Licon et al. 145, where in addition - to a\n\u201clarger carbon-chain skeleton\u201d, they found that \u2018fatty\u2019 - molecules also had \u201caldehyde or acid\nfunctions\u201d.145 For the \u2018pineapple\u2019 - scent, the following natural language explanation was ob-\ntained: \u201cThe - molecular property \u201cpineapple scent\u201d can be explained by the presence - of ester,\nethyl/ether O group, alkene/ether O group, and C=O double bond, as - well as the absence of\nan Aromatic atom.\u201d31 Esters, such as ethyl 2-methylbutyrate, - are present in many pineap-\nple volatile compounds.146,147 The combination - of a C=O double bond with an ether could\nalso correspond to an ester group. - Additionally, aldehydes and ketones, which contain C=O\ndouble bonds, are also - common in pineapple volatile compounds.146,148\nDiscussion\nWe have shown two - post-hoc XAI applications based on molecular counterfactual expla-\nnation\n\n----\n\nQuestion: - Are counterfactuals actionable? [yes/no]\n\nDo not directly answer the question, - instead summarize to give evidence to help answer the question. Stay detailed; - report specific numbers, equations, or direct quotes (marked with quotation - marks). Reply \"Not applicable\" if the excerpt is irrelevant. At the end of - your response, provide an integer score from 1-10 on a newline indicating relevance - to question. Do not explain your score.\n\nRelevant Information Summary (about - 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, "temperature": - 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '4438' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAA3RTy24bORC86ysac5YESVbixy0IkCwSxLuH7F7WgdBDNmeY5bAH7KZtJTDgz3B+ - z1+yICXLioNcBpiu6mb1o75PABpvmwtoTI9qhjHM3rw9/fOy++fjt/78wx/DX98+fn7/4dPf1+31 - p607a6Ylg9uvZPQpa254GAOp57iDTSJUKlWXp6vT9avl6fJVBQa2FEpaN+pszbPVYrWeLc5mi9f7 - xJ69IWku4N8JAMD3+i0So6Xb5gIW06fIQCLYUXNxIAE0iUOJNCjiRTFqM30GDUelWFV/7gno1lAa - FawXk0VIQHuCLATswHCOSsmh0YxBwEcYE1lv1McOBg5kcsAEYiiqQJYSRnh/eQm1xzm8fVEBE4El - 5yNZQHkqQQLWO0ep5LvEQxXhYxFv6MCCdgscw7aiQoGMkt09Pod3nIBusaxg+otwdjXnK8rA8ajg - DedgoSVwGITAcaq8x/uHQvWRHu9/7B6ANiuMLF799Y74eP9ww2y308J5vH9wgROG+ofRlkhPqcVw - KCFzKBPHiGErvopaTdczSwatp4gBpOebsgBUEE3ZaE4YwPQYO5IpSDZ9mRoWYlJKYDC1HAvDx/oq - tkJlZGV5FdsG6BLnUabgowu5ol6lKkbV7UFeUecFJHcdie5lvBykwXICfO0tAZpy69iGuirf9Spl - Qz5ab7BeiIxkvPPmuJuBbQlhyRWIRJYsKAOG0tBu1k9HxlHmV/Eqnh0fcCKXBYt/Yg5hH787OCJw - NyZuZY8f4s5HL/0mEQrHcv2iPDYVvZsAfKnOyz+ZqRkTD6NulP+jWAqen5/s6jXPXn9Gl+u9Lxtl - xXAELE9+m7expOiDHDm42Wn0sXsusTgIrZ02shWlYeN87CiNye8M7cbNyq0Xr9vz5fKkmdxN/gcA - AP//AwDzfvso2QQAAA== - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99c84da3252d-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:41:58 GMT - Server: - - cloudflare - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '2157' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29998954' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 2ms - x-request-id: - - req_6afb578836c2390464966ed7361e5365 - status: - code: 200 - message: OK -- request: - body: '{"messages": [{"role": "system", "content": "Answer in a direct and concise - tone. Your audience is an expert, so be highly specific. If there are ambiguous - terms or acronyms, first define them."}, {"role": "user", "content": "Answer - the question below with the context.\n\nContext (with relevance scores):\n\nwellawatteUnknownyearaperspectiveon - pages 11-12: Counterfactual explanations are described as actionable in the - context of XAI (Explainable Artificial Intelligence) for chemistry. They are - used to determine \"which smallest change could alter the outcome of an instance - of interest,\" indicating their actionable nature. The excerpt explains that - counterfactuals are generated by minimizing the vector distance \\(d(x, x'')\\) - between features, subject to a change in prediction \\(\\hat{f}(x) \\neq \\hat{f}(x'')\\) - for classification tasks, and \\(|\\hat{f}(x) - \\hat{f}(x'')| \\geq \\Delta\\) - for regression tasks. This suggests that counterfactuals provide \"intuitive - understanding of predictions\" and can uncover spurious relationships, making - them actionable at the instance level.\n\n9\nFrom Geemi P. Wellawatte, Heta - A. Gandhi, Aditi Seshadri, and Andrew D. White. A perspective on explanations - of molecular prediction models. Unknown journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, - doi:10.26434/chemrxiv-2022-qfv02.\n\nwellawatteUnknownyearaperspectiveon pages - 15-17: The excerpt provides evidence that counterfactuals are actionable. It - states that modifications to the carboxylic acid group in a molecule can enable - it to permeate the blood-brain barrier (BBB), as counterfactual explanations - suggest actionable changes to enhance permeability. Specifically, the addition - of atoms increases surface area, overcoming the hydrophilic nature of the carboxylic - group, which hinders hydrophobic interactions necessary for BBB permeation. - Additionally, in solubility prediction, counterfactuals indicate that modifications - to the ester group and heteroatoms are crucial, aligning with experimental findings - and chemical intuition. These examples demonstrate the practical applicability - of counterfactuals in molecular modifications.\n\n9\nFrom Geemi P. Wellawatte, - Heta A. Gandhi, Aditi Seshadri, and Andrew D. White. A perspective on explanations - of molecular prediction models. Unknown journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, - doi:10.26434/chemrxiv-2022-qfv02.\n\nwellawatteUnknownyearaperspectiveon pages - 5-7: The excerpt discusses the evaluation of explanation methods for molecular - prediction models, focusing on attributes like fidelity, robustness, and sparsity. - It mentions that Shapley values, while providing a complete explanation, are - not actionable because they do not offer a set of features that change the outcome. - In contrast, the authors argue that counterfactuals are \"better\" explanations - because they are both actionable and sparse. This suggests that counterfactuals - provide actionable insights by identifying specific changes that can alter the - prediction outcome, making them useful for decision-making processes.\n\n9\nFrom - Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew D. White. A - perspective on explanations of molecular prediction models. Unknown journal, - Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\nwellawatteUnknownyearaperspectiveon - pages 12-14: Counterfactuals are described as providing \"local (instance-level), - actionable explanations.\" The actionability of a counterfactual explanation - is demonstrated by suggesting which features can be altered to change the outcome, - such as modifying a hydrophobic functional group in a molecule to a hydrophilic - one to increase solubility. The excerpt discusses various methods for generating - counterfactuals, including CF-GNNExplainer, MEG, and MMACE, each with different - approaches and limitations. Despite these methods, the excerpt notes that counterfactual - explanation-based model interpretation is less explored compared to other post-hoc - methods.\n\n8\nFrom Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and - Andrew D. White. A perspective on explanations of molecular prediction models. - Unknown journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, - doi:10.26434/chemrxiv-2022-qfv02.\n\nwellawatteUnknownyearaperspectiveon pages - 35-36: The excerpt references several works related to counterfactual explanations, - which are a method for providing interpretability in AI models. Specifically, discuss - counterfactual explanations in the context of automated decisions and the General - Data Protection Regulation (GDPR), suggesting their potential role in legal - accountability. and Numeroso & Bacciu (2020) explore counterfactual explanations - for Graph Neural Networks, indicating their applicability in complex model interpretability. - Freiesleben (2022) examines the relationship between counterfactual explanations - and adversarial examples, which may imply actionability in terms of model robustness - and decision-making.\n\n8\nFrom Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, - and Andrew D. White. A perspective on explanations of molecular prediction models. - Unknown journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, - doi:10.26434/chemrxiv-2022-qfv02.\n\nValid keys: wellawatteUnknownyearaperspectiveon - pages 11-12, wellawatteUnknownyearaperspectiveon pages 15-17, wellawatteUnknownyearaperspectiveon - pages 5-7, wellawatteUnknownyearaperspectiveon pages 12-14, wellawatteUnknownyearaperspectiveon - pages 35-36\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nWrite - an answer based on the context. If the context provides insufficient information - reply \"I cannot answer.\"For each part of your answer, indicate which sources - most support it via citation keys at the end of sentences, like (Example2012Example - pages 3-4). Only cite from the context below and only use the valid keys. Write - in the style of a Wikipedia article, with concise sentences and coherent paragraphs. - The context comes from a variety of sources and is only a summary, so there - may inaccuracies or ambiguities. If quotes are present and relevant, use them - in the answer. This answer will go directly onto Wikipedia, so do not add any - extraneous information.\n\nAnswer (about 200 words, but can be longer):"}], - "model": "gpt-4o-2024-08-06", "stream": false, "temperature": 0.0}' - headers: - accept: - - application/json - accept-encoding: - - gzip, deflate - connection: - - keep-alive - content-length: - - '6452' - content-type: - - application/json - host: - - api.openai.com - user-agent: - - AsyncOpenAI/Python 1.46.1 - x-stainless-arch: - - arm64 - x-stainless-async: - - async:asyncio - x-stainless-lang: - - python - x-stainless-os: - - MacOS - x-stainless-package-version: - - 1.46.1 - x-stainless-raw-response: - - 'true' - x-stainless-runtime: - - CPython - x-stainless-runtime-version: - - 3.12.6 - method: POST - uri: https://api.openai.com/v1/chat/completions - response: - body: - string: !!binary | - H4sIAAAAAAAAA5RV22rcSBB991cUeophNIzG44zjNxM2YJa9hN3AhvViSt0lqexWt+gqzXgS/O9L - S5qLk+yCXwTquvSpU6e6vp4BZGyza8hMg2razuU379e/fdz8/PC5ePvrori8/f2j6Ntf3oWb4sF8 - ymYpIpQPZHQfNTeh7RwpBz+aTSRUSlmL9XK9uizWxdVgaIMll8LqTvNVyJeL5SpfXOWLt1NgE9iQ - ZNfw9xkAwNfhmyB6S0/ZNSxm+5OWRLCm7PrgBJDF4NJJhiIsil6z2dFoglfyA+rPJDMwofdKsUKj - PToBjARoUhVYOprDrQdtCIawJ4VQwU9PnUMezHATlSs2jA5uvZJzXJM3BG/+urk9hypEMA21LBp3 - 314FlPJ4TFeN11oSE7kkCygnGKAkg71QwrEDS0qxZT/8grToHImCadDXBJ7IkgUNgE4pDj6hVxNa - StDRA/tEiRl+OcEh0Tn82bAAp1sbpg1ZKHfQsueWv7CvhzQbMhoiWJ7iS9ItkYeKUPtIAtuGHQF5 - 6WOKwT0m9tBFsjwUNEtupoEuhg1bkhGS9qy8Iei9pZjy25QhVCeBydOCQQ+9N2FDEaTrI4deIJIb - WWy4E3izJedwi6r0yT/6sPU7wogdRenIpGuChw5rEiiKvFiez+/8nX//AxmY4IUtRbJwl5WkSvEu - e9m2pHmMI+MtaROsgONHgj8a7BztYIOuJ3nZwpS7DNqc9jgVJx1GodQM2gFb8srVDhLqpLGJTgFt - UAciji0+0rTv9gxafJxa10IvVPVu0KMlw8LB55O9i8GQCL2KuMt8fT6HDyECPWGa+lnqchscmd5h - hDbYBHkk6fsZk76uk2hP6p+Km4H0pkn6H1Ls9uIzGMvwtHNsAA1bqGPou0Q6+WZUowvB5mVE9lBi - jEwROootYcmOdQchjnyljCSJuCGHpCTcJjkSSHD96D4DdFz75LxlbVLTKXJLXtFBxYM+J0Wm+Tbo - 9ioO/lUKvMyL9ajAD33UhmIbIn3P2DQu4IJBNzsMce5oQ9+8JOVuz++Afpi2w4wm2ZQ0MjGqdhrS - k4fiRz2oej/2yp3S5tMTL6e8var2ZV6szge5C+2FJGCpDV40otKk7SSTxDB2nWOzb2gify+gqcXV - /z6x7GGD44sxveYyP90MkapeMC0m3zs3nT8fVo0LdRdDKZP9cF6xZ2nuExPBp7UiGrpssD6fAfwz - rLT+xZbKuhjaTu81PJJPCYvVxcWYMDtu0aN5ebWarBoU3Unculj/V9y9JUV2crIbsxEk+/qYYnFA - OpSayU6U2vuKfU2xizyuyqq7p0tavSsu6OoiO3s++xcAAP//AwBVC1SrMwgAAA== - headers: - CF-Cache-Status: - - DYNAMIC - CF-RAY: - - 8c9c99d72a4a22de-SJC - Connection: - - keep-alive - Content-Encoding: - - gzip - Content-Type: - - application/json - Date: - - Fri, 27 Sep 2024 15:42:02 GMT - Server: - - cloudflare - Set-Cookie: - - __cf_bm=LWPff5zX3ICbbT9QScjPn280Ah9S0inGrZ6ybIIewDI-1727451722-1.0.1.1-q5KgaN4N5L0dHLZN75tHjBW_AtJ5vsnl6zmfkPZuX3nQb6ofrTsGnI.tov8..L6pDl6EchlzSw2ix2jwjXBTjQ; - path=/; expires=Fri, 27-Sep-24 16:12:02 GMT; domain=.api.openai.com; HttpOnly; - Secure; SameSite=None - - _cfuvid=PWVdd0wya80sePSgKjRZyOELFVjKfBs7ecUlL0T1EI0-1727451722792-0.0.1.1-604800000; - path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None - Transfer-Encoding: - - chunked - X-Content-Type-Options: - - nosniff - access-control-expose-headers: - - X-Request-ID - alt-svc: - - h3=":443"; ma=86400 - openai-organization: - - future-house-xr4tdh - openai-processing-ms: - - '3986' - openai-version: - - '2020-10-01' - strict-transport-security: - - max-age=31536000; includeSubDomains; preload - x-ratelimit-limit-requests: - - '10000' - x-ratelimit-limit-tokens: - - '30000000' - x-ratelimit-remaining-requests: - - '9999' - x-ratelimit-remaining-tokens: - - '29998421' - x-ratelimit-reset-requests: - - 6ms - x-ratelimit-reset-tokens: - - 3ms - x-request-id: - - req_9670a7a1e5925640b8feb8354787ee0e - status: - code: 200 - message: OK + - request: + body: + '{"messages": [{"role": "user", "content": "Extract the title, authors, + and doi as a JSON from this MLA citation. If any field can not be found, return + it as null. Use title, authors, and doi as keys, author''s value should be a + list of authors. Wellawatte et al, A Perspective on Explanations of Molecular + Prediction Models, XAI Review, 2023\n\nCitation JSON:"}], "model": "gpt-4o-2024-08-06", + "stream": false, "temperature": 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "429" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAAwAAAP//dFHBbtpAEL3zFaM5Q2UMhIRbmkZtD1HTSlUOcYSX9WA2Xe9sd4dAi/j3 + aG0H0kMvc3hv3tN7M4cBAJoKF4B6o0Q33o6ub+bfvjzQTn2f3Gx2Uv/+8ffrx9tPq2r/+ecLDpOC + V8+k5U31QXPjLYlh19E6kBJKruN5Pp/OxvPsqiUarsgmWe1lNOVRnuXTUXY5yi564YaNpogLeBwA + ABzamSK6iva4gGz4hjQUo6oJF6clAAxsE4IqRhNFOcHhmdTshFybuizL58iucIfCARQoRiwVuIAC + r+GeQvSkxbwQsIPbvbfKqdQuAq/hji3prVUB7gNVRicC7lKxWOCw81Nb2XCIyfGxwAeyVu2UCAEJ + KFvgU79XsUk7bmtt4Y6FK8vyfeJA621Utt/o8ePpBJZrH3gVe/6Er40zcbMMpCK7VDcKe2zZ4wDg + qT319p/roQ/ceFkK/yKXDC8nnR2ef3smJ1c9KSzKnvFxnv9PtaxIlLHx3b+wC2hcfXbITinbmhj/ + RKFmuTaupuCD6d639stZNsvyyYVeVTg4Dl4BAAD//wMAjAmQ4scCAAA= + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c999d5d1867b2-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:49 GMT + Server: + - cloudflare + Set-Cookie: + - __cf_bm=gK44ohFwDd5WKiBoRjoA0YGKDmimgFNiQ8ukJRMnG.A-1727451709-1.0.1.1-V.Zf3Hhy8YwzkEJok22089NmBFWmCEmiPmP.GiqkV157_lXG1woQd1UeERTqCaqseJgfOhGmnEFH1L3FUqTMNA; + path=/; expires=Fri, 27-Sep-24 16:11:49 GMT; domain=.api.openai.com; HttpOnly; + Secure; SameSite=None + - _cfuvid=DoEGhZd5DcgECkMyAWnjKWwxp_c7OpMX6A9fxXzv9sM-1727451709777-0.0.1.1-604800000; + path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "506" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29999904" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 0s + x-request-id: + - req_74f9b5e3a661e39a92ed20c8b465b356 + status: + code: 200 + message: OK + - request: + body: null + headers: {} + method: GET + uri: https://api.crossref.org/works?mailto=example@papercrow.ai&query.title=A+Perspective+on+Explanations+of+Molecular+Prediction+Models&rows=1&query.author=Wellawatte+et+al&select=DOI,author,container-title,title + response: + body: + string: + '{"status":"ok","message-type":"work-list","message-version":"1.0.0","message":{"facets":{},"total-results":17288,"items":[{"DOI":"10.26434\/chemrxiv-2022-qfv02","author":[{"given":"Geemi + P.","family":"Wellawatte","sequence":"first","affiliation":[{"name":"University + of Rochester"}]},{"given":"Heta A.","family":"Gandhi","sequence":"additional","affiliation":[{"name":"University + of Rochester"}]},{"given":"Aditi","family":"Seshadri","sequence":"additional","affiliation":[{"name":"University + of Rochester"}]},{"ORCID":"http:\/\/orcid.org\/0000-0002-6647-3965","authenticated-orcid":false,"given":"Andrew + D.","family":"White","sequence":"additional","affiliation":[{"name":"University + of Rochester"}]}],"title":["A Perspective on Explanations of Molecular Prediction + Models"]}],"items-per-page":1,"query":{"start-index":0,"search-terms":null}}}' + headers: + Access-Control-Allow-Headers: + - X-Requested-With, Accept, Accept-Encoding, Accept-Charset, Accept-Language, + Accept-Ranges, Cache-Control + Access-Control-Allow-Origin: + - "*" + Access-Control-Expose-Headers: + - Link + Connection: + - close + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:51 GMT + Server: + - Jetty(9.4.40.v20210413) + Transfer-Encoding: + - chunked + permissions-policy: + - interest-cohort=() + x-api-pool: + - plus + x-rate-limit-interval: + - 1s + x-rate-limit-limit: + - "150" + x-ratelimit-interval: + - 1s + x-ratelimit-limit: + - "150" + status: + code: 200 + message: OK + - request: + body: + '{"input": ["A Perspective on Explanations of Molecular\nPrediction Models\nGeemi + P. Wellawatte,\u2020\nHeta A. Gandhi,\u2021\nAditi Seshadri,\u2021 and\nAndrew\nD. + White\u2217,\u2021\n\u2020Department of Chemistry, University of Rochester, + Rochester, NY, 14627\n\u2021Department of Chemical Engineering, University of + Rochester, Rochester, NY, 14627\n\u00b6Vial Health Technology, Inc., San Francisco, + CA 94111\nE-mail: andrew.white@rochester.edu\nAbstract\nChemists can be skeptical + in using deep learning (DL) in decision making, due to\nthe lack of interpretability + in \u201cblack-box\u201d models. Explainable artificial intelligence\n(XAI) + is a branch of AI which addresses this drawback by providing tools to interpret\nDL + models and their predictions. We review the principles of XAI in the domain + of\nchemistry and emerging methods for creating and evaluating explanations. + Then we\nfocus on methods developed by our group and their applications in predicting + solubil-\nity, blood-brain barrier permeability, and the scent of molecules. + We show that XAI\nmethods like chemical counterfactuals and descriptor explanations + can explain DL pre-\ndictions while giving insight into structure-property relationships. + Finally, we discuss\nhow a two-step process of developing a black-box model + and explaining predictions can\nuncover structure-property relationships.\n1\nIntroduction\nDeep + learning (DL) is advancing the boundaries of computational chemistry because + it can\naccurately model non-linear structure-function relationships.1\u20133 + Applications of DL can be\nfound in a broad spectrum spanning from quantum computing4,5 + to drug discovery6\u201310 to\nmaterials design.11,12 According to Kre 13, DL + models can contribute to scientific discovery\nin three \u201cdimensions\u201d + - 1) as a \u2018computational microscope\u2019 to gain insight which are not\nattainable + through experiments 2) as a \u2018resource of inspiration\u2019 to motivate + scientific thinking\n3) as an \u2018agent of understanding\u2019 to uncover + new observations. However, the rationale of\na DL prediction is not always apparent + due to the model architecture consisting a large\nparameter count.14,15 DL models + are thus often termed\u201cblack box\u201d models. We can only\nreason about + the input and output of an DL model, not the underlying cause that leads to\na + specific prediction.\nIt is routine in chemistry now for DL to exceed human + level performance \u2014 humans are\nnot good at predicting solubility from + structure for example161 \u2014 and so understanding how\na model makes predictions + can guide hypotheses. This is in contrast to a topic like finding\na stop sign + in an image, where there is little new to be learned about visual perception\nby + explaining a DL model. However, the black box nature of DL has its own limitations.\nUsers + are more likely to trust and use predictions from a model if they can understand + why\nthe prediction was made.17 Explaining predictions can help developers of + DL models ensure\nthe model is not learning spurious correlations.18,19 Two + infamous examples are, 1)neural\nnetworks that learned to recognize horses by + looking for a photogr", "o infamous examples are, 1)neural\nnetworks that learned + to recognize horses by looking for a photographer\u2019s watermark20 and,\n2) + neural networks that predicted a COVID-19 diagnoses by looking at the font choice\non + medical images.21 As a result, there is an emerging regulatory framework for + when any\ncomputer algorithms impact humans.22\u201324 Although we know of no + examples yet in chemistry,\none can assume the use of AI in predicting toxicity, + carcinogenicity, and environmental\npersistence will require rationale for the + predictions due to regulatory consequences.\n1there does happen to be one human + solubility savant, participant 11, who matched machine performance\n2\nEXplainable + Artificial Intelligence (XAI) is a field of growing importance that aims to\nprovide + model interpretations of DL predictions Three terms highly associated with XAI + are,\ninterpretability, justifications and explainability. Miller 25 defines + that interpretability of a\nmodel refers to the degree of human understandability + intrinsic within the model. Murdoch\net al. 26 clarify that interpretability + can be perceived as \u201cknowledge\u201d which provide insight\nto a particular + problem.\nJustifications are quantitative metrics tell the users \u201cwhy the\nmodel + should be trusted,\u201d like test error.27 Justifications are evidence which + defend why a\nprediction is trustworthy.25 An \u201cexplanation\u201d is a description + on why a certain prediction was\nmade.9,28 Interpretability and explanation + are often used interchangeably. Arrieta et al. 14\ndistinguish that interpretability + is a passive characteristic of a model, whereas explainability\nis an active + characteristic which is used to clarify the internal decision-making process.\nNamely, + an explanation is extra information that gives the context and a cause for one + or\nmore predictions.29 We adopt the same nomenclature in this perspective.\nAccuracy + and interpretability are two attractive characteristics of DL models. However,\nDL + models are often highly accurate and less interpretable.28,30 XAI provides a + way to avoid\nthat trade-off in chemical property prediction. XAI can be viewed + as a two-step process.\nFirst, we develop an accurate but uninterpretable DL + model. Next, we add explanations to\npredictions. Ideally, if the DL model has + correctly learned the input-output relations, then\nthe explanations should + give insight into the underlying mechanism.\nIn the remainder of this article, + we review recent approaches for XAI of chemical property\nprediction while drawing + specific examples from our recent XAI work.9,10,31 We show how\nin various systems + these methods yield explanations that are consistent with known and\nmechanisms + in structure-property relationships.\n3\nTheory\nIn this work, we aim to assemble + a common taxonomy for the landscape of XAI while\nproviding our perspectives. + We utilized the vocabulary proposed by Das and Rad 32 to classify\nXAI. According + to their classification, interpretations can be categorized as global or local\ninterpretations + on the basis of \u201cwhat ", "cation, interpretations can be categorized as + global or local\ninterpretations on the basis of \u201cwhat is being explained?\u201d. + For example, counterfactuals are\nlocal interpretations, as these can explain + only a given instance. The second classification is\nbased on the relation between + the model and the interpretation \u2013 is interpretability post-hoc\n(extrinsic) + or intrinsic to the model?.32,33 An intrinsic XAI method is part of the model\nand + is self-explanatory32 These are also referred to as white-box models to contrast + them\nwith non-interpretable black box models.28 An extrinsic method is one + that can be applied\npost-training to any model.33 Post-hoc methods found in + the literature focus on interpreting\nmodels through 1) training data34 and + feature attribution,35 2) surrogate models10 and, 3)\ncounterfactual9 or contrastive + explanations.36\nOften, what is a \u201cgood\u201d explanation and what are + the required components of an ex-\nplanation are debated.32,37,38 Palacio et + al. 29 state that the lack of a standard framework\nhas caused the inability + to evaluate the interpretability of a model. In physical sciences,\nwe may instead + consider if the explanations somehow reflect and expand our understanding\nof + physical phenomena. For example, Oviedo et al. 39 propose that a model explanation\ncan + be evaluated by considering its agreement with physical observations, which + they term\n\u201ccorrectness.\u201d For example, if an explanation suggests + that polarity affects solubility of a\nmolecule, and the experimental evidence + strengthen the hypothesis, then the explanation\nis assumed \u201ccorrect\u201d. + In instances where such mechanistic knowledge is sparse, expert bi-\nases and + subjectivity can be used to measure the correctness.40 Other similar metrics + of\ncorrectness such as \u201cexplanation satisfaction scale\u201d can be found + in the literature.41,42 In a\nrecent study, Humer et al. 43 introduced CIME + an interactive web-based tool that allows the\nusers to inspect model explanations. + The aim of this study is to bridge the gap between\nanalysis of XAI methods. + Based on the above discussion, we identify that an agreed upon\n4\nevaluation + metric is necessary in XAI. We suggest the following attributes can be used + to\nevaluate explanations. However, the relative importance of each attribute + may depend on\nthe application - actionability may not be as important as faithfulness + when evaluating the\ninterpretability of a static physics based model. Therefore, + one can select relative importance\nof each attribute based on the application.\n\u2022 + Actionable. Is it clear how we could change the input features to modify the + output?\n\u2022 Complete. Does the explanation completely account for the prediction? + Did features\nnot included in the explanation really contribute zero effect + to the prediction?44\n\u2022 Correct. Does the explanation agree with hypothesized + or known underlying physical\nmechanism?39\n\u2022 Domain Applicable. Does the + explanation use language and concepts of domain ex-\nperts?\n\u2022 Fidelity/Faithful. + Does the explanation ag", "lanation use language and concepts of domain ex-\nperts?\n\u2022 + Fidelity/Faithful. Does the explanation agree with the black box model?\n\u2022 + Robust. Does the explanation change significantly with small changes to the + model or\ninstance being explained?\n\u2022 Sparse/Succinct. Is the explanation + succinct?\nWe present an example evaluation of the SHAP explanation method based + on the above\nattributes.44 Shapley values were proposed as a local explanation + method based on feature\nattribution, as they offer a complete explanation - + each feature is assigned a fraction of\nthe prediction value.44,45 Completeness + is a clearly measurable and well-defined metric, but\nyields explanations with + many components. Yet Shapley values are not actionable nor sparse.\nThey are + non-sparse as every feature has a non-zero attribution and not-actionable because\nthey + do not provide a set of features which changes the outcome.46 Ribeiro et al. + 35 proposed\na surrogate model method that aims to provide sparse/succinct explanations + that have high\n5\nfidelity to the original model. In Wellawatte et al. 9 we + argue that counterfactuals are \u201cbet-\nter\u201d explanations because they + are actionable and sparse. We highlight that, evaluation of\nexplanations is + a difficult task because explanations are fundamentally for and by humans.\nTherefore, + these evaluations are subjective, as they depend on \u201ccomplex human factors + and\napplication scenarios.\u201d37\nSelf-explaining models\nA self-explanatory + model is one that is intrinsically interpretable to an expert.47 Two com-\nmon + examples found in the literature are linear regression models and decision trees + (DT).\nIntrinsic models can be found in other XAI applications acting as surrogate + models (proxy\nmodels) due to their transparent nature.48,49 A linear model + is described by the equation\n1 where, W\u2019s are the weight parameters and + x\u2019s are the input features associated with the\nprediction \u02c6y. Therefore, + we observe that the weights can be used to derive a complete expla-\nnation + of the model - trained weights quantify the importance of each feature.47 DT + models\nare another type of self-explaining models which have been used in classification + and high-\nthroughput screening tasks. Gajewicz et al. 50 used DT models to + classify nanomaterials\nthat identify structural features responsible for surface + activity. In another study by Han\net al. 51, a DT model was developed to filter + compounds by their bioactivity based on the\nchemical fingerprints.\n\u02c6y + = \u03a3iWixi\n(1)\nRegularization techniques such as EXPO52 and RRR53 are designed + to enhance the black-\nbox model interpretability.54 Although one can argue + that \u201csimplicity\u201d of models are posi-\ntively correlated with interpretability, + this is based on how the interpretability is evaluated.\nFor example, Lipton + 55 argue that, from the notion of \u201csimulatability\u201d (the degree to + which a\nhuman can predict the outcome based on inputs), self-explanatory linear + models, rule-based\n6\nsystems, and DT\u2019s can be claimed uninterpretable. + A human can pred", "atory linear models, rule-based\n6\nsystems, and DT\u2019s + can be claimed uninterpretable. A human can predict the outcome given\nthe inputs + only if the input features are interpretable. Therefore, a linear model which + takes\nin non-descriptive inputs may not be as transparent. On the other hand, + a linear model\nis not innately accurate as they fail to capture non-linear + relationships in data, limiting is\napplicability. Similarly, a DT is a rule-based + model and lacks physics informed knowledge.\nTherefore, an existing drawback + is the trade-off between the degree of understandability and\nthe accuracy of + a model. For example, an intrinsic model (linear regression or decision trees)\ncan + be described through the trainable parameters, but it may fail to \u201ccorrectly\u201d + capture\nnon-linear relations in the data.\nAttribution methods\nFeature attribution + methods explain black box predictions by assigning each input feature\na numerical + value, which indicates its importance or contribution to the prediction. Feature\nattributions + provide local explanations, but can be averaged or combined to explain multi-\nple + instances. Atom-based numerical assignments are commonly referred to as heatmaps.56\nSheridan + 57 describes an atom-wise attribution method for interpreting QSAR models. Re-\ncently, + Rasmussen et al. 58 showed that Crippen logP models serve as a benchmark for\nheatmap + approaches. Other most widely used feature attribution approaches in the litera-\nture + are gradient based methods,59,60 Shapley Additive exPlanations (SHAP),44 and + layer-\nwise relevance prorogation.61\nGradient based approaches are based on + the hypothesis that gradients for neural net-\nworks are analogous to coefficients + for regression models.62 Class activation maps (CAM),63\ngradCAM,64 smoothGrad,,65 + and integrated gradients62 are examples of this method. The\nmain idea behind + feature attributions with gradients can be represented with equation 2.\n\u2206\u02c6f(\u20d7x)\n\u2206xi\n\u2248\u2202\u02c6f(\u20d7x)\n\u2202xi\n(2)\n7\nwhere + \u02c6f(x) is the black-box model and\n\u2206\u02c6f(\u20d7x)\n\u2206xi\nare + used as our attributions. The left-\nhand side of equation 2 says that we attribute + each input feature xi by how much one unit\nchange in it would affect the output + of \u02c6f(x). If \u02c6f(x) is a linear surrogate model, then this\nmethod + reconciles with LIME.35 In DL models, \u2207xf(x), suffers from the shattered + gradients\nproblem.62 This means directly computing the quantity leads to numeric + problems. The\ndifferent gradient based approaches are mostly distinguishable + based on how the gradient is\napproximated.\nGradient based explanations have + been widely used to interpret chemistry predictions.60,66\u201370\nMcCloskey + et al. 60 used graph convolutional networks (GCNs) to predict protein-ligand\nbinding + and explained the binding logic for these predictions using integrated gradients.\nPope + et al. 66 and Jim\u00b4enez-Luna et al. 67 show application of gradCAM and integrated + gradi-\nents to explain molecular property predictions from trained graph neural + networks (GNNs).\nSanchez-Lengeling et al. 68 present compr", "rty predictions + from trained graph neural networks (GNNs).\nSanchez-Lengeling et al. 68 present + comprehensive, open-source XAI benchmarks to explain\nGNNs and other graph based + models. They compare the performance of class activation\nmaps (CAM),63 gradCAM,64 + smoothGrad,,65 integrated gradients62 and attention mecha-\nnisms for explaining + outcomes of classification as well as regression tasks. They concluded\nthat + CAM and integrated gradients perform well for graph based models. Another attempt\nat + creating XAI benchmarks for graph models was made by Rao et al. 70. They compared\nthese + gradient based methods to find subgraph importance when predicting activity + cliffs\nand concluded that gradCAM and integrated gradients provided the most + interpretability\nfor GNNs.\nThe GNNExplainer69 is an approach for generating + explanations (local and\nglobal) for graph based models. This method focuses + on identifying which sub-graphs con-\ntribute most to the prediction by maximizing + mutual information between the prediction\nand distribution of all possible + sub-graphs. Ying et al. 69 show that GNNExplainer can be\nused to obtain model-agnostic + explanations. SubgraphX is a similar method that explains\nGNN predictions by + identifying important subgraphs.71\nAnother set of approaches like DeepLIFT72 + and Layerwise Relevance backPropagation73\n8\n(LRP) are based on backpropagation + of the prediction scores through each layer of the neu-\nral network. The specific + backpropagation logic across various activation functions differs\nin these + approaches, which means each layer must have its own implementation. Baldas-\nsarre + and Azizpour 74 showed application of LRP to explain aqueous solubility prediction + for\nmolecules.\nSHAP is a model-agnostic feature attribution method that is + inspired from the game\ntheory concept of Shapley values.44,46 SHAP has been + popularly used in explaining molecular\nprediction models.75\u201378 It\u2019s + an additive feature contribution approach, which assumes that\nan explanation + model is a linear combination of binary variables z. If the Shapley value\ni + \u03d5i(\u20d7x)zi(\u20d7x). Shapley values for\nfeatures are computed using + Equation 3.79,80\nfor the ith feature is \u03d5i, then the explanation is \u02c6f(\u20d7x) + = P\n\u03d5i(\u20d7x) = 1\nM\nM\nX \u02c6f (\u20d7z+i) \u2212\u02c6f (\u20d7z\u2212i)\n(3)\nHere + \u20d7z is a fabricated example created from the original \u20d7x and a random + perturbation \u20d7x\u2032.\n\u20d7z+i has the feature i from \u20d7x and \u20d7z\u2212i + has the ith feature from \u20d7x\u2032. Some care should be taken\nin constructing + \u20d7z when working with molecular descriptors to ensure that an impossible + \u20d7z is\nnot sampled (e.g., high count of acid groups but no hydrogen bond + donors). M is the sample\nsize of perturbations around \u20d7x. Shapley value + computation is expensive, hence M is chosen\naccordingly. Equation 3 is an approximation + and gives contributions with an expectation\ni=1 \u03d5i(\u20d7x) = \u02c6f(\u20d7x).\nVisualization + based feature attribution has also been used for molecular data. In com-\nterm + as \u03d50 + P\nputer science, saliency maps are a way to measure spatial feature + contribution.8", "com-\nterm as \u03d50 + P\nputer science, saliency maps are + a way to measure spatial feature contribution.81 Simply put,\nsaliency maps + draw a connection between the model\u2019s neural fingerprint components (trained\nweights) + and input features. Weber et al. 82 used saliency maps to build an explainable + GCN\narchitecture that gives subgraph importance for small molecule activity + prediction. On the\nother hand, similarity maps compare model predictions for + two or more molecules based on\ntheir chemical fingerprints.83 Similarity maps + provide atomic weights or predicted probabil-\n9\nity difference between the + molecules by removing one atom at a time. These weights can\nthen be used to + color the molecular graph and give a visual presentation. ChemInformatics\nModel + Explorer (CIME) is an interactive web based toolkit which allows visualization + and\ncomparison of different explanation methods for molecular property prediction + models.84\nSurrogate models\nOne approach to explain black box predictions is + to fit a self-explaining or interpretable\nmodel to the black box model, in + the vicinity of one or a few specific examples. These are\nknown as surrogate + models. Generally, one model per explanation is trained. However, if we\ncould + find one surrogate model that explained the whole DL model, then we would simply\nhave + a globally accurate interpretable model. This means that the black-box model + is no\nlonger needed.79 In the work by White 79, a weighted least squares linear + model is used as\nthe surrogate model. This model provides natural language + based descriptor explanations by\nreplacing input features with chemically interpretable + descriptors. This approach is similar\nto the concept-based explanations approach + used by McGrath et al. 85, where human under-\nstandable concepts were used + in place of input features in acquisition of chess knowledge in\nAlphaZero. + Any of the self-explaining models detailed in the Self-explaining models section\ncan + be used as a surrogate model.\nThe most commonly used surrogate model based + method is Locally Interpretable Model\nExplanations (LIME).35 LIME creates perturbations + around the example of interest and fits\nan interpretable model to these local + perturbations. Ribeiro et al. 35 mathematically define\nan explanation \u03be + for an example \u20d7x using Equation 4.\n\u03be(\u20d7x) = arg min\ng\u2208G + L(f, g, \u03c0x) + \u2126(g)\n(4)\nHere f is the black box model and g \u2208G + is the interpretable explanation model. G is\na class of potential interpretable + models (e.g.: linear models). \u03c0x is a similarity measure\n10\nbetween original + input \u20d7x and it\u2019s perturbed input \u20d7x\u2032. In context of molecular + data, this can\nbe a chemical similarity metric like Tanimoto86 similarity between + fingerprints. The goal for\nLIME is to minimize the loss, L, such that f is + closely approximated by g. \u2126is a parameter\nthat controls the complexity + (sparsity) of g. Ribeiro et al. 35 termed the agreement (how low\nthe loss is) + between f and g as the \u201cfidelity\u201d.\nGraphLIME87 and LIMEtree88 are + modifications to LIME as applica", ") between f and g as the \u201cfidelity\u201d.\nGraphLIME87 + and LIMEtree88 are modifications to LIME as applicable to graph neural\nnetworks + and regression trees, respectively. LIME has been used in chemistry previously,\nsuch + as Whitmore et al. 89 who used LIME to explain octane number predictions of + molecules\nfrom a random forest classifier. Mehdi and Tiwary 90 used LIME to + explain thermodynamic\ncontributions of features. Gandhi and White 10 use an + approach similar to GraphLIME,\nbut use chemistry specific fragmentation and + descriptors to explain molecular property pre-\ndiction.\nSome examples are + highlighted in the Applications section.\nIn recent work by\nMehdi and Tiwary + 90, a thermodynamic-based surrogate model approach was used to inter-\npret + black-box models. The authors define an \u201cinterpretation free energy\u201d + which can be\nachieved by minimizing the surrogate model\u2019s uncertainty + and maximizing simplicity.\nCounterfactual explanations\nCounterfactual explanations + can be found in many fields such as statistics, mathematics and\nphilosophy.91\u201394 + According to Woodward and Hitchcock 92, a counterfactual is an example\nwith + minimum deviation from the initial instance but with a contrasting outcome. + They\ncan be used to answer the question, \u201cwhich smallest change could + alter the outcome of an\ninstance of interest?\u201d While the difference between + the two instances is based on the exis-\ntence of similar worlds in philosophy,95 + a distance metric based on molecular similarity is\nemployed in XAI for chemistry. + For example, in the work by Wellawatte et al. 9 distance\nbetween two molecules + is defined as the Tanimoto distance96 between ECFP4 fingerprints.97\nAdditionally, + Mohapatra et al. 98 introduced a chemistry-informed graph representation for\ncomputing + macromolecular similarity. Contrastive explanations are peripheral to counterfac-\n11\ntual + explanations. Unlike the counterfactual approach, contrastive approach employ + a dual\noptimization method, which works by generating a similar and a dissimilar + (counterfactuals)\nexample. Contrastive explanations can interpret the model + by identifying contribution of\npresence and absence of subsets of features + towards a certain prediction.36,99\nA counterfactual x\u2032 of an instance + x is one with a dissimilar prediction \u02c6f(x) in classi-\nfication tasks. + As shown in equation 5, counterfactual generation can be thought of as a\nconstrained + optimization problem which minimizes the vector distance d(x, x\u2032) between + the\nfeatures.9,100\nminimize\nd(x, x\u2032)\nsuch that\n\u02c6f(x) \u0338= + \u02c6f(x\u2032)\n(5)\nFor regression tasks, equation 6 adapted from equation + 5 can be used. Here, a counter-\nfactual is one with a defined increase or decrease + in the prediction.\nminimize\nd(x, x\u2032)\nCounterfactuals explanations have + become a useful tool for XAI in chemistry, as they\nsuch that\n\f\f\f \u02c6f(x) + \u2212\u02c6f(x\u2032)\n\f\f\f \u2265\u2206\n(6)\nprovide intuitive understanding + of predictions and are able to uncover spurious relationships\nin training data.101 + Counterfactuals create local (instance-level), actionable explan", " relationships\nin + training data.101 Counterfactuals create local (instance-level), actionable + explanations.\nActionability of an explanation suggest which features can be + altered to change the outcome.\nFor example, changing a hydrophobic functional + group in a molecule to a hydrophilic group\nto increase solubility.\nCounterfactual + generation is a demanding task as it requires gradient optimization over\ndiscrete + features that represents a molecule. Recent work by Fu et al. 102 and Shen et + al. 103\npresent two techniques which allow continuous gradient-based optimization. + Although, these\nmethodologies are shown to circumvent the issue of discrete + molecular optimization, counter-\nfactual explanation based model interpretation + still remains unexplored compared to other\n12\npost-hoc methods.\nCF-GNNExplainer104 + is a counterfactual explanation generating method based on GN-\nNExplainer69 + for graph data. This method generate counterfactuals by perturbing the input\ndata + (removing edges in the graph), and keeping account of perturbations which lead + to\nchanges in the output.\nHowever, this method is only applicable to graph-based + models\nand can generate infeasible molecular structures. Another related work + by Numeroso and\nBacciu 105 focus on generating counterfactual explanations + for deep graph networks. Their\nmethod MEG (Molecular counterfactual Explanation + Generator) uses a reinforcement learn-\ning based generator to create molecular + counterfactuals (molecular graphs).\nWhile this\nmethod is able to generate + counterfactuals through a multi-objective reinforcement learner,\nthis is not + a universal approach and requires training the generator for each task.\nWork + by Wellawatte et al. 9 present a model agnostic counterfactual generator MMACE\n(Molecular + Model Agnostic Counterfactual Explanations) which does not require training\nor + computing gradients. This method firstly populates a local chemical space through + ran-\ndom string mutations of SELFIES106 molecular representations using the + STONED algo-\nrithm.107 Next, the labels (predictions) of the molecules in the + local space are generated\nusing the model that needs to be explained. Finally, + the counterfactuals are identified and\nsorted by their similarities \u2013 + Tanimoto distance96 between ECFP4 fingerprints.97 Unlike the\nCF-GNNExplainer104 + and MEG105 methods, the MMACE algorithm ensures that generated\nmolecules are + valid, owing to the surjective property of SELFIES. Additionally, the MMACE\nmethod + can be applied to both regression and classification models. However, like most + XAI\nmethods for molecular prediction, MMACE does not account for the chemical + stability of\npredicted counterfactuals.\nTo circumvent this drawback, Wellawatte + et al. 9 propose an-\nother approach, which identift counterfactuals through + a similarity search on the PubChem\ndatabase.108\n13\nSimilarity to adjacent + fields\nTangential examples to counterfactual explanations are adversarial training + and matched\nmolecular pairs. Adversarial perturbations are used d", "lanations + are adversarial training and matched\nmolecular pairs. Adversarial perturbations + are used during training to deceive the model\nto expose the vulnerabilities + of a model109,110 whereas counterfactuals are applied post-hoc.\nTherefore, + the main difference between adversarial and counterfactual examples are in the\napplication, + although both are derived from the same optimization problem.100 Grabocka\net + al. 111 have developed a method named Adversarial Training on EXplanations (ATEX)\nwhich + improves model robustness via exposure to adversarial examples. While there + are\nconceptual disparities, we note that the counterfactual and adversarial + explanations are\nequivalent mathematical objects.\nMatched molecular pairs + (MMPs) are pairs of molecules that differ structurally at only\none site by + a known transformation.112,113 MMPs are widely used in drug discovery and\nmedicinal + chemistry as these facilitate fast and easy understanding of structure-activity + re-\nlationships.114\u2013116 Counterfactuals and MMP examples intersect if + the structural change is\nassociated with a significant change in the properties. + In the case the associated changes in\nthe properties are non-significant, the + two molecules are known as bioisosteres.117,118 The con-\nnection between MMPs + and adversarial training examples has been explored by van Tilborg\net al. 119. + MMPs which belong to the counterfactual category are commonly used in outlier\nand + activity cliff detection.113 This approach is analogous to counterfactual explanations,\nas + the common objective is to uncover learned knowledge pertaining to structure-property\nrelationships.70\nApplications\nModel + interpretation is certainly not new and a common step in ML in chemistry, but + XAI for\nDL models is becoming more important60,66\u201369,73,88,104,105 Here + we illustrate some practical\nexamples drawn from our published work on how + model-agnostic XAI can be utilized to\n14\ninterpret black-box models and connect + the explanations to structure-property relationships.\nThe methods are \u201cMolecular + Model Agnostic Counterfactual Explanations\u201d (MMACE)9\nand \u201cExplaining + molecular properties with natural language\u201d.10 Then we demonstrate how\ncounterfactuals + and descriptor explanations can propose structure-property relationships in\nthe + domain of molecular scent.31\nBlood-brain barrier permeation prediction\nThe + passive diffusion of drugs from the blood stream to the brain is a critical + aspect in drug\ndevelopment and discovery.120 Small molecule blood-brain barrier + (BBB) permeation is a\nclassification problem routinely assessed with DL models.121,122 + To explain why DL models\nwork, we trained two models a random forest (RF) model123 + and a Gated Recurrent Unit\nRecurrent Neural Network (GRU-RNN). Then we explained + the RF model with generated\ncounterfactuals explanations using the MMACE9 and + the GRU-RNN with descriptor expla-\nnations.10 Both the models were trained + on the dataset developed by Martins et al. 124. The\nRF model was implemented + in Scikit-learn125 usi", " on the dataset developed by Martins et al. 124. The\nRF + model was implemented in Scikit-learn125 using Mordred molecular descriptors126 + as the\ninput features. The GRU-RNN model was implemented in Keras.127 See Wellawatte + et al. 9\nand Gandhi and White 10 for more details.\nAccording to the counterfactuals + of the instance molecule in figure 1, we observe that the\nmodifications to + the carboxylic acid group enable the negative example molecule to permeate\nthe + BBB. Experimental findings by Fischer et al. 120 show that the BBB permeation + of\nmolecules are governed by hydrophobic interactions and surface area. The + carboxylic group is\na hydrophilic functional group which hinders hydrophobic + interactions and addition of atoms\nenhances the surface area. This proves the + advantage of using counterfactual explanations,\nas they suggest actionable + modification to the molecule to make it cross the BBB.\nIn Figure 2 we show + descriptor explanations generated for Alprozolam, a molecule that\npermeates + the BBB, using the method described by Gandhi and White 10.\nWe see that\npredicted + permeability is positively correlated with the aromaticity of the molecule, + while\n15\nnegatively correlated with the number of hydrogen bonds donors and + acceptors. A similar\nstructure-property relationship for BBB permeability is + proposed in more mechanistic stud-\nies.128\u2013130 The substructure attributions + indicates a reduction in hydrogen bond donors and\nacceptors. These descriptor + explanations are quantitative and interpretable by chemists.\nFinally, we can + use a natural language model to summarize the findings into a written\nexplanation, + as shown in the printed text in Figure 2.\nFigure 1: Counterfactuals of a molecule + which cannot permeate the blood-brain barrier.\nSimilarity is the Tanimoto similarity + of ECFP4 fingerprints.131 Red indicates deletions and\ngreen indicates substitutions + and addition of atoms. Republished from Ref.9 with permission\nfrom the Royal + Society of Chemistry.\nSolubility prediction\nSmall molecule solubility prediction + is a classic cheminformatics regression challenge and is\nimportant for chemical + process design, drug design and crystallization.133\u2013136 In our previous\nworks,9,10 + we implemented and trained an RNN model in Keras to predict solubilities (log\nmolarity) + of small molecules.127 The AqSolDB curated database137 was used to train the\nRNN + model.\nIn this task, counterfactuals are based on equation 6. Figure 3 illustrates + the generated\nlocal chemical space and the top four counterfactuals. Based + on the counterfactuals, we ob-\nserve that the modifications to the ester group + and other heteroatoms play an important role\nin solubility. These findings + align with known experimental and basic chemical intuition.134\nFigure 4 shows + a quantitative measurement of how substructures are contributing to the pre-\n16\nFigure + 2: Descriptor explanations along with natural language explanation obtained + for BBB\npermeability of Alprozolam molecule. The green and red bars show descriptors + t", "tion obtained for BBB\npermeability of Alprozolam molecule. The green and + red bars show descriptors that influ-\nence predictions positively and negatively, + respectively. Dotted yellow lines show significance\nthreshold (\u03b1 = 0.05) + for the t-statistic. Molecular descriptors show molecule-level proper-\nties + that are important for the prediction. ECFP and MACCS descriptors indicate which\nsubstructures + influence model predictions. MACCS explanations lead to text explanations\nas + shown. Republished from Ref.10 with permission from authors. SMARTS annotations + for\nMACCS descriptors were created using SMARTSviewer (smartsview.zbh.uni-hamburg.de,\nCopyright: + ZBH, Center for Bioinformatics Hamburg) developed by Schomburg et al. 132.\n17\ndiction. + For example, we see that adding acidic and basic groups as well as hydrogen + bond\nacceptors, increases solubility. Substructure importance from ECFP97 and + MACCS138 de-\nscriptors indicate that adding heteroatoms increases solubility, + while adding rings structures\nmakes the molecule less soluble. Although these + are established hypotheses, it is interesting\nto see they can be derived purely + from the data via DL and XAI.\nFigure 3: Generated chemical space for solubility + prediction using the RNN model. The\nchemical space is a 2D projection of the + pairwise Tanimoto similarities of the local coun-\nterfactuals. Each data point + is colored by solubility. Top 4 counterfactuals are shown here.\nRepublished + from Ref.9 with permission from the Royal Society of Chemistry.\nGeneralizing + XAI \u2013 interpreting scent-structure relationships\nIn this example, we show + how non-local structure-property relationships can be learned with\nXAI across + multiple molecules. Molecular scent prediction is a multi-label classification + task\nbecause a molecule can be described by more than one scent. For example, + the molecule\njasmone can be described as having \u2018jasmine,\u2019 \u2018woody,\u2019 + \u2018floral,\u2019 and \u2019herbal\u2019 scents.139 The\nscent-structure relationship + is not very well understood,140 although some relationships are\nknown. For + example, molecules with an ester functional group are often associated with\n18\nFigure + 4: Descriptor explanations for solubility prediction model. The green and red + bars\nshow descriptors that influence predictions positively and negatively, + respectively. Dotted\nyellow lines show significance threshold (\u03b1 = 0.05) + for the t-statistic. The MACCS and\nECFP descriptors indicate which substructures + influence model predictions. MACCS sub-\nstructures may either be present in + the molecule as is or may represent a modification. ECFP\nfingerprints are substructures + in the molecule that affect the prediction. MACCS descriptor\nare used to obtain + text explanations as shown. Republished from Ref.10 with permission from\nauthors. + SMARTS annotations for MACCS descriptors were created using SMARTSviewer\n(smartsview.zbh.uni-hamburg.de, + Copyright: ZBH, Center for Bioinformatics Hamburg) de-\nveloped by Schomburg + et al. 132.\n19\nthe \u2018fruity\u2019 scent. There are some exceptions ", + "tics Hamburg) de-\nveloped by Schomburg et al. 132.\n19\nthe \u2018fruity\u2019 + scent. There are some exceptions though, like tert-amyl acetate which has a\n\u2018camphoraceous\u2019 + rather than \u2018fruity\u2019 scent.140,141\nIn Seshadri et al. 31, we trained + a GNN model to predict the scent of molecules and utilized\ncounterfactuals9 + and descriptor explanations10 to quantify scent-structure relationships. The\nMMACE + method was modified to account for the multi-label aspect of scent prediction. + This\nmodification defines molecules that differed from the instance molecule + by only the selected\nscent as counterfactuals. For instance, counterfactuals + of the jasmone molecule would be false\nfor the \u2018jasmine\u2019 scent but + would still be positive for \u2018woody,\u2019 \u2018floral\u2019 and \u2018herbal\u2019 + scents.\nFigure 5: Counterfactual for the 2,4 decadienal molecule.\nThe counterfactual + indicates\nstructural changes to ethyl benzoate that would result in the model + predicting the molecule\nto not contain the \u2018fruity\u2019 scent. The Tanimoto96 + similarity between the counterfactual and\n2,4 decadienal is also provided. + Republished with permission from authors.31\nThe molecule 2,4-decadienal, which + is known to have a \u2018fatty\u2019 scent, is analyzed in Fig-\nure 5.142,143 + The resulting counterfactual, which has a shorter carbon chain and no carbonyl\ngroups, + highlights the influence of these structural features on the \u2018fatty\u2019 + scent of 2,4 deca-\ndienal. To generalize to other molecules, Seshadri et al. + 31 applied the descriptor attribution\nmethod to obtain global explanations + for the scents. The global explanation for the \u2018fatty\u2019\nscent was + generated by gathering chemical spaces around many \u2018fatty\u2019 scented + molecules.\nThe resulting natural language explanation is: \u201cThe molecular + property \u201cfatty scent\u201d can\nbe explained by the presence of a heptanyl + fragment, two CH2 groups separated by four\n20\nbonds, and a C=O double bond, + as well as the lack of more than one or two O atoms.\u201d31\nThe importance + of a heptanyl fragment aligns with that reported in the literature, as \u2018fatty\u2019\nmolecules + often have a long carbon chain.144 Furthermore, the importance of a C=O dou-\nble + bond is supported by the findings reported by Licon et al. 145, where in addition + to a\n\u201clarger carbon-chain skeleton\u201d, they found that \u2018fatty\u2019 + molecules also had \u201caldehyde or acid\nfunctions\u201d.145 For the \u2018pineapple\u2019 + scent, the following natural language explanation was ob-\ntained: \u201cThe + molecular property \u201cpineapple scent\u201d can be explained by the presence + of ester,\nethyl/ether O group, alkene/ether O group, and C=O double bond, as + well as the absence of\nan Aromatic atom.\u201d31 Esters, such as ethyl 2-methylbutyrate, + are present in many pineap-\nple volatile compounds.146,147 The combination + of a C=O double bond with an ether could\nalso correspond to an ester group. + Additionally, aldehydes and ketones, which contain C=O\ndouble bonds, are also + common in pineapple volatile compounds.146,148\nDiscussion\nWe have shown two + post-hoc XAI applications based on molecular counterfactual expla-\nnation", + "scussion\nWe have shown two post-hoc XAI applications based on molecular counterfactual + expla-\nnations9 and descriptor explanations.10 These methods can be used to + explain black-box\nmodels whose input is a molecule. These two methods can be + applied for both classification\nand regression tasks. Note that the \u201ccorrectness\u201d + of the explanations strongly depends on\nthe accuracy of the black-box model.\nA + molecular counterfactual is one with a minimal distance from a base molecular, + but\nwith contrasting chemical properties. In the above examples, we used Tanimoto + similar-\nity96 of ECFP4 fingreprints97 as distance, although this should be + explored in the future.\nCounterfactual explanations are useful because they + are represented as chemical structures\n(familiar to domain experts), sparse, + and are actionable. A few other popular examples of\ncounterfactual on graph + methods are GNNExplainer, MEG and CF-GNNExplainer.69,104,105\nThe descriptor + explanation method developed by Gandhi and White 10 fits a self-explaining\n21\nsurrogate + model to explain the black-box model. This is similar to the GraphLIME87 method,\nalthough + we have the flexibility to use explanation features other than subgraphs. Futher-\nmore, + we show that natural language combined with chemical descriptor attributions + can\ncreate explanations useful for chemists, thus enhancing the accessibility + of DL in chemistry.\nLastly, we examined if XAI can be used beyond interpretation. + Work by Seshadri et al. 31 use\nMMACE and surrogate model explanations to analyze + the structure-property relationships\nof scent. They recovered known structure-property + relationships for molecular scent purely\nfrom explanations, demonstrating the + usefulness of a two step process: fit an accurate model\nand then explain it.\nChoosing + among the plethora of XAI methods described here is still an open question.\nIt + remains to be seen if there will ever be a consensus benchmark, since this field + sits on\nthe intersection of human-machine interaction, machine learning, and + philosophy (i.e., what\nconstitutes an explanation?). Our current advice is + to consider first the audience \u2013 domain\nexperts or ML experts or non-experts + \u2013 and what the explanations should accomplish. Are\nthey meant to inform + data selection or model building, how a prediction is used, or how the\nfeatures + can be changed to affect the outcome. The second consideration is what access + you\nhave to the underlying model. The ability to have model derivatives or + propagate gradients\nto the input to models informs the XAI method.\nConclusion + and outlook\nWe should seek to explain molecular property prediction models + because users are more\nlikely to trust explained predictions, and explanations + can help assess if the model is learning\nthe correct underlying chemical principles. + We also showed that black-box modeling first,\nfollowed by XAI, is a path to + structure-property relationships without needing to trade\nbetween accuracy + and interpretability. However, XAI in chemistry has some maj", "thout needing + to trade\nbetween accuracy and interpretability. However, XAI in chemistry has + some major open\nquestions, that are also related to the black-box nature of + the deep learning.\nSome are\n22\nhighlighted below:\n\u2022 Explanation representation: + How is an explanation presented \u2013 text, a molecule, attri-\nbutions, a + concept, etc?\n\u2022 Molecular distance: in XAI approaches such as counterfactual + generation, the \u201cdis-\ntance\u201d between two molecules is minimized. + Molecular distance is subjective. Possibil-\nities are distance based on molecular + properties, synthesis routes, and direct structure\ncomparisons.\n\u2022 Regulations: + As black-box models move from research to industry, healthcare, and\nenvironmental + settings, we expect XAI to become more important to explain decisions\nto chemists + or non-experts and possibly be legally required. Explanations may need\nto be + tuned for be for doctors instead of chemists or to satisfy a legal requirement.\n\u2022 + Chemical space: Chemical space is the set of molecules that are realizable; + \u201crealiz-\nable\u201d can be defined from purchasable to synthesizable to + satisfied valences. What is\nmost useful? Can an explanation consider nearby + impossible molecules? How can we\ngenerate local chemical spaces centered around + a specific molecule for finding counter-\nfactuals or other instance explanations? + Similarly, can \u201cactivity cliffs\u201d be connected\nto explanations and + the local chemical space.149\n\u2022 Evaluating XAI : there is a lack of a systematic + framework (quantitative or qualitative)\nto evaluate correctness and applicability + of an explanation. Can there be a universal\nframework, or should explanations + be chosen and evaluated based on the audience and\ndomain? For example, work + by Rasmussen et al. 58 attempts to focus on comparing\nfeature attribution XAI + methods via Crippen\u2019s logP scores.\n23\nAcknowledgements\nResearch reported + in this work was supported by the National Institute of General Medical\nSciences + of the National Institutes of Health under award number R35GM137966. This work\nwas + supported by the NSF under awards 1751471 and 1764415. We thank the Center for\nIntegrated + Research Computing at the University of Rochester for providing computational\nresources.\nReferences\n(1) + Choudhary, K.; DeCost, B.; Chen, C.; Jain, A.; Tavazza, F.; Cohn, R.; Park, + C. W.;\nChoudhary, A.; Agrawal, A.; Billinge, S. J.; Holm, E.; Ong, S. P.; Wolverton, + C.\nRecent advances and applications of deep learning methods in materials science. + npj\nComputational Materials 2022, 8.\n(2) Keith, J. A.; Vassilev-Galindo, V.; + Cheng, B.; Chmiela, S.; Gastegger, M.; M\u00a8uller, K.-\nR.; Tkatchenko, A. + Combining Machine Learning and Computational Chemistry for\nPredictive Insights + Into Chemical Systems. Chemical Reviews 2021, 121, 9816\u20139872,\nPMID: 34232033.\n(3) + Goh, G. B.; Hodas, N. O.; Vishnu, A. Deep learning for computational chemistry.\nJournal + of Computational Chemistry 2017, 38, 1291\u20131307.\n(4) Deringer, V. L.; Caro, + M. A.; Cs\u00b4anyi, G. Machine Learning Interatomic Potentials as\nEmergi", + "307.\n(4) Deringer, V. L.; Caro, M. A.; Cs\u00b4anyi, G. Machine Learning Interatomic + Potentials as\nEmerging Tools for Materials Science. Advanced Materials 2019, + 31, 1902765.\n(5) Faber, F. A.; Hutchison, L.; Huang, B.; Gilmer, J.; Schoenholz, + S. S.; Dahl, G. E.;\nVinyals, O.; Kearnes, S.; Riley, P. F.; von Lilienfeld, + O. A. Prediction Errors of Molec-\nular Machine Learning Models Lower than Hybrid + DFT Error. Journal of Chemical\nTheory and Computation 2017, 13, 5255\u20135264, + PMID: 28926232.\n24\n(6) Duch, W.; Swaminathan, K.; Meller, J. Artificial Intelligence + Approaches for Rational\nDrug Design and Discovery. Current Pharmaceutical Design + 2007, 13, 1497\u20131508.\n(7) Dara, S.; Dhamercherla, S.; Jadav, S. S.; Babu, + C. M.; Ahsan, M. J.; darasuresh, S. D.;\nDara, S. Machine Learning in Drug Discovery: + A Review. Artificial Intelligence Review\n123, 55, 1947\u20131999.\n(8) Gupta, + R.; Srivastava, D.; Sahu, M.; Tiwari, S.; Ambasta, R. K.; Kumar, P. Artifi-\ncial + intelligence to deep learning: machine intelligence approach for drug discovery.\nMolecular + diversity 2021, 25, 1315\u20131360.\n(9) Wellawatte, G. P.; Seshadri, A.; White, + A. D. Model agnostic generation of counter-\nfactual explanations for molecules. + Chemical Science 2022, 13, 3697\u20133705.\n(10) Gandhi, H. A.; White, A. D. + Explaining structure-activity relationships using locally\nfaithful surrogate + models. chemrxiv 2022,\n(11) Gormley, A. J.; Webb, M. A. Machine learning in + combinatorial polymer chemistry.\nNature Reviews Materials 2021,\n(12) Gomes, + C. P.; Fink, D.; Dover, R. B. V.; Gregoire, J. M. Computational sustainability\nmeets + materials science. Nature Reviews Materials 2021,\n(13) On scientific understanding + with artificial intelligence. Nature Reviews Physics 2022\n4:12 2022, 4, 761\u2013769.\n(14) + Arrieta, A. B.; D\u00b4\u0131az-Rodr\u00b4\u0131guez, N.; Ser, J. D.; Bennetot, + A.; Tabik, S.; Barbado, A.;\nGarcia, S.; Gil-Lopez, S.; Molina, D.; Benjamins, + R.; Chatila, R.; Herrera, F. Explain-\nable Artificial Intelligence (XAI): Concepts, + Taxonomies, Opportunities and Chal-\nlenges toward Responsible AI. Information + Fusion 2019, 58, 82\u2013115.\n(15) Murdoch, W. J.; Singh, C.; Kumbier, K.; + Abbasi-Asl, R.; Yu, B. Interpretable machine\nlearning: definitions, methods, + and applications. ArXiv 2019, abs/1901.04592.\n25\n(16) Boobier, S.; Osbourn, + A.; Mitchell, J. B. Can human experts predict solubility better\nthan computers? + Journal of cheminformatics 2017, 9, 1\u201314.\n(17) Lee, J. D.; See, K. A. + Trust in automation: Designing for appropriate reliance. Human\nFactors 2004, + 46, 50\u201380.\n(18) Bolukbasi, T.; Chang, K.-W.; Zou, J. Y.; Saligrama, V.; + Kalai, A. T. Man is to com-\nputer programmer as woman is to homemaker? debiasing + word embeddings. Advances\nin neural information processing systems 2016, 29.\n(19) + Buolamwini, J.; Gebru, T. Gender Shades: Intersectional Accuracy Disparities + in\nCommercial Gender Classification. Proceedings of the 1st Conference on Fairness,\nAccountability + and Transparency. 2018; pp 77\u201391.\n(20) Lapuschkin, S.; W\u00a8aldchen, + S.; Binder, A.; Montavon, "], "model": "text-embedding-3-small", "encoding_format": + "base64"}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "50126" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/embeddings + response: + body: + string: !!binary | + H4sIAAAAAAAAA1SZS/OCvLOg9+dTvPVunSq5KEn+O+53iFxEnBUoIiiiQALkfPkp/Z2aqdm4AEql + k+5++sl//9c///zbl211mf79zz//Pptx+vd/fa9di6n49z///O//+ueff/7579/n//dk1ZXV9dq8 + 6t/jv5vN61ot//7nH+7/Xvl/D/3nn38hmCWsOfEDMHYdi72ynkSsdq6i8bDkGhi3vkUzQ321a/TJ + TJSFUocVCY/tbGl2jMAxjempShtvlq9lB8Tt60x9WbonyxJNPoTPPY/9Q/zWWIiKg+Sm8YEg12Js + MS5lD8dBfYfTozKH+dI6OmSwUmjQHhVtdo7RA44FZ+N43RTlSD+ggVery6hVzsmwBttNCqs4jLHc + LCc2m4X/gedXreJTiAtt8s46gdO6qlSGhcLEcGAqyurHGReFrgM+8YwP5GPpQzb0tmFs464x+tgX + DuNrFwzkzJQYVrEf0wNUg4HxVmwCR4Ax2V0ezrC4/q6C56J40nDgXCCE3MUFr+LAcLJGwrBMZRuj + 4xNe6eH7+zzAdQyjYcCEi85SsoJKa1CiWjtshDVMVqCoJto185Xe0mufzO+mP6DXnBb0lNu2t4be + roHLc9wT7lFO7QyDKEaXVlZotanebJqznQ+u3sbFvquZgygP7xUZt6NJ8aWehyXpsSm1KGqx5qOc + scc1sZHskQmrYkbaOesOMrqO7oBDajYeJ+6RBLvNxcS5bz7AOr4vF+iMU45DNN+9uYHgAco+NwjD + TswYfDkf8CnbHQ08cB44y7ldQNyGFj0jrx245ia56C2gFhtvNHrLq1p9dOx5mcbH1Nbmi+x2gBUz + DoeuDVum3VuCBo/L8Vm/VqWYZMUKSae9sH3/dO1C7nUKE9XYUe08OwPfNixEioN9qvloB97r6XMA + ikuuNPjuL26yVIiQh0oabh8dmOX20COfTzmabcQ1GQMvqWGwuVWhcBq75D29jAvC7mOgt2p0h1lq + 2wf8+z5gG8l8id4C+rwOGs7u3YVxZU98KDUepq5dt9qqnFiNbpLrkUchMTAHl3VFFZN07FSLPXCC + 27porTOeWtW0AYsrhC5UursbCt42HhjRByi9ZYCo+9R35ee1E3uoWy9C5dOil4vpkQrGrOhp8Msn + /hzmQBZsm+aCx9pJm2QBYV42af4xWm3C9bQDnbOs1P9sD1p3U6IeNWZf4MM2fbTLEG5NqG7ihqpv + /qhxwvTu4SPi91h+3eWSa9V5ROdXo4ainfsa22xzFXG4Najfs0wTfEPJkaxuDWo/XhEQDrm+oiFz + uHD7fPOAna4AQqvGbrj6XgN4/hwW4M57DjWX4cHmvg9y2J+WGB+bfhxmlX9DoBePA452WjzMFBYr + 2guuQj1zajR2FtwMDpo5Y2V/18D8W997oxuEX+JhWIsmNZGQzgj7a4oH7uWed2A/AAsb+yNj65vz + O9je0BM7fIUZ96jHENaCc8apMOXa7DPRhtR9wvAlBlrJt1ECYQ3lClvyUWPCqmEV2PPoET7czcPs + aFIGcktTaEDLZZibsJCgFgu7cCVg1dZbvJOQ0n5e1LexzGaiiBD6yVsNd5F/Z6zezRWMuHyHjQlr + 5Zxp5watspvibH6+wJqJMAN4PFxDIJ5fg6hquxBBMlzDOFYPjM84FiNyFE0sx0niEdGzXWkyxQxf + j5w1iGfr/QE3L4X47AA8cGxNCQqneKL+uQxL0X96O+mbbzRlqu6RzY1+4KV8f7Cj7F7lvPetA6z8 + k4+x6lzZ/JEqAtW7Y2LvyRtM9M/vERpP36e3eqezKTry6z76OBtqXx5OS0IDXKDijleq+l7DZqDc + XNjQR4uP5aAkq9xeCcw+U0K9aDq2c332VJicNohsDkoH5qGNM3TYfhTqjevikQs3FTDIyx31hvTW + rsXlJMP3JpAwlm6EzbckgLv32FsEjtd7u97ieQdf0jUN+22agyUUcxc5lWzT6zd/GJmsDggS1+Ki + z/DAqXSUgF/PN3phxdiy+4FAsFPMAYdGcfA+7yjUIX3tJVwetj7jBsI18LURbWqhy5ONg5wI6MV8 + hLGveoyTr0mHeiTsqHHkS7YCxTVBVYOGGlrds2XO5nAPTPrA4V7Oh/U+P3L4OhUH6m+UtqRbKZDh + tz9815dp9NuP9uL2eaaZw3YDiefERR0Vc0Ji8B7miR96qOzUktpSazGO69YagrU/4BtAtdbb43UD + S3BzKD6dBm1B9+XxV1+83U0HfPe4peBmIpv61yobxGHpG/gxHhG+haEBBNHQdHhO7JAIcnzTlm// + QZ97scHBxjgMvAB1CenonWL5Zmpg6c/Uh9lF1vClZ4LXu5nXw8Woc1pE774cp/hpI9zWPD0rdaIt + 92nI4JB2CrVO+a1c7eLeo9xSFHzgjSYRn8DskXAoHYwD2wYzeWwqMDqpSK9ffpiiHXf46zfXea95 + wq4ROnhtS41GI0m87/vUKJk4FefxUA6C4kfpr/7QnIV9Mmcl3kCXuyy0XN83T9T3tgzN1EJ/8RGy + RYyRErxk6mJjD94vwD6ojDRKlVWaNJaDV4XUvrxiZzvF3gJK9wJdMNn4UF525SrLtxC6rpLhUr7b + nlA/DjYCoqpjb3KfybyEow1+/cH0Xptkqi+6iaR32WGbhX3JHmQtoFZ2NfWVpPJWRVpleOTeE03E + ogSLIu42aEeKnur2gddmOdrncPsyc6rybtrO8jV5oIxaFpX19dwuqVE3KNSzB7WOj3NLbuBSg+IE + dOrHVav98hUs5HLHpmnrw1/+cPtNjcOI94fFljcZnHF8wH4+PdtlE/k5yFO7ovhbD7iAxDJyNuE7 + zE7p6lFtCnZQ3UuE+iqxNOF6vctI9buZSN2u0mh0rAmcLqQkvNx13tInuQyv7VnD2nl+D4uxLWz4 + 3FZZuE7WWI76A84AVheI8X2jMF7WEl+SZbOhzgWpTPDcrobNfr8l4qHkypUrrwX8xVfepo9h/RjD + Do5ZlWI5en6SmdeJDq/72KaBQ7fDUrsjB0/jq6PuIjrewp/NHJpDlWD/zNR2+fLCHy+VCp5Bf3vv + QkiUa0wdFvPDxNpNDUPvAfD5ghq2qPeQwP7E4nDKz5uBDUvdoPOdr8g7I3LJt/0uBgcjfJE6lxzA + aXndQLovPeqvKW2nzIwfgGeGR71pF2pj/lLN3f7RzzT/MA+s8VzakmZMJU2CY+DNssdlcNrkAb2p + DmIsgH0IlXnWqV3JwBsrGX5A7uwfWO9er2HsVtTswbi41L21m5aVAixgGsotNj4Dx+YHmj7AupOS + euLZapfbuD9AOXE6eusDvRS+fPvrf9S9gZixW3nS4dgkCjX8URjGX78eC8Em6MhZ7cLRXQ/X4kap + G0SpRjjMfJQUfE2Wtnkm84//+/1Np94mW8GUVz0ECixwmC9FM8zTx6ngLcQ3HAzatVzFfS/AJX9e + 8GnlscfCAaggR+aLCMH1oTFdb2JU+Ucfx18eoRZHTXAu8ic9xS9SMtfnG5BK2kL2+FyD6Srec9gc + jx8sf0wxme01kaF1ut3DLXgOgM79pYHQtWwiPDeRNgYXaYWnc93g8nhSPPF2v7ig21S/fvkEdBcq + ITy/LZ7agfQZZsEdXKCKqkcDMtlMjI71iH588uNN8Re/gsdq+N68jhrrHu4BnuKDR3M1EMBUPnMT + WvexpKfWYB5reu8ATWe+Y/+znT1yHHodRuEpwraXXbzlDZ8jbLaLHn7qxErErjJ8GHZyRk9kNRkj + zP9AWD7OX37UNH7SQyhJn51Ilq2MhgUd4gyVQ9wT8ZRvyxlVcw1JvGmxfjWJR/VMaSCq2yUUFxp7 + C9m8VxAmCkcDU/eGpdlvO7iNggQbh6zxyNV3VqhHQk8PstIP8wM9e6A/9iNJv/MSvWejCn68L9/M + FqymnWfwLI4tPkm7U8kr3ZlAAnYjKURy0lhpJyPiue5Cjab3W9rx7xqW9HPBf/23zyITkinZUEd4 + jGyV5VOIjNvJJCs2nm2TPOYNyhUPUv01PdrlKr5z6fnEV1IrltFOuyWdwY9X9YscecJpLGxYRgql + 9ocNYOGqcw5PwE/pTbCkYUTVrgYcpR01hmX55f8Idh+Zw26dvMr+oX1W8OMrF8ikZGAkK7wqNKLe + 5BoJC9OKwHayVKq12wHUifA8wK2Rd9iwvAugh/GpQ24lTgg60DD6Vl87OLdcRy1eeSfLrYxlST/r + W6pyl4Gx15w18EiIEQo8CTxBue46mMzACp+R/gEfb3qq8MWRMNw/5cGbquMFwh6eXt95MRxWYSIj + CPhJwM4FNWC4rayA3cuYqcs7dblq+dRBU/JaGh4oHT6+oRRIXT8nqr425tc3PHJ4DOSElnRMQK2f + 4wd8yoONvdfbG+bTPVCREEkFlUvOb//qd0XkDTk6SjNMkta5wLyIT2wje8MW4eZVsD/HLVYTewJr + Vkg7GBnPN87N5+BNl2XM4PW1ilTZvGpAOV8MQY+VNzY4dPubj+HzGVypC+SwXELx4MItz3rqfn3B + 6GhrCljwWbH1m38s7mXCgd1WMskbMjBlLDnQX/sTvW73WTLr5+IB75dwG27NvVnOie0/pG0zPnA5 + uUY5L0jn0AmEKfUKoWDrVtQ2iNMkicpeZrHpOw/uv/xCw1WbyrUJbxCScvTwGRrmsGQciOEFCwgr + uDh6U/aOdERs+g5F1p7Y58sXwApsgwb6wQDs1C4yqrKHSYsg4v6n/q16Y2MzMDDguSrKQXvbPkOp + sSaNFWdZQoNVPbDrt0OybtUuhHR/9rByz7yE/8UrntrmG++Affk8hXsrW2lIZEPj5luUgiDy7liG + xZ3Nb6JekOpZN+qHJmFLXK4ces1ZEfKWBxn57g/IeU5EHXyuGTvK9gW094uI5W++caumbOA5kxdc + wc9Dm19uJMFN9jzRcJR6728+KU2SUGO7Vz3Ock4VWL35hhNTntr5nF128OMLPQ338q5l/YPLpDdO + HfydH77rb+kw57kDPq4yV86qpRyglw7xz4d5vKoHMbgfHly4306xtgQ3wMFd6lHyHlDmcad1jGHU + gBe1vj6McJty94tXCBtogREJZAPu03jA5r2DbN6cnw38xjeceTZorH6dfQg2cRmi7T4r59eU5/CT + RvdwY6iADXTvENi9rJlcx3vIFjiVM5CtdB++w0Rj031qU9T0qU/xsX0mi+LkB3iOekTP52Fov36s + h6Wa3UION6PW02A10c+fWd5H0/hvPklUOz7DhRfz8juPFz/+Dpdz0nrzzphNVBzVe/jhHblc4ee0 + QllwbRzhgveIyr9kyPU3TDgYLmwRVFVGzfH0weZzE3k/vwF0AnV8iu5RuWD0SpH2SWYitYGrMTkV + a8jMvMTJ7SUOv/4PteVjhdtAiNpZeb84CKsKksm1EsarL28FKJ10ms73jTe/mzre4xOxp03fsO9+ + 36vQ0osr9r7zGT+OhwodrhtMtX0naLOwlSU0+rgI9+T+aJfDzvUR2csOvsz2cfjWUxOe9bb+xqf1 + Zsn8uJDDd4PMXdmWnNQOHfjOPzjx8ewtXWW7EKVUx9bxsW/Jd55Az+0lo44TdIw1+08DG1yoIR/p + H8ZOfjpK3lkxqXpMe40o0qpCdb8j2Bo3ijYdSnlGp7UasRW/wv/hVSHIdOpr3ENbnTIZkfcOXvib + bx6vle8Obp25oHjua9BtC7eGmr9dv35UZPOu2TeQMXLGlmGmTACVV8MrVUJsWvCTLGgDZLiQ6k7t + i38s52yR/N+8QiPS7EBvBHMOmi3TCXHrSWN7MZPQl+/xLYRh++WfGkbDG2Ovqs1EuFI5Rr//c+Qe + DVvl9kigBTUhBKNul0w76BXy2ljG2pAOyeofhg7ulttEvYdSe7Mat5c/nxdU+KGtNqpt9OXfEK4p + HQb/fCco59ItVfMt9GhnyRw8KoVNMU5VMAVLqcONtbOwkgedt3K7jINfX4N//piLPpGKHPO9UkxW + oL3J5j5LjSeH9CYfd94KgyqFidwYBOw3Vcvg8VP8/CN16j4Dy8azMhDdhIbqMIwAmRPOhz9f4wdy + xJZVqlfIH1QrBLeuTX71Hnx9C1aUfTm8rNOzB9/1xKYcb72fz/zjjVB03EQ0trcQetFMaSVHLlg3 + 1uLDrdpo2NM0YfjjbaVrXRoYWcPGrx8D3/sh4Kx5+MZ7hN/5LKTyMdeEOLh8wKvqXzTCq1POcUJW + sN20hy8/QLAgfedCS7potAL+2r4GcUvAYWPP+OZUqCRUkHskcKiixiqnJS8m91haXr5Lw7DlywmM + 7xEOo7ontaz07bIhgQAOGs2osWGGxvZ13MDIeL2xv48KMJY9CfcKjVccRPrNW1fNUmGZSRw1HEVt + Bctx8x8PEyQKRFv9Q/uA57tY0T+fcHXkETaVFZPxzNRhLRnLYOCMPdaPS9hyzjHqIFROdfjL1zlS + jhXIgvORWhrCJeuqwId+LgsUS2LrLXDHbyDSMAvfH+eQiCdZqiEPOIa1+iSX/HvRavh24ZHevI+Y + rC7e1/B7nkC9L7+wPjnqP39MZbPbJqwhdvbjW+peac5m8hAusGk3ORG3B69d12LI4HO8cRhv+1Rb + 3nAi0qDpMzW/PmeGj+MHSjs5pSe609ji26cGelt9pLKw773lnbombIaQ4kDIL4AdxkmHx8nncPnk + n4x9nl0uKTc9omoiZMPifdoNuO16CafRuSinHHwKWE/lHHZyfPOms7Uh8HYjMznvylKbb4mxQUny + dGilXnUw7evXA3apOmLDoUP5qAX/Ar4+E8tfXqdX7HKAPY0B68+7O6xrZzTwvsYddTbbd/k+lPIK + W6N3qZK2xTDuojhGl36x8M8vjfK1fMBCF2Sq3LMhmS5pI6FvfQ3Z4fVIlswsOsCXnxgrQUfLCbGc + g9W2xvR3PjBZppNKSSHW2Li6IFlbsWzg11eEvSE2CXvI51Ga641G9nhFbd9hUf3xNdZrbV/SnM8P + KDVXn4b3gCtJKJ196LpaFnYxROAtdY8PfJeuQOagUT3xgIcebBZ5xV7jFeU8RMyEU2hNWElbqZ1+ + Pvbbnynmzwv72Kh3f/MK9TrQgEVyWh8ZCQFYrtHq/d4PBiZn4K8/BWyPyg14vMYN9tb3Vlu9aVLh + 4a6kOA05rXy5j0hAYadmOJC7IpkvZIgBwFef9OJ2Xy4f0e6AkeuUhnFQg89W9DbAq3YevlwroWWX + MF+hmRoo3M2p8T98VCD4+c7fnLda4qDDTW721Nw+gTZ+fRGMhcijNhFS769fG8U2DrmAdOU6P5ML + CF5q98eHjEy4k25eBkOEn4Y2zQdmw+nhpH/+hEtsvwPZS5hDiXfkZDrPsv+bx8PBfszJ4ia4Bx0/ + VtikVvbzcRCKyTyFWy5JGWNyA2G6LgWWd3OWzJvz1OyjvPNCvng+SsY/yvrnA8je5tevL4pthM76 + LpTUqG7n3LU/MNhcK1zxx8cwDqedC49TyIXSra1aYomDCQ+CdMbq8JZA+yn8EN6G+5W6RugBAc0X + AfrvxiLuNldb4eo7M/z1Dw07KyOjvONgQ7sWh33D2ONN3As8EVOkeNtzHjECVIHa5ksa3KXcY8lL + kuH+vgZYsziWjLlr91C1QxiC5zVqv3xeQNyVORGTaSj7Jowl6btfSbK+P7/zPBcduWEie3XzSCY3 + wR/Y5lOI5Qd+J29VDw5/vGY/J8JYrccjOksvk8rcRLS1huEDboeLj1MZfPkHD70E1s8BG8DZDyyN + rBXiEN6oXvmDtvJ69McLOMey1fKcifTf+lDZvjy1sV3OOewi26H6l8+o66MaKEll4kBS7fI5Omco + ffcPduLM18RdqPjwiCUhFEMsefMuKmLIgn6lDmQ0Ec4HdYVd5Dpkc23khBOzbQ4xr5pkft3rhDWX + tw66y5pi55ycEmr7WgV+PinsmwTMW/Vtw9LTLZxOjxj8znshf4qrEM17TSPOsfrAb32gTmf23vrz + T3u9wli/5EY5C4fa//EVvp76KSG5GxBAxGZH8W5wvGE9vTJoTWmH7d0pAKJNHQjsF2y+/Lt4bERw + hbdm4cP7bXdmf/mfnCDCMTaeA7vdUxe5gNr0/wAAAP//nFrLroJMEn6X2TKJXO3updCICEhzE3EH + iALCQUAaupN598k5/3ZW8wKVVNV3SVU+3Li9yy33nsE6EQbi2ejTLvji5pr5KDm1t9yYfv9DCxof + +LxoeL5gfn3oKgBCUBAvPd9c5WetXlCk354SMU7A93G2F6DARabkXD2KLVAND/3rLxXwn3//H4kC + 6X8nCnT/HPra/XDEWzTMDtSrjRFvO1ut/Ox2BjxJt4HqMG2mNbk2Doo+cUdImgt8tDo1RwEWzzTv + KZjYJ44rdFJanZo338KrfLfeUPNCTkh5GTG1yq4BoHvUxP0GS8HZIX/Dk8uuC+9wMSn5/pChLpSB + L+AiLvgI4QxftZERw6E3l5u3Ooc/zZJRPccnPGv0K0JyVhpiCbYer7f5nMOim116kfU05uFXMqBf + mTJ1s1qNqd7EDmqFWCf6Jqtgi7f7AqOWE7+6ZBZX1KPYI9v0TV/p0DhtxyhoYDU8XBqZI8f8oS8j + 3I11Rsw3c6etHoQcbpVyokbdhoVkDncV3mIUkiiLy0n+ON0ebkc1pQHFX7zenDWD4mP40MCaLq1i + rp4MwqS3yTHYEGazkfpAKUuX3pPHAlhJbiuKhTyl5EFvYC37YgVq5ivEdRADi6suPnyE/ZUe9+Tj + rlUxHiBZRGkBeQ8LHjRdjobqiSkOgw/49iisIL9lCcGTvWtXW3Mg3HyvJ88z7DHtL/sDbL61QcJo + 8V1Z8j89nIT2hxipr8bf3wQBDNIBU/Ic+bRGqpLAkSavb4asquA3I3NgpoOU2tvPw1VCeXxDVEgJ + vVrZUEi54eYIdM/a31vH1qW1lqrwZ3Nf1BwTLWZ2FJro9u4YcWIrwizfVAdph8ORXC/fuhDlc7Cg + 5nB50hOZrXZNtYevHVUR0UuTeUAsqZ4i1n0FGr1dl89yLTogsfcHWpzSA5B3EitRz0lBT3cLAyar + Lxv1/FLQyKGKu7h6LAOhr0riyi+MOQobE+UJ9GjUB5QznfUJ5LHeUHK8U7D9VE2DQN1gQuTqwhVd + N2wU+9qd3sz9250tr0mQQGLqq9exdWftZuRI3hKdOHNjgVVJhhRW+KlSc4Ac8FBuepTApFh2v/Xp + GewTEIu1Tx0lORdsk14exKFv03MKDmArZl1A9rgrqc4B5Nty2o2wJ3ZFr/6ccV5r6R7h98GgUXJO + CtYfPwxga6mI/cI3vt1M8oaAIofELy2OJQNoPjyVuUwNq/O59LyoKvSeOSOWHDCXX5dIRXClDrEG + +ep+1n3LUHrMJmJs3p6z3IkHBEzlQO3WfLli1/YW+tUTak/AAhKJTBX98ZUY27WQ4xwYsDD2IbU+ + dzdmhYdmuEvymvzta/MtKYCnXn2RckqLYlXcPUMMRmdyGrwbZvvw00BWsxfBSxwUMt37JQzLfUZO + aXUu+LO650g5j6GvMTUHSrqKAkxCP6TWRY3a1V1BBCUcVP5+kgb8D5/fQvYk0e7suyyTqwwukb8j + xuN+dqVHNg6wjR3F35AaxOtPGkWIHPemj7rDT7Fh6btANahtWnJ/nJj7Mhh4cTmiJklTzoLWWNDo + vXrqtV3trq9eG2C6XEtfQ9m73bAQB0B+jwKxvNQEmzDZDjIXIpFLk818ZT81g/aISupLa19IUnry + 4b7Ou6UOeBQrc5wYaLd7TSRmsjUxfo9TqPtuSILa0vg3LoIUaV7Mic5gV0iWdzSQpKoRjXRcu8ot + 8g8wFluf4sl+tj8XGPhIQBeRYLudirk3dzakj91ITifxVrD+nS3QELILiZJ5bJnd+BA+mHKjDqjC + dtZuTg7WD4mpN95azmDaDShfrdvvvE6x1DFdRo16mMkzUUOX309JhOzznNNiy+WWe3wc4V70nkuv + PoWJpYEeoZB/Nmqup3fLnpe4R30o3OmdhsM/eAFtBzeSPTVS8DbVFyiQkP76h4LZHz7s8pWQ0zh+ + XF47zELht9HosXJz/FGlxICmJ3DieEoHZI5DASpSjqiOGcaiL7whFOPzgRg/DE0rH13xb38kIuW1 + Vdgh76EElZUSWx+LrVTCHjlt86HOdZnbdUxVByW1zOix/MwFU4Djg9DHkOKT7E2rlM4vuFO8glTO + Ohfbho4y8JPLQo9mbBZcfF8GyL396CvtdXW/UWDP4E9vAtfHQDHVfQ4HwH7IQTscJkVWDRH+HDPR + lzuCsTLs8wHeEX+SXz8sPkg/9+DR+PKi/ulB2DsWPIJDSq/auZyUJzokiH3vX2Igc3SlBvIeDdUD + E3OHSaGUO7OBpG86X+1ib1qiLINIGUFM3Fbe3FW++z18jsxYtK5CYIC3VwDPQReSsDCEiZW7jwhb + R+x9mLomVgTwGtDrqF7oM1gGziun9Pa29LCI5YSvaT0MYYIuK4/J0YzfsXgEJxvpEnySIG7EeAmX + PIEHnR3pg7yjdvnj768/kEgRr654MdXX/k//S0MowFaYXYmktTd8UYd3oFwSsMDSHyH1h/QcS7/z + h5fhaVDvU+ympTY+8j7xpIoaroQnUT60Dto+yUZvZO7bWUyMCu3No07//IOJ36BBO3SNiFMJdUH/ + +uuPYUoCJ9q1Izr/XoSVKZOo+2oTdzpVAEOWer6wXQLM57g8wF//WJRWtWI2BFoOfzJvoVZHWvd7 + uRxW6HPLIUfVObqS3byDP3zScigHzGg9GQiHnv3XL9+shL1R/rNaNNcg5mKszxB24ihR42QfMR/u + 5xEO7udMD88ct+u79g6QfSryp9+crvMj/S8AAAD//6RbSbOyvBL+QS5kTrJkFpmCgAg7QERBRaYA + +fVfcd7t3d2lVZYHku5n6j7SN9Gu2LNVHfDfz7VCSs7EOINh6tAHsmJoNdHk7/UExjSxVOAX4D4z + Rw1HbG4vEBazXGCzHUew7PcBJbe4zIdXc9DGy1UPUEp8EWvYMYb1wzQSDAVLIfbtZxXrySc6UNIe + YENbJkDvP8eCh7MtYG8e7xHlXkoFvhoLMT7FTsOvSKj+6mvvR7vY6HpYYMb33QyU6eBs9CVLiPXB + QAxeF4rlygELXrF9xcbp9qCrctk6FLBkJGfBwoAH0RzAsT2XRFl4peD4sgjAVQKKD3C/RnPJMQFM + XjPGmgECQL+fewX+9IKsPo1ibS0tgGJ1d3HQiy3duN/XhFOfrcQzrmbBvFAgoevlEGC16a7781AL + yPnMEfP17YZFbYkEL8OtJq4s8pQI08UEX42H2L1NF7occjuGOx+T84WNI7L3L7Sb+key+6QNXHMn + sSQFi06K9JlGS5cyCTxUtoM1z3wOW3qXILz4CsQn6WwX25qsITin4ElOdnmhO//1cNNLOIO+o/Rn + cloL63Q8zaDhVm0790wNi9/wI9rOP1ydDj40c+NI9Ps9bH4a+x4Rf6qEeXmaGViz4zP4p698o3+A + kSkXFSo5F5NT2kzNfD0VI3yxbOT3y1mgK07THBb2Rybu+3IraKyXCXSFW+tv8dgP0yING+Tml4nt + S6sXc2iyDEg0L8T5U3sPS3i0OWCOUUq0e6hE5H7abEj1NscVHzLOcMKXEUp+25MH02zOXz1CprOj + +dCYTLQZVHfhM+ldH6zIpfwVDwfx732NQ7ZGJOveoZTMt3KGaWAVNPYUAS794uDEbMSBOo93CbvP + GmNDHDc6q+V4AIKpVMTAxges9gQkmNk3fWZwwtGlkgKIgrfhYBO9HWfz3KaE2hKt2AuY00AYp3eh + 9j3WxI90CCjtMxuGzy9H7IIuw1IY9gKnTxnh8n3hi7Gpngeg36o7Pi/o6KyeXriwtbbB395T1myQ + LVuYimNDghhTjXr3XwhfObGImwuvZkvWMITEGHL/yDShwzd8PkOGcOp+3m1Bb83TQq+D0+CzzrwL + Kj9WE5lzuRL/8PUc2lfGAuxQsogLwM+h0MwreDq6C3Ye57DZ9C/8QFNxfOJqz82Z+NiRoRj0H3+j + p2SYzKlSxQj6iOx87myI53rJkbIHLvlVBpzRVz1Un5MzHxPEOjN5NjLa9Ss5KYRG7x0/YAONDtvO + mjnkSzQVISorJDn9TG3ajl8JZV/1hLXHr4rG8Heu/vQO8cVbGq3yw6zgN09P2AHec6CHDZiQsYYE + +/huA0Y59ync+W+GmlQ7dHsoNXwkRUDOx0guyMyfTGk6bA3BDQPokiykA1NfrP62eE305+/EeZw7 + 7PKrTNcbc+nhv+cfiRfRR3XJ0QLSkYTm2wdDlF9HSPp+moHUD8Wa1GUlHdqLNsND0wzTszuk8DjO + 6zxN05uu7LePYezyFTEvmhYx8klOILBvybx6bl1sSG5DFHOMgeWtf2jLAUQJHFUpx84rbqItaOzx + r79n7qWHxfgwuh5MUZfNR3m8FFR1sw0IiBt8/rVldLjl4APN/HQkyuf5oNtfnlBTJiSJ0aoO803C + ADk6HYjhjwKY9EebwAfIG+z+isewFcPY//kngsnoaH96BP5ikmH7ovjO5k1MCe/908TBH19lmHkh + M7Gf/qzU/LBccdJBbk0UbOodol0QvnxEcP0i98FwCuYdCS4cB97BRrQZBUO24gXWdVBnHi0w+k3j + poPG5j5ElUo8rL8xdmHTrCl26uzU8Pmac9Di+wfxW/fprMxbDpDx0bTdj45gfsEqkJhrnO39d4tG + UYhymNxngK3dL09Dygho97/zRxDGZnk7ao/EVezmw1NZok38viRo2IIw81PYDOP12M2gsPIvtgbw + AUtrdR84HwQFh4mfFkteuBbMdN/DRmpyGrn/NBvKLKvMlDWo1qnxJqOP9/2QszB8tLWfuRA2TCFi + 1RePxZZHio/8QrzPbfBx6NL79geG7/iE45wVtH/5yp8f0Y4822wtv6jocLaEmXcOj6EnKVuB/feI + cYmjiGqmH8IRPkZsYsQNvbikJRSw9CEKb3rDailRgFD2A/hfPyBhDYBX29t8kO46/U2jpPO9ffOw + MsaOtijnVwrnc+Dhy1LR4vfntz6nkpAyas4Dx12cDh5LPSCP7E2G1QVTC6RMj3Fk7RsRWX2V4SjU + PgkF6UkX4AYdfC0J8ueBhnR6ZnINB3WB+FKHdFjQpVwguKkZPrWqD3iWXAPw8Mpqzt+Lrm2j4yxQ + HT6Rz/SxWKx/fiaznME/Zm88DKs354CR0XdeYyQ10x9/yW+3xoVJ+2aVl+cH8XybYDmUP83GWZUJ + 0FnJ/XHHR6JxQQKjJ4l9eg+fxZrnMwerVxzP3K4Hfm/luIAf8dS5TQZC6WGjOpIb7UT+9Mui9CCF + f/rFqyOLbtos6+hPT+vrbysmJa99pME4xEUsfSKa6psrvh/D0+/bnm+Gn2aFEkMYFVfSvQX0Fprq + n17EOEo0jex6C16ZzcbloWma8TobLfxcRX4+Hq9xs91/XQkPy8HE592fLfLJSkApMwnO9vyC10wz + QC8yFnObcXLzz/+Hau1h7fbgnF/s3gJoBzOZawxe0are1QCoevZX/ybll0iUwIJ7mah6Hzn8X56I + FqbDKheeouFZgwqcF93H+MXqESs//AqeTkUzH33Tj+jxoM7o3TvOjAZjiIa/5xkLt8f5rje6ZN9h + hCo1/vnB5QdJDt+n22UWD7fS4Y3cqeB+39gT3ZPDP1W9Bn946f1s0qysmcWw/eh4XiuPRIvxGkZg + 1r2DrXiuKNW/twCWfgcJ5pVWW6PBdKHqhwR7Ms7ATzYbGYEfE+E93wGU9d0QMlnEE+u7bzwHsFtQ + prsevpPtDLjCRTOUbCzMY2DWTqu8DAalZX7EZ/z50q1HuQpvtPRxBY5hw7bBwsCIPCUsB0ZPl5Kc + Y5h8fhMxnolN+4LJXShg4UMevjkXnf1uelg+ort///PvB63pkcukJt7zv6jrfbVFYWTPxC5rHDGn + 5Z5DKH1CYv7h27Fs8398pr3L3zDs+gsF75ODTSnPC8K0Xg9rMemI66ais4xawCA4EG5mb3BqJqYM + RzS9jCfBZ/hxNi5bfQSlNsR//NDjfUNuWZ8tdnC/FuvtxdWQbq8HVvlLBdbHp4yhlEgA2683LZap + fXJoTceAPNT1Gq2LlqXwPAURyY+dCOal9XS4OID49K2BZl3aPofhIRSISk/csP3sSQJ7nkzs6FFr + 9NB7NRhntcXmzfpGS6LaHNz9yMy+60jbQhNxoGCcyP8w1yRajOxg/8sr3MCUHT54TSns785pz2vP + lGmP2wFt19OduC4gBek6/gX2vA/b19kd6D4eRSRM7uQsLb1GeWmU4KUUUmztfpwz7yoDNZiEPrvn + gfyf/jos0CTJnpezIK4lUB6WHKfnjYLdPyTST9cFEsvoVRDwPufweis/OEtSu1kfhpNIf/xmfA/7 + BsuhCMWCi8e5Kd0YbFQaY4kPin7mz/oY0RPORgGXMiCBcO+LblxRD+qZOeBTFAkDua/LiID7Ij6K + 5nzPO7cYPmpBmg+ghNFsxvcQfM6dhU/Wk9PGn1O/AKzN3wz2/GK8XN0AelJX+JsTTtFkX9MKYjF5 + +OJwkhvOjK8haK5xMHP2RR7Y7idZkLF+CXETR3emcm5lZHR+6SPVZofFpfcE3pxBJs6nchoOi7oO + 79eG+uKulze1bCHMytedePoWaEu+GjWc2zSdj8FVor8g/kC4oI9B5PqkRqt2siQYm+OX4OCa07/8 + Ap6f3vaHJ3THhwTiyo3x1R8FuqjxpsKtgm+sLaAHy3NTwr/6Idhq62YTUBtIp9j8YD2L7GLLf+oH + 7XkA9sFxa94bkWS453k4P4+stgrJZII9/yLu7kdYBjUzkIjn4Pt98KPtnnoJHKMaEBmDV/FPz74v + DMCK59YR9e7PEHGvx5s4xatuxl+R7BOukmK/rCRnuuirDvb5kI/I9gOzKPspfBez5n86jnWWQVBs + VMKvQrxEHIZVvdsB+Jtn2GI2DRMWnzliP3qKT9m+IVzZ4gv+1ae/5wHELKcXTI9sjl2We2ur9GRG + uOc3/nL9CsP4gFSGwbFlyP3zYrXfnp/CTnz7/loOSsMYGWeDI7qFZJ/30PUWWR24ldHTF27uAt4B + 1Kq/eRJWindMCdMaHdJc/uoX8W3Rxtqyxz/9P/MR54Pdn7TS1Yi8mex6dwmFY8J/2i0iVvKTmj99 + B9J4CnB0e3Dav/cvfr8flq9foSFUhT3kOO5G7KdmDIwfWBZ0f1KJleiiF2zVbCGqDcnzxQTUgAJt + SWElGiPGtdQ1K1OGM8oLQ8PZk3fA+IrCBO153T8+2axVVsHffEkeD2eHyQvdgg+lQDOyVZ2OVCUx + cNxHjj3VZhv619+B0i/YUQMm2tTomEJ99liC59gYaDVMrkRODCX3eUTR0lRPCHDlx7NgOgpY/vxx + +tNVn40f7e5PhwoUwsvAHtnOlKot68Pwd3nPxx2v/92PhfkLUbI6ACv0r6k0nW3ir+7P1sgx1m24 + z198qWMedJX6PoBCngfYNc9uQ/NwEMCffjLM1Io2Hqg+ODf7huvffdvvpoN5uNB5+wQEUFkoa8hw + 38K/lMNz9wNVB6Hbe/Oe9ztE1VofbmZ3/Tf/ooxRzVB/i++5+/pdscryo5YCdhqx+YPPYT26mQ8V + Ss258Y3MWeJrb0vH0LPI3zxtnP2uhOIKOuLs+TXxYOrCfX5F9Olha7QLSh069xMgp08tU3ZpXznc + /Qoxfj0oVqbMR2h45oX4zc8e6PP3kyHpHjE2NO/WbKiTl7/z9NGuDxb/lH9gzmisx/NPQtc6yA/w + ZT0JxoetKra//OWP/+xZj5tF6WkKc9UO5+NhO0SLI9YMiiXa+JX8lLXl7MkuMoom8znhcWiGbBYW + IAHBx1qn685o5bEMubY7+HHG1Q1lk5X5Nx9Idv4eTF3V0T6f3POsT/R1e6RDwdQqokf1Q9vmh3wQ + hdTliZXflOZvHgKHFFDsnR9zQ010eIF//u8Oq2IZstD/01czF9STxvVX9gXfHOP7myAplKFOKUjm + +V7Ni/o0IiZ2mRYG8yHCJzx+hnW7OLooZWZMZJyOzrr7CXgz6jO+PalazNd58WGjvpKZ71ze2fXn + DNjiM/tInRy6nWNqoQTID2zMyxfQfT4K4KM84WwVm2hmCjaFz5Jqc73nzys0vpwkB9jzB5P2A/3y + ywcelfaLH3fC/81zNyiiw3mWGuvX/P0etPIEYJNUuPibZ8L6Kack2PqjQ9sV+0DUUINPF59p/vge + Vt5Nnuu55sB+/hwyLyuZt3ZWnTUMrBnu8y68520De7mX9d98bBaq5xl0p+Wegv59T2fuLtnOqgt4 + gf/HRgH3vzcK1GP5JB7lkbZq12yGlzT/+GKXb3Tz8xoCtdkqYp6vCuD4pzvCvEV0RtpqRKzhWj5C + FH7J43b+AvrW0AytXz+Qc/JDdL26uQCO4nYmpyQVNXqeziVASXgk2uX9BotCnwESRqYj2s+/Ofxd + aSxA6P2CzYr/Rj1p3BbakhfjxPs4w1Ytdw4qU3ciYWPKA+cb7xTuf3+u08UG/GT8Oug0X5nIuUmb + KV3XFgrn5U5y+JWK1bXlBMHufcOy8ZucdVZKH5b9eiXR/fJr1sPo9lBhVYyVdWa11Q/fPmD0SvSX + 53NqNudRq4h79s28WOwNLOf1mSMo1jF55LoF+Ew95AD/kIMzNtcBixO+g5NuZcS8TTeNiqVjwxkP + IfFzI20W/se0UAtfE97fX2O8Iv+AsHpBYrG2TZe54VwURqCeEb7/nPU7iRzMq7rD4VRx2mhY+47w + OznNxyubNZtUej2YdDubt7m5OrxypQmSjdojt7/zF0vHAte3Gfos37qUf6diAKpYYMjJ44xh2ygZ + 4SqGFHujpABGzz49ej+YBedf8QfW/X1QhwDvi2zeggmKCoTAGBPiOP4BUP/xeMG7x9bkPP5eDj8Z + zw7R4L6SLPk2BeuHk4/evC8Qg0aaM/1wu0jevBrY0VY7YqGdLdB00tw/xv5XI9fZ2tDZ6TC+eZzR + zPVrFcBfvZpX5d381RM8XFbRF8ef6vAjahIUuk1KFCXznYFzzQ6+8zdPchv2xWLcwxJwZfYjXuy7 + zrJdIhc6y20gZ8bkhpHlVgm6JRywumQ3p5v9UoYsPYB5yW+oGBLu3cGDTEJyRlAolvf94qMuH1Xs + xp4ScbEbpEh3hJTE8ufQjBKOYuR0qCHmvRGd6ZQGEnKjwMXJJ3wWDJEXE/2u0ZG4Y38CnKp/THhT + 24WUzAkWW7/ebZhZaYhTezAaarOfBFSZu5CKHUk0FWq8QWvfzi7Zom8o/SghUvDhQaxchHQ4NOUG + lS7zSSKRL6Xt1SjhzbtxxBzEiM7FYxv/6gMHhLci3lN4E6RZr81H4SM3nP7ya+CaPwe7iRgCShpV + Rx3kA6zntyjiQRdxCKdC69/fy9rQiA8DdF+5n8/ESw34z12b0Q0cCmL2jePw5FvWkC91FiffznaY + 6DoLMFCDgJyzlx4x0Dxz4NLnNg69qwNYwZlkmK2gw84xUAZGOTsvNFgwwtmOL9sY/iqQKBzCBhdU + A2/9rj0wJfuH3UfPDpu0Kj669iXrX7NLOzBDVOjwg24+Uc+66hBy2RLIGv6I1cas/50v6vhyws5v + 4YZFIWwJ9dunxDKvRQ7PZg4ErrSi+XiJPhrfHhsf1ZBtiXaYVYePgzqBLGYscsuFzdn0l/mCW9AB + gh8/GUxW2STIm6kxX3KjAV3fvhjwjE3V51O3bhaMjAqGVQ2JUhTjMFqq6CNMbjO2U88sZqFdXkAs + h8pnp3lwmKjvK/jMGxdrgAUN5SWOQ02u+eT8e6xg5Y5xCFcMTjj9ij7lWe25oS3owfzNIsmZlldd + odFXVazb+RusgxkFiMexTTTTHkD7Y6UPbB86moF10RpaPZwWnsTjg+jPzStWfhViKJy3Ozb6Nqfb + jVNqSK70N7NPKjgE3lwd3lpymLfvrS/Go6IuKKqumy/y7NisB8UK0XhtGP/dpIE2jfVJQveV+ZGg + dnTKijjggGqa3IyEz7shTyKbyJjGjZhmLtOl6PoPbLMxJ49DqUZLudQMvKaPC9aW9Fss/NSl8FGH + Oj6Haawt0VUOUG3Nm4+G668Z+alOQafd3+Qx3B7OYvJYB4/H1cSazt0HltMsG1KahiRgbs+GdW5j + AMnzMc1bKDVgcVW9RZsmBTgVd0Mv3JYRvYeamY/vxR2W4fZ+oUoFgc9M1eoQHYQqWtGHEN0NToAV + ccqA39W6znBIxGJ2xbhDA87u2PczBmxi1C0w7/2RnNYjAd9fsVSwfp6f/pqpx2I9afcY2m6u7v19 + AnwAjwdw5z0X49v5BLiBlSEylNNGrA2fAXnxLUSxC284RWAaPq+sy+Gdxy62SGxR+g38FEZrcSDq + 92YX7ORnNVyfwvivPpjow/Xoj39vX6HVpmKUXtD6dQN+gKcSsWdWjFHpy72fus/TwFJr4uCTtzxy + Q5IKZjV7dqLPctme8DkD20e8C98aXvzVEE8ae5qojnTmeiLJju9LbSQlJN/Vw/qSbc7mKyFE388s + YfXCPyLu2j9KeIelga9iPoPFSosUlGAOiTUdamfto6MLTXk7YBP1FmC6CcYAvmyPVIf7/j/y7pNB + B3kKcUW5qdiUr/cB/edACE7nWuMOS/z5wxty1ur3sOPtAuXbUBAnTImzaTQJkSDUIwmu7Jn+45/9 + foj1FX0wNzypYAu3t0+D2qTzUHspxDf0xSeZo8NsSUSCObxKOAtPfTEFVCwhQapJZIvl6aRlrQXG + M79gJTkdolV4YheYx8adD2HXOqu1eS+QV6+O+PPr4Wzv+BdDdJgmkia3c8SUI6si2Xh55JR/rWKp + 81BG2el+JRV5N82aHK81nAKgE4znCSwO2kzkLNdh1ydCM4ckqRFIFuQfecWKliftZLgIjewzn9CM + Zvq6HRC+uzWWzeMVbMfmEYBVxzpRvFhxuCoHPcwzyyCP+BFQ7p48XSiHoYCd92MEcwyICy9p+iGB + n2O6lCMrozDF5R+fFowkriasN7Mi2ngWHGrEnQ7X/NViY0xfxSInMoNUeY4wXsxTw9mTA+EfnzlA + 2aL13uIe2j9rwvcw7R2iEiWG8zZVBPup4bDYUEL0zJ8uSeFbB+xNU3LwfcTbfDSR41B2u0CYLaxC + jNvoFrwxlyPU2UDH+mhZxVIW7wDOYxj5oo1yh8bHtIZLgKF/3gJmmPZ6ht4vfxFLe16L9cioAtRK + 7PpIOc3FWrK3HOVepZDTVzoMa83oNnzyEY/xJTK17d07L2C12UaMJNGGiV+FBIoBKvznzUXR7Ces + DXe95x+Z21rM/a2rQf9ov8SG3K+hXmHoUL1sEvGHJC4Ij6VKSlTQYqvpj8366rcOAGF5+uU1rIv1 + jGcV3ofG8jnBCMD+WYbNewrJnZ0e2gy7KoEP9VOSXQ/Q7TK4pvT9jBLROuTt/Xgr4fiyWJLteowx + TksAP2P49X+nlUQLn7YMZJNLhGV/yIalXDoGplmn4QevWMXaZL0AT0wLsULVpti8Thcg/fLCvBC+ + izYPySXMQHUn9iteAfmVD/Ovn/zxGsrRoh/FA9QXzcGOOPyGmSulEH4ziHH5rvmBviy4wfPpjojJ + iB9t0Xouh/E7M3Gai+W+8edxQL64F3KT3z86x0HPwEf8kYmH7z9tLthYgNkqdliPHKvhVzq58PcE + MtGontI/vIfTk1PmVhnuDe1StwP6ATbYHcE6rP632wAVNpbYCFgNZWLxJUlFoJFHI/XDTCrJ/sOX + Ge56kr0GUwz47yZiPQvjggYX0AKnOzZYD56XgfqL1cOBH1/E4HzdWcR6MEHEThIx7aEftujeWZAp + WBl7Rc4Nw8GTWli6UMPBpa0dLubLBfzp1fvXlCLaBFRGObxJRD2QlVKJ5TeoiRJDdvzVtojaCWy3 + 85lYquMO3O10SeDMwxtRi8Ec1ihLXbj7C2wz+4aEUNcfdHZ6PC+GlNDBX6wOPIcvnaWtnofFUlcf + cnRdibu6oFifXu7DPvd1EmdKP6xc2sew5VOVKEf6cSgvpT70701MzrqaFSvNUAvUZqlwya1RwSh5 + FKLuTA3sMebY/NOr+doyJDCPV7pNY6XDHgUYB8+n1yzGaQnRPXYUcjaS00AlNrNR4QU8PiOYRguT + ZRXY+50E9/AQrS2vJfBtlT98XiSHzjGQWzieZxN7X1MqFqJUI/jzO9jTOUCfuV6Lw+E84rPq8tH6 + 6U0Zvq3qNwtmyUXTvfmpgOqdhqOzMYLlfHJ6INvS8qdnm2302R7qs/bBtq6cm7V9Pm0kTAATRxzO + DQ0utIVQfMXYzDvB+ccnV9CUvnBV6mG7XDgdzmffwvY1PlJSoqCTuk9uEk/4GA0v/KzqH55aTWc4 + ozwMKkT08MX+zb41NHouJtKbsN71Pi369r3qyHeryb9c3j2YPWRV8HChIrGyJna2p5zn0I0Kjsie + fnKWQ5i/4KYfeKJKWKDdAQoh3M+DXEDLAYIMGMPW3zcYffAb/vAHRCanENn4eQ43NqKP2mte+cvO + h0tWZhIUg2OBnfN3/Ie/YLOvOb4b329BP6fnC/35T01/R2ArW8+CwyhXuKw0zZmGRVThX71fXHkD + a/hhLBh5kYP942vV5j/8khJ4IjsegOlohDPY+cjnTGcdlsVQX3DXiz6TKX2zii1Q4f3tdXv9FtFi + 47MAk/VHCRY+9bAW5tcHrtRH+Izdc7TeZm+D7nzAfjyIKVjf4Lqgw5SE+NSzj4apfCVB3/F0x+dP + u9AdHznYxo/nfEgrc9+Au0GJd84C3vlQG0XHkiD9sgJWQJuAdalgB/MsIvj0iVyH9YTl9cenxO2K + tNmmBAigfj/uRFWPC11Cwczh+pRGbPCtC5hZEDdIcXHGpg+ZojvAJURhy93n44t1mlXTzxB83scT + Np401Whuvnzk3KOHvzA3ZRiYI5Igpc99I7NXKXOLox5y3Wny6WF+aZvrpyEM1EwivuP1Gv1wlxGh + 2+3jo4zqGmdOSwueWfnG2D6zlF4ykYGR8rXwH/4uqneHcPenWNlhfueD6h/+ZcLVAFwyO4FE9V4j + /pl/FhsaVevf9wPzyNIxF6UYviyXwarzONONMHEMy2ye/eNwbZsVJxcZNXE/YDk3o2G5PzIdytL0 + wdren/Sei5b02LLQp4yUU/KieQ0P0wPOyz2sim3vBxBI8wH/+etP93YryL/yZGbHM3XW+P7qoWsO + Dtbd4AsWoFYqfA7Vmci8RjVyc54hCO4wwtGt+A7rhJQZ3uowxYpnrGDVdOWA5DAQsCerJBofyccH + Az+/iK0rv4ZOhcYgsfxVWJPjXtse50aSJEx+RO+bRduI8FlQ7B5u83G4PbTtW0sqbPnm3/1TdoV3 + E4iM8plfop2BJfocOiBiieKzcH0DqsgXG+bed/DF7NUWS+ipMkzcMsPXAnHaekubFkYmo2Df6EJt + tNtWh0fmGWCzs7SCuVIcA9cRdPxQUupsA2tBlCgMIicN13Rl+ESG8bswscb5P42ksWlLJXfP5mWx + wmg7d+ILfq9dhItBb+nWfm4z9Mz4NnOh2f4HAAD//6RdS7OyMBL9QS4uL0my5P0mQUHUHSgqICKv + APn1U/jNcnazvGUVdUm6T59zOnR6HuIaw9yWkwk+d7XPvrl0A5u/iOFBP+jCPf1iuOH9xC00A7N+ + UNeff0aDal9vfF3hUBGWA7Fhw/XLCegWNOTuQ9Sp9HJ6HW8evBb8jern3O75X33jDc8kZvYJ8xlX + Fw62kI+IdoS7fDocdgYcmmLBoIw1f5scFsE0KK5U78p3PIY4m6HTjhMxx/srp6t4VUDaxT3RcM9V + I7vy9Y8fElWjAmBBbAnw5rQ2/emV4ecP/vRD3KIxp2bSWtCfsTotYfLSB3m/GOiblRJVjNiIF0FX + POTzTb3NwLEr8Z4tDogvf4Ro0j7yxxs3z+BcNwlmgKn+tB2uhtrXV6nxaak/j7wJYWTVAQlcsYkZ + 2FcNcurIIec3DnW+UN0nVAl8UM3Gii78VY8j7NfHl+LX81iNLr+k8FSsCw1FdfynZ6T5VPDTLB7L + eG5HmMIu6sk/P3HzY1tYPRNK1A2vWfIXPZFykxPqmFbY/+M7Wz0jm/+iL0G+m371lQSqU/z0vAUi + axQI7lMunx/9d4VWxbtE+8V3R5wOju4TUj+VWn+tnKhF7m0QaOGHnr8EvDHA4fTisCSZPtviN4Xc + 1rK5VXsDiPUpLEDoT/K0c6pY5+3LRQKb30FczkqreTpOBfj5QbGAXZ3+/PEN36hffT2w+ZcFRPJ1 + IV7aTv5kvtcErmP+ps6t4PXl9EIW2PgMNZX3BMZ6lSSopA030UcP8vEqpzdY9OcdNUvtE2/5o8HH + 42zhaRc4Oc9HtfWP7zgbv+VP71aC13DmqbPxj2V42pJcFO+YKFF0AFs9uMBPrgmUTC+aj3lerqjL + AoPE5r7U2eZvwOB4IjSK/2wgTi5nAH7ss2nTt4BVzSmARJYdjGJU6uOVZAZs3cXEzXkUfRqgugHC + Lg+JbnCVz/ZWuwP2FOdESy3cr3j2HYi/5EIxlzbVcrkJGjwXESPk8X2yxfXjGpYZvlC9EF79vLB1 + QOn8CSaGh0f8z/8cpM8ffq0s0Wfjb9nBj6se8bwrRn0UEuEJH8/IoEnsO/0CcR1ASpICV135zv/1 + A/j9qaD4kJwY96Qclrd8IsGXWmAJ9w8P/OI7UgxeX0vq1T9+T/3LXYmHB1xKuMXnpod5tlQ0NGTn + 3NsYvp/nasOvJyTFwSa+arN4MpOugeuQ7jAaIIzXg9KkPz9i4jc/Y8VqtoOe/zBpkISvfL0P/gXe + VvdODsil/jB9Ou3nZxNlm8cxLMXlCSM+yUkWvMpqCXJhgOqNehOVQ+b/W09Ldr7kYWOOsSFdss3H + Vsnmx+vTba8M4OsWh6nKUqdnmMstmArSQKOPZVbcj28pB3wg9u3e5eOraTFsV8Ojyol3wRzmWQ03 + /52ErnTw+eHxOqL25Ye//eyH00tsYCFIF5o9rhGY7ij20OaPbvXOrgaM79HeVm4evcYTZIslzxlK + 671DyOuhMVZ3qgYzJ3SI7+OCjfk5Sn7+E31sfHvNtGzz5xJ92vyxeH6axQ3I4/VA43vJ4knavT3Q + SXo6CbxH2fw3EA9Mpx3DkL7baiqpVqOosBGe+dGL36Zo/OvP4Pl+3OXzdeEMeOvYacO3p87yfS2A + e+O9SRCsn3z4G4gDnDpficvqEoyVoUzoer3tyeXhDDpNhXf3y0dK9tDNuenTKfATXTPi/736fNmT + iwDXIdn98s1n5V1KoXvJZAwvRa+z21GW4cycO8krOeknPPsezKFuEes8ivpwcNMARIv2oJ7bqL0Q + VHIBNn5CA+025GtzAfOPf+M2Wg/xErnb2edU2PDSqqolYsttmzFSUNXcf/zlhzf6FMz0kolaPg/W + aYZ/yqWYWJa2VRc7QwP4D46IuptsfwkrbgdDO9sR76fv6rsogM1/IsHmL1Td05PB5h9Rxz2L+S// + ocDYMn2u5TcuzfeaoqczrEQ51IrO73euApyWTpMQ+3q1zJLXydfESnEXvLR+2pmCAMv++If5Xvrq + mx/2BOnydGlS7yR/Dj2Ng7fFEoh5ELyqe58IhJG/o5PwJTcw5ufdTk75P0o05bT/9UcKGFBzT7ym + Un0Rz0qLBnewiDoGas9v+hRs/YBJ5F3FFwT7YoDwb9Kok56/+bD1F8DPf3P5rGazyr5HQIK4wjO/ + S3SW114Lf/3EckcPjKU2X8DCidmm172eu3v3J2SD5lFXeOhgeh46B/3w0QFKUAmAjztwBjDHlkAw + mFn3wohFiU0yzmzYwjfWEfJtJBJTeWM2L8Wf928/tvWPl7UPU/mJzRNx9jlXsaNTGsg+C4dJrFEV + L6f3U0LD6eJNsPYmsF6/QQA3v31auoPSC7511KDgRTnR5KL1mTuqN5TCmBB/FyJ/Ft6dA3160Kmy + +XkDloACt/4ejZBT5//iH/l9Se1xsvzpnn4DsKS4JWaaVtWAsGPBWnwNeLkfcD9veIo2f3paSqGp + +C6OGrDpXyxtfvUcnxoJoDT6o4b9+ovXnVu28Mff8V259tv7yKiNjIaevHzu1yQQFXi/vmpCaIPj + Vf2YNcA1uG0qj/pTXOAV/vx550kLxmyfD376hLjxoOX0m0sF+PVH8V3ZVzN6CAVknwlTrJ7bmJUs + K2E3nI9kq6/5HNHXE+RCWxGy+Vv8Vs/h7Z2teOvP6syJwQ2e7oNHDkpCfOGuji3Ye2VEtatK+wEM + YJXHD67Jr5/HzORpAbQ/e1s/OuybHUc1IJetT/OG5/v1rr47SI5vg9hBec43/I2QcMu/mLMfBltx + yjvQ+3rjJKzKoxrl8SH9PycKxP99ouDyWHIaBP0Ys/0p8+CBuSbxTumosyJWVxgPnUy9wGn0lZOE + JzjQ8ESsYU+rFXGljMR9J1GzRqnPfOXbwTBSavp4BaCap/RigG8qn6bZBS1b3kCI4MQlFfUre9HH + N9hFkKXlGaNEWP0pUx4WyJy0Jo6i2owpeXMD6j6RyS09DD0rMnrZDzI30sNBLPNVkXMOfnfbN8R8 + xPJlQu0AYjON8AI1TV/jy+yAEYctTQ3T8Oeu/oNAF8eAHLv4AWbZqiZk0vFJC9FzmHB26wat5mFP + VHBc4ik19yuMGIQ0MsYX+60PwKLhU7epv4Bdg2eKVGU7Y/mNRMDkt2tBv9UIiVFt+sJ1QEdYRbZD + DRnosZAJjYcUha0TPGqveHadOkPOZRcTq5dZtbp6D0FVvlJqqCyK1xB+W+Se3IxaeqHk2+TDJ3hP + jytRz0myTSlWjygfz8IkGKahr0/iCiByhhJXra/3fHa1MtReuj11hyVgTI/2F+hCw6dW3zsxB6td + AteOulTBZRBP8XE10P0ay8QaDzbgMk7ZoWWRduSOWNfP/g07cD2nmAa8ePCFPf+xQFVZHiWy3VSz + eL/J0HD6/bQXzoHPj+MyI9ZfGM0L55Rzo5fv0HSHb3Kpm9Zn+YcO8BQPJXk8dkk8ofXRArTb1fQX + fzNs1AZt8UP8YpsNo7zPKRiOSknJXy2xdZZyB7pd3dHLFNi5+NeXJdr7nE41RYrAInl9C37v54bV + s5/8p1DDj62VNMjFHswOUDqg89GOpp8SxHPUSRK8p+KAQRJ3YOZtKkG9Tg6TXMAgH21nKaFldUdi + EynKubPT7FBEjQfRHqrGuC4nBrT7vqZ20Ko6f5PqCF1dL6dkdom/JA5nocr7isQO2pcv+nJVov54 + eNPjo21ybhbvJXy/bjd6etp2vj3Pgg9V1InqOKkuviGT5cbMfHqvVzX/tz5lgAg9Iemlr4u5k+Eh + 5zLiTLocs/moeGjyn2eanaZ3vIQlHGAZKCXNdXeb6hTpN/Qi0YecpsTTOVmTJJgqxZ3qCUlifuyb + 514lSCZ6m2E2W6dhgL/43Tl/bjynqpqifeVNVIkvuS7sjIOEgkDuSZiFls/dP04Hywc8EAUvYs88 + brxByivhJN6OaS4misPJx11eEPVvz4HJSrsEovGukxM3S/k65coTSfw2Rbp8jj3TxGuJpLD5EFdo + xHjlaNjB4Q0PUytdvZwz/v5aGJa3lSjK6dsvCfyT4f0qTMS4ZXt/ya44k9kcRyTI22c/f8/bN/GX + /DGtZqvGfNrgCX4GsZ/mKuniafQPNRqI+sDgDI45l+52CniCRJq4FDUVj6dZgMuovahx2HVg3gtw + glHuzdQm0pyvC1YwkiJFwGifr3mL6FqgmvPHiRtSJ15fTXqBqXK7E6I6Sr/cnTgCpRR+p92AGp9m + maGhPV9DEpdvQ19WLsYogWyegL6vATezboK7dKvoH/FVDQ41ShR88IVanzLPZ4HsMayf60qug+fG + PBDfGNLbW54SzXf6GY4fDHOa+dM0IMv/vR98fRyL2rlBfHYK6xLpnlJT10eqvp6Ggwy237EktXy1 + mhbL4CfL78R7VIRNiOtkWHGopoTGT33ZJ0kJuQT/UVXoR7CG8NX+3ofeAvlSsVj2JrT3kpI6xfWt + U0t2PPjD63gVIFv8vwqCa1Cl07v4K/I5PsslCqLDh274zWhjDQLsTlVHtVoY+2+0za52xhgQF7yZ + z5zwBRHqNX5a+gfV6bG+DzAMmxNeu9vAqBO3GspdBdN79LHANF34COZc1BIf2WnO1ba6A30yMnq4 + yZw+P4w4hUzTORJGih4v7eeSoPo5r+SSBFosbPgJ9lxL6GPbz1XqBQz/lAbip1Q82fThVQ/tNCmh + lyKS/OU78fAfnv6ex/8Ns4PeVAmJceARWMljvaBBFkbcxwkGS+A/PfBU//Z4GYQqXw1ZDGCKIaD3 + vS7r9HCbLZS7GsZ8PXVgKY6BA+3RNWl+kaeYftl0gUVafwl57JJ8fmVKim7m6pLffq2y/U3gnusI + /vs7jT07xWUN70ZeTABbNhAUORZQxlsPqgLd0vmL5MigrohC1T7SmQjWekZRhnWCwyYAArlyKbrU + vURVwfDymeTuDjbL+0jO72+Xj0AcAygJGaPY+fvGrHEkAyW7w4EWjnCLl7uTH8Ffuttvt7B883U8 + igEqq1NIruG3YMIDkRpy50Alx+hmAzHmjBZu+EM1Hz/74e+qRuhzKfVtqvSzX06XXoP2NUxpFO+V + WBRV6YKc714hufQ8AtHsDgFaZrjNMOmrWNRtpURfFB1pdLXtnlOJH0EFy2fqkZeWi/PR8dApdWaa + pMjq55K1HrotSkWd6jP47FHOE3rt9SfxDcrALLjAAc89fyZuq8KcCauaoWZ9/W0zY29gGkBZo7NY + GkRTI1cXs7vAwe0UBbF/6/Xmog6eUm/GfUklsKjdYwWnj32n2kMtwXTUfQca3oRpAjxcsd27LpA9 + mSv1jTOoFrWdA7TVB3p47SlbXb2C0BkkbULfqwH4Sc0baN3uObUE0dbZIas55Jh/B+KjB67m6Sy2 + 8MefdMO4grmQzh765cutP54Ae+m7QN74E26cwe7n7qNj+NrXCTU2vBNCoa+hQkSD4k+zAOZ+3Bmu + lyOgIRf68ap/nwM6v++UePuh9OeTeMmkX/wY72X02UsOj3KsSRrR9o9dT6sqMWS9Wu40ZPzKZvHR + OlCVD59JUq6K3iW3bgenh/iHwWi2bNHgFcIyfOr0erzSH58q0Ybn9D59ejAE9CND7RBcyS++mSUr + Dlqv641qfBTnbNKuASyFhBBbjU/5mJdiBNfrfNvqxyuem+d3Bg/URuTgHCp/MU4LB0375lJ7xu/+ + 3/OOZ4sjFjjdq42vTuhxmlfqFNHFFzzytFApke8EkC3E799+3S/ncGJuPoJWrI47hC46m+oNv8dV + PkSgTYOeWudnyOg4FztoOplAgxup4xWt5xZW3F9NvQo4+uINGoa5VSvUdXKRLeO4rNBXvIX4W/xv + 6+fBxyeKtxOfJlhhJytQf50ILT7Nwqgw6Cma7vlCTVYe/F5yU0UeGmlH8fHtVNxJjC7w9DHv9Dyb + Uj+XY2dA3d/vqP6rX/TgleBxCnfUdHe8z/74/AJTrOFpucRzNQ/GXYKcui74FCWmP4PMvEFw7V/U + 3NlBvNWrAGQuZ1I1HwbGYlEs4JtqIXHQxfIFA3kFxKLlk9DUKWAvGJXIMdGBJMvJ8LlFvw4gFCWN + +CMQK8b+3AnuYB8Qw/aMeHQubQF/+f8PL4qLF0B7ahuq7AzFF8ApVODGd4m5s4d4acL5ifxDeyQa + zdtqLualhOf6pVF/BOdqjrHbgYswhZi/LmI+L6HW/OPL9o49e9Y4swW6TgupYa2DP91Sw4KvfZNQ + IqcLW/g8neFWfwkuL952UO9xgcK7iYk/Jom+/PL1+00qeir2T33+C00P+oqz0PzzcPM1qsYAHszb + nXiBY+mrYkUKMvhdiyX0OeRi++VLaM+OQ+9mx7Zb2JAENz5LT1Hy1mcPlwU62GeRGlBBFcuScAdu + mbn803uiZqAIoD9woNZFLLd4KBTwp1ZfEixR2C/Hb+HIsXbG0z6slGrdozkDD9RtE9uuXs98uXrK + 2EsionljnvfG5aHB0PdseqafF1j1BSew0D8fap7KwWen8z0DZ+Evwu037PuZ15EBrwKvk0sh5r1I + A81CQ/RSiS/ILF4V+6+Gy3y903tmXP2xM3oNcPrxSPT7TqkEaVILGKPPh6par1cU5aBB6znBRONV + R18j+k0hdaIjvUjXLl4HM3+CMKxPJKD7fT5mD6OAh3kSiduvM1h8veDgD4+UeP/M2VIVLYDZ2Sf6 + uJSAPW5rB0p8m/HwuvcxjSdcA1WOP0RtDo4/fw9pAO3F4kmwqGW8CrhqQDMFKXVvO69iAN47cCnD + jga2GYKtHmKQPT9wOmE0s8WjvAZDUdbwEtoHMCXlVwPed2UkuI0yG3j7I0Nr1Ah1Ro7qc9kXAnS7 + psPin+aDxYJ8CpeV76dRCV/6csKyB/2THhI9IVw+iso6o00fEUPTQD/a1nSEn52sbHjxBmNiukdo + FKZLDUeJ49FtDA9u+UY3fpS3GZgaeFtQRGOUX/SJo2EL4lkXqHFbl3z2jqcIKp+hImGnOLkwkiBD + E3+IiOMKx00vD5kc75ycXDa8ZOk9vsG+IQnxpcOnohyeJ5Q2XEZ143b01zMpW/QKF4tc+T8GPp4w + T6jdGS7NdmUEhrJPBeCeBkzUXrTzBaD+CH768brMbT/75Tbl9DSvRPOENl/4e6fBfeQO00iiYzUf + fSGDq8mNBF/81l+bdxbAH/8rzoz2K8frA/zp102PshlWQgpuy180iVQI/RnR9YZcpEhEMRMzFo2s + PKLzo30T8ldfWOcIDoa//bdPY+7/+CQQDl1NwomwalnnZ4byhcuodtrL/trdggFu9YIEWz1eBU9a + ocl7y8QjSdUFHFxaUAVDQItzEGz7uzpQfV99Yiqu14t47DEA9moT87qc40XeOnxS8Uq2b8Acxqrz + AwJhvwdEe+M3m8VraaGZfUuiGz2KZ304yeh4qU+ThIRbJfwd+Aa+p/sVy4/EqP7h6QvcJxp6ml+J + Rfa5oNdhEIheBZ+emlOggbbzMmJY3z+dHtJVg646P4huhBYT7m+gQCmsP+R2dP6q5TuhHTzXlUbV + Sxz13GN3x3tkvt94q5dsuNZHRTrv4ZEkcuXrPEBVBCsxXzH/3W51yuCAQRVMAQ3U6cq2z0YgfN/L + Af9xf3c2tK5Uww3/fno6X72zOf3q5SR2baOzlR5ntPvrMNHxNwHicEMYFvr7M/HKjP3ly0FN3vQo + xZeMbbdcXSGU892eKOPerJbsFDdo0wfUcga7Wo799yLvReNOQtbYvZhen+2P3xBvJ3/1UrvXDYp7 + /zqJh4Kyefv/ZJljJl4JX/ur7CcOeh6fIykuvqMzopoK9E9qSJNudphILUNBVtaoVKljK+aZ4WFI + mGARZ/vqlZv3zxn+9nPzL/zV2YPtWqJnSozeUX3uqJklcAZZm+ASjf2Gd4L8869M7hXq7Lb/SiCU + BEq1nbBW/UsXgp8+J274LcD8TiUJEv/KiPd2Sc75N8tBr3f7Jt7pLeUDbykJCn3HJvF47frRpNcW + BnH0pbrYhPH6cPUZ8on0nWYluFWbfzTJT1wj6pbB0ovOyW7gvKwOln3X83l09j0YyO+YONKiVdw3 + ft0AtGBEoxIe/ZXcqQeHw30i4bed9Vm7QAf27+OH6O1q++J8HbebST+faS9QIxekBR6BEq/cdDtq + r1x8Dt8G+vFrIKq97/q19uUMtH+KMIE1pmzLzwv68dEbNBL/9/c2U9AjSvy33SLVOA5YzQ3n5CLp + 1xQ63D5BSU08mq46/dyuHFSUZaX4g6/9agZLgfZ+TGhYsXe+xukgwdFfMnJ73fuchZWcwpNGVMyb + 9AtGa689//l71vHYsJl+9wIU3nVMLuJqVeLfHgfQPX9agoXsEncDKBsQPdWQmDEM+nnRDwOCtowo + 8eup74a/+wXYV5JSXQoXfchA0yAvOVWkKD+Cvu7B6EDbFHLiIHLVeeU5yf/4ZxiRafsyrZLg/XIK + CT5yDKym7jRQJY1KQrOL4zVYpRVW93ZPHzdb7YXa5BIErKih5FBa//Vj6tf4oNrpmPaDl7Ql3BfR + Ac+4HHLaOLMBEqXpMKry0F93xkGGmz4jigtaMOd2uaIfH0qX79Pf+KYMw8PTpPEnP4Ifvsm9cvHp + 44minHFpmsDi78JNEJiSPv4Ns4ei7A0nH0q+z3m4K6Cn7Fq8Kt93tTzFpwEjphBy2vhxQx5yBs8i + QJP4p/Vs/oZVho6GveJqJft8WOzgKbfD5idMF6VfW0NyIKdHRxqL51Kfjq6K4cHe9Cwn5foSBfoT + 8on8xXvhPPjLPbAM+GbbF0LlXc3HmM8iwCHVo142vvwVvlQHbv4v0VfhxkTJ/0YyvlOfmrtHXY2v + P3OAwuc+EvVo/fI5FaC23cppq7LEGNNgA9+5VRB/0//M9osBvDBlFJcfwZ9bXX6C/X28Uye+LdUv + f+H1bnnEOu69nKKjNsDFsmeqV7rcz+QiDGDzh0i41zOftU+jhX/tLSOHY2GyRcJpAceoF6l7vvT6 + 8pUmAW7+GzEnMFazrM0ypHdnokezfeVzIXcZaHvTJm6o5Tnr0PkIP2P2wMvFeMcLVFPv53eT4Lbd + UsbJhgP3r29Io53x1Md6DTOw8b3p/XD6nLofdYWCXFDqvS9ytfUrjiLrbhbxL33lL24TeOAsFQs1 + YzhUM+LnJ7pNTUL1seXB3N2zBJ73uyOxn/srE9QYTDCoVw+Llv4Gwz///LreiO4GZcXqMdhB9gXH + 6R1VUcyVY2nBMvgjxCGhlYtyy2S4+f2bHlF64ecXurvySjxutGLGN+YRrpcIUF+fRbAEbiLBGL0/ + 5Dq0f4wWoT7ATvliGmbGVafb+shhqL9I8IdNfXk8KAf8d6NOzLitPmWO3EFhX2rU3PTE4u/fMlpW + saeW6xX6Itf6TSa3fUJM7vgCm59t/PwXqs2Hcz8+XsxAtXfhiFNjuvkT5xX+pUpAjz5WKmGLZ7RL + OZ2eUzPO129wXyF+DphcG1j6Y6PCCzTtwiU//cqOz1JG1nKW8E7IpJiCXStD7mvnWFo7xBgREwNm + bnyglj32/YBJyEG0gzUxuVmKV0f8rD89T4tzfo6Z8uka0B+FK3EvONPnV+Yk4PrKGYb+cMj7SXpG + P70/sXEp2Wg8ugFcQlklzjX75hueSHAg+mMCW/6vinVRgNPxI8HCOlfbeklw2G+3wnZJnc+Cyzz0 + TU8v4pmroA9/5JL98mkC99cCVhILEaz48kL0VcH+OuVOCSs/6ElmedN2wiVuYENd/adn2dLYcgEf + CWeS2DD24Bs50QQ3PU4sYcr0dcEOBt9OMKi+/U6boJSgp4ITdUxF75fNrwDj3nCxRnhDF27HwxGu + xjkgus5P8Wpa4AJBhk+b//T1mRRb/973h98968xn++vXbP52FK9HXfdAn1AWDlt/7ls6uw7IVG6n + fZjF1WAd9hIE0wsRN13s/uutzhFGTz2k1t0Kc3E+rTsIbl1FrMvDZf18fa/w7bQRcWLOZXx5/DzB + bBf5hL5tpK9/Z9bCrR82rY+2ielbaWaYYgUTZ3xX7Lc/AJ1ta0poc+5XZ88U6J7fLd5dfMfnxGtp + oD9ZbqfTxoenv0QR0Kmtd+Sw+cNzbj6eMr1hleBEy/1N/yU/fUWUyOkBo4STIdIHf6r0Rvf5X779 + /NHHxtfnjHMgRLp7pqpwK6u+iNUZ7k5JSVw7jPoJW8cash2bpr/D9RqP34nfQR6SmPr6Uc3FG0oG + iDrLmd6ZS6r1dtMcpIs0mMT3E8UbvkU/v4NaZZLkjAvGAmx+FjEn/67TMHOLn/9DlJVy8fBQniV6 + KFJKHvPrE8/i4+nA2+BZuK+tUWfK+5wgVbqE0/6CZf8b01sAzdWlU8fQqV8rKaphVrjjNBL1k7Ox + XVK06pcHcTzsVcuuGySIXXYjP31K/5ZnjVTlSYkZVVG+7LNjjWyTy+kVniowrLHSAUnaZqah+u1P + 4HyJYILSmuAxWGIq8BD/+CPRnqTz2eYH/AcAAP//pF3Ltqqwsv0gG/JO0uT9liCgYg9QERCRV4B8 + /R2sfZrntk7TsddGCZVZc84KVXCvn83i6P5i5u9+LupJmZek1LUleSwNrN/Nb152P3e9JIuP9nyO + s08qDeNWiBV0QZgQ/Dow8fh4qw0aPuEXnyPzVG8ZN7sg4OrbXJkmT//qB3/+Djl9FH9Y3vXqI/5D + nEC4b0pNZ5ibwA+ZHKufx1njtNW8wtcFH2ZGoUu84LetQvx42lhGowX4ablC0G6uiTGGJmB3vxa2 + pbUG4pE5xP/q37te/tMzw9rMUQsKc+SIjXUxX6+tOf7xyb//T/lZPfvweXhaM7fXK4fe6Hqofmyb + 5Jqo0w024AD+PtvdQc7Z/OOW0MUv85+/NKHkKIFBzjxids5RG/NTx/wvJwqE/36iQBi2/QzzeNRo + qi4d5MqoC5hwONBNIKkt2D9ZIDdaDGB+PdceJRF7wErWQG00mzVCC2sa+5mtdlhSMgUQRttElONw + opzs3X3JpkseHONvpvEH7u3Cdy8kxBDDtV4swYfwe9X54HDXtZx5MJcInupIJeqx5bwxtr0roKH9 + wfmhu3mbPGsqHMRCJomyELoc7CUEDgMLbOpXTePe2+kKp0pC2Gj3riuIbXvYLCFLzi84xSONb08k + nlL/xE3RllOdV120LY1FTlhQNd77uBG8ws0NuOSnxbO33K+Qsy4V0ff1WOqVVmD/HGzFzFLyVaoZ + ma9bvDts55w7nR86OHGjih+vRM3Z2xy2KJyEI7lqr4qulnLc4Pdq8kTzsy+l09GWYA1OZ3x9DX09 + q05tQm4+VcSF1bcepVaS0akOVfLgbIFOnLLqiPpjg09yoWrMgt8SWpdVIedUar01p3YPPZ+yxC+m + 98DYGcchvL4lrL/WvYuPfAsg6J5jAH5HW+MUAGSY+dIJ277z8TamL3WEFsOfWS5wPb5Wn1fohZjD + GH1Gb8szpUFzNWoEz84AthcFT3j42ATrddLRrXDrAsoeHuYjf80G5u/5npcXJAqLJsB5sRcgCw4b + to6XOKckSjoke6eBBL17HwgwfAbOQ1pi3IZ6zHFezaEhamWst9cw57Z1leD1d3Xnpa3yfMW1epV4 + zBtE6fkiZ/fPiLHzMvhompPTGys3UOvsK/FWAQ8LFcQdAV8C8QOdz9eKcxsoxLpPiiFSasZ/tYJU + f+kZO9ffKd8ufZygrwPvJMQnWvfVw6ugZlaUaB+NA6vxs6/o4+cRdkJwihmoRjqaev9Fzh+niNf7 + rezQ7Wt6AZKOAtjSuM7QjWwdtpmAybnnWelRPHYWeXS66/HB6rsQXNSRGBr3yinIgh76bs1h2VH9 + mD9wPxdM132u+/N2G2g7zFd4rKRlv38Y7/tDhpfv6zizMQ3pUidTBsxrz5GAKGK9bXjveug8s1nk + R1ZbwKvzkdpdQvw8y2LcMkJoIph4HVEZ7VFzF+tdov37sH93bcAPl0VADnMosDzc83jdXhKHxCY8 + kaKDPN3SZS0RZnudqE/14XGCHqlIPyIyk7rxalJl8gJfA4ywfEiTnO2GooRhur+zoq+axlW2MiLi + NBqOuf6Qk1kfezj5Kpyp/i7zsY+8HrHO5431g+/ETABMCZqo5HAod/HAKslmI3X6Tlghc5GvzuO1 + wTrv6AyUuzWw2qlLYCUp4bSJ+UyX41CZ6KoWbnCkske3JTD2ObK6jYPbeKbU0nQdvXU1CdChWgdS + q88EtLV+nxfu5dOlarwS/KbpSgxI3XxNn9IT8pbbEf3tvIcFFnwL8GyI2OtcMpBUGiNwSssXsYpD + E4/lR+zgxTm0RN7jaxqKRYIFifC8khnmi2o/N1g9hA3fEWiHNfluGeIS2w2QP6Q585y8J7TjN8WO + oRqUd9olg/oms8Tp4aleu6Go4LvjMY5X+szHQqNPhK2TiOUxBANlG2aElnrZiOEaC1iXd3eQmm9z + I+mBJvHaTmqPQHT8BuJlO8VUPcQHuAjnx8wXBz0nJiqviDWdNzZw9BqofmdUoBwUTApzfGu9KQ82 + /NGgmJdSMWJ+DdsSxDdPJcapc4aNHt8SaD51hG3YTdpydtYnov7czECdAsr7of2ESpGPM4m/mbck + PevC1a4MkrjoWW9ro0jozt9toqCyHbb6ss5/eI6vcI7qLY2H9G//YyN6/epRKHMGiOOhx46omwNd + PZVDsvoOyAU/VI1tmOIJ2ar64tP39ANMtJkHhBP1TWyGGPWk52kCLxhrWG0Lli66okOkj2+CX46Y + aSs7wFmSFNoSLfoQbyzO9RM68VXEXlWQmAZ14kJxSQfymPGaE31mGSh1v5bIBQPi9fWGKZSHIiZ+ + HfD1JBceBNHIKtiaPx7dbrVTwUVWK3JazynYpgaVcKGKTK4Z0w/bWzr3iPvUBsayJ3lNm3sN1JL5 + SrynKdecFeQ2XAr3iO0DTfLv2w83dNfTHhdtrgwca8o2orpqEZk8jJx1Po2OQIS+Aft5VgOn3JUA + qas2Y8tjakqcdknhdC5V/Fo9Gi+ebgpofM8Q21T47K8qPlq4x/9ffvXYYd1GFPUUYn95LcMmTZWM + /p6H3LefmDKvu4weZXEmuRl+61m8TTO8BpcjOZ18Z2DIuMjoYCIfvww10/ilSyFqlojFbngP8jX5 + SimUnu2JPFGvgU1TzyEiZaHj13Mahu2znmcIT+cvyU43E9BPq3Zo5AMFm/a4guX3PRVQz12GeFpQ + anwmcC443OUzecyLOnAvSgtELuM7uH/vCWDaiI8Q7QWLJAf/l2+ydw9gxDcNeQnz6G23WqnQNWYN + fCq9qqaemhZw//udQb7zVdgiCQH/GuD4Gzy1+SpoCXLeZREg++LUrHpcQnRxYDszpvkcJnzJrnA+ + cCXZ4w+s53gM4HTdXsGosG9vjfmwggw5lyTZFkcjX6WfEcqjkKisUXr/8P9Y8gPxQtainBXELshc + 3p5RcRCG8bNPEdHE95l4+GANf/sF/fEDDM8Nna6PI4QhTQ1spKcfoDB3XBR2XYXD1cqHfT8c4KN8 + noPX+97E5Koc9rm5DUPUekHeFAZ2iWwpp8SPAJ/Ty9yncJtcEwfI2eoFn7UQza9QxUnj/bxFDWsd + Jn1qB0x5OHor31QS7NhnHhxMo883gYQ2fNNmI1krs3SBaqZD4axZAUXZW1tq7IwgFQdrFi78Rxu/ + HW2h/hFu5NkPLliHZGzRv+elvmvAWq6uwyVbdKJsoI1/nnBcINMrGilq0oG1YZJCxE54IMVbW/Lt + b/18vwWBVEQsoOeqLMEXQh7vvwdMe36Ev4sqBgMj8d4yv2wZeqb9wMm2/LQRvLoAhZegmpvCOnpL + UP4yOA9ZiZ2n33rLpJUyms6VSlQxDwA75e4CfAEqQRML3bAsPc7A6yeeiKujWVtTZ+kQWiyfaIJK + KImQKYNo5JUdH81hUUolgFAmHdGNuKupV7AQIIH+Zu6Yed5mHyAUsYVFcvqJ6sBfFc4FNxr8iKfc + v/XSHIcKzklZkhM113wKtrKAjuAxxD9/Q4111C1EBQkxuSa/OqcH7TmLY5/fCY6/kjbOA1vCIK/Q + zD50BzDD55VAYfl8sGvZl3zpoBXBPT8RM/nVMXXxCULXTos9H690OxSPDebXZSSYygP46pV8RSvZ + 41FcK7BkJTZh8JY1fP3c+/2dIikCl+/jiK/q4TxwafcJoCqblwDeNj7erpd7BXMtGfGDftWaqMIS + gZ0vEBnWOVgk3Q7gar5k4oXJdRhvRy+BGsYh1rYE18uP3Bf4K6450bDI5atw8ReoJiCcx0Dn4zEl + kw8M4RFjl7EDwFtl5EOuDDvi8+8boMNyzeDmlTW5KZ9tWO9jv79zmwGy/z3l+14aIWWO5bwxTUDH + iyyWMJ3K30z3fL4U7RSK0f2ukKs9ngFvwA8HRV37ENW4z95a8sII+bSuA+Hm+flv7EsBWJd1w4pn + 52Dp5lcI37y84ZO8hZRXjA8DecOqicoEpbY90viJuIL5kuztXCgzB98U+H4DgncqmR4LzJ8LH1Q4 + E3WIL2D9PpoS7vpg1w+9ttKpZ6SAlzSi3Qrs/d4oyeDOJ0jgO2K+oA1IQL6gAFsBN8SUt38mVOzj + SPDINsOyXWUdFnaKSZJJi7eeD/0u0e9f4twQM0w3aDfS8XHv/+HPP73yx7fpxsqU2/k7HMNEIcnh + eIz3fFkBYpUHrMAjG6+SIJrwlZYROdevVluvD/6A/KCX53qyZEBxoSYi8xA+5F5VDOillyRL8/dQ + zlXQQkCk488EqqxfcPJ23vXsLecrFNnBnhekT97yeOUFuIvP0/x+5YU3S5/HDN4wxn/xnS96zS3S + eXVN7EuzQnnWmltAfvYbJ/e2o58eyxJ6GWZEXHoOapo9ww6uC1XmxjSf9SBc7zoUfqTB6pmagLMP + 8ADShvOD885/5zmV5b/4wr4z28Mg8cUGUvFnkWCaopg5uR8T7nwNnw7VeZiyT9pDgQcWscxiq+n3 + KTAge55trB/LcNjCGw3RI2D2OdBg8lZ7H3O0NOwDy5XvD3TXw7Dfbum82O9cW5yFuvDS2c4/ftJZ + 2y+CviY9iWbzxBvhFProI2wpOR21CswomBb4Lt0Nn9y9K6qGcI/2eA7ojmfU0LcAAly3+A9Ppj8+ + 79dPiDWuf8a0SJ4BPK5mTHT1/Pb+6b2d3+HLHz8XT9sBrdNNJqfoLuQLny0u/Ms3O18elm8HWtj6 + fkr8uLAHNqdyj/7w2rNz4q0xQyRY5z3F3naXKad/PBMEzwP6p6/GVfYWuG6cSRxD/dBF/sYyFJbv + BwfUqIc5uhouTLTSnvk3q+eseNogCgpRwMbFU2OmmNcWwV8/YTM9bt7EtGdZ2uORaNYLDbNuMQcY + zYKJsZbc6Vpctgoe+WmYmd2/mQWHHf/46zx/YnEgod8EsG7TawDNUfGWrLR0aF3oRrDGeh7949dA + ZL/Y7MMebKsSmvBEHB7buKP1egnxAbbMKwloviGwdNzQQmhrLvay+kuXprpkCD4qjG2LjNrS3S9P + 2D2bjPiNwdOVU0QTtq8+xZhTaTzp7nGWvuV+ovHEJDUbbF2BXmkVzWIPp5pmAmeDHW/nsG+NP/+k + gbu/FPzTDx8tZKCpOGtArdejJnJFDyjfe3gF2n0dFmr+oHQiHk/86/GdL3NAUun0EzMceJGoLUX7 + CSXOSb+BGOCk/ghbJoDyUAczA9w3Xe7ZkMIonBt8ouY53/0Z4S8/Buzw+wHqSb4OK0kLg/qGyngx + mzVE+vFIgvXZtR5lRK6CwzikRC6VT740xTrDXe/gQGzvYBletIN88HSxsl9+afh9/jxyS4ynLqv/ + +Cq05/iCPTJ13ng33i36vNgIB4XZA2qmJwl6wJfxo/SqgWTCwYatLnEz4n699kuVJkF8+q7ncj2A + eg0qzECUhyFJ5amhywuzCYLbxSPPR7Zpy0sCT3iRwHFu4/OZssLjbEK5ssZACobZW99Gt0j3k2Fh + LWXkgXVUKfrzM7CRpatGz+SqwofThLjwRExXI/2OMOWwFdQrfcaNrvgQkJU7Yne/3tLNt0hqjz6H + C7T84ia8zhDu/g45wVsf94yQ6nDPL9iQbwvdtt/UwIVqMrbydu+Zen+nf/4c1sD7GW/2gTnAwWIe + RL6ZucddijqBY70p2K+E7i+/RbCfqgKrlfnVyP58gFr4G8b8NavpfjIQzopZYq3PbW/c+RkKKBWx + Ol9Q/PnLby9BXwIJcTNYnG7mAPYSjTyDV+3xjtNH0j98+MY4H1OdmDDNzB/WJfuVj4vUyfDeqqd/ + z3/b/UTAErMlp1soxDRYdXd/Y8uaD6ivAbmHZQ+Pw+eEvXTG8fQ2ygWtglkT7IiZtwZ94qKHVXvE + L/wl/30lqYNveMY4PVAmp8zNH4F8kGKCgyXTlh85L4ityi9RKxF53XaVzX98/Wq+3ZgZqZVAIPJf + jN1syMf1cReg7765GY3A9dZnkY7AdoqMXFxjoRP4RQfogUAOBLE1asblrhw63z8Aq3U2xxsXLza6 + /N76P37EXYohAcJZscjtnVBtXnPUgE0MG6Jcg1O+kHFR0XPtIuKlM4nHxXq00GkmhpgI17SxLrGN + 3odrEsC724FFUl4NjC8GwobGHfM1ehcBvA9Rv59wXfPJ8LYAirfXJ5De/ddbh0IQ/vIzMYnVg1EE + lPvTh1j1YQv+9BJKK2qRU/Ne8zmAsQ+uF8ElmfVr69GuPRfcvdnHge/c41WUXQj3yVPkZaiSt319 + cYN2xl5IsAVaPTaFOMLANhH29/jeft19kf78xfPuz5H+1LTQ4M0Um0F2pt12Aou082Vc5JyS08A0 + VcBs05vc/uJ95xNI949ncgqJmq+/azSD3W/Gdu8a+Z9ehvqmskTHij0sqQ9SsPO5QOIYK9/xqITw + ++mwXh5eGquSVw9aHiREU5JrTntbliD/STyc1IkN1sCqrpCVgU70a6PEzL5/gPy7BzjhYuU/fFoF + NJxJKVBv5mLBhuRrFMEaXD5gegTliC6gcbDJv7WaIcUtlR4PNSTOn3+FCc+J6qrMM6LmGq80fcvo + HrUnrJwYZqDt2ebgdvspRHl2rbYw/qmHpxenYFWQHjXtq3afybUoJBTfTE4u97KB+/XxibP3qUkH + 1YW1HdDgqa+19qe3JdYOWuzdcVlTx+lDIOttTHC3SvtUqKyBuz4nVtIMw7TrcdgIx4iodw9p6w54 + UOTH07xuwMx5T02ff3yBXJzmWC+JMNnQFKI7cbrw4v3WyOXADNecuFFgUka8u+o//wrJWwi2Lvst + cPzYEDvTgdW2mwEySDtWxadLq8ZcAONAaiIjInqeKznvPBsTNvJ4I45wageiJekVrNXniY2nWOV0 + 19d/enGG16OSz7cM+f/qH6aQ1mC637oO/tVvLOfo5VQI/AXab3LFf/7S8llOGQjPWzVz5NCDObj/ + MpixjhN85mGNqek+JHiTfY9kUdBS2h9hKfz5s7f13AyTHuuy5H1G8C+eyF4/Qv7av2Zi8ltN0sfT + BbvfQ5ziNtXT9yMwQPkFSSDlOTvM9WnKYAPUG/b5Nw/6zL83cAgrQPSEdJRUmb3B5Hqwsc1JIxiT + 01cFf3pA03rZ+727qIHt4xMTNWgLQNW8stF9CHsckMqli38+SCBB3oG46vkS/6t/sHnf4p2vDaT0 + nqFo6SmPlbhiBlKIQofKyY6xhuvvQG637Qlg60/k5DGCN5VZ3wDKqx8SdGVd1+R7LSDLwu+8Xiao + DX7/OQCV6hTfikfszQoAKhTS5EdO0jGlO3+OkH8InkSVq9mjJ7lP4EWJnAC8Mgsw7u99RefVNnHI + sfkwTo42g/vL9bEvy3a97f48PJbsgB3+auZ/+QNtS2sFa3NtY/pLuBQk3U3b/Q/F2/FVhoY7RSQr + ogulT3LYexjcZOwgYNb8b9NLsPv3ePfrNSKtq45KlYmx7PIN2NzsIUMVcZe9j4cU917s+cDUfI4E + mqSC8UfOm/Snf4rdf6LZVb/Cp3U94z0e4u3zO2Vgr+fh0y1M8/G1fW14tcbvLMqeAfhHZfp/enNe + Zsh42yPNCwhsIcC4W7Oa8rXZQChPXcCgixOvyg9AeDL0167PAdjK5pGAaDmr2D+CpV5Wnk/hn/8a + yJPobUNb98CrBw77q/wGm6QOAnQ28Ub840ehmxxCCYzvEeKgH3pKSdYdYPBoVJy08rumB+5nA/99 + 5/HpXAk53cDY/+lbbAT5cZjkmO/gFS5uwNVBHa92tPeovluEWPE0/Of3Dnn0CJYsNjz69sMFbTx+ + EWvM9JrxDSeD2Zs4xK8EG6wuKFPo+StLbFUYd/9KGVGO+jGASkIo3f0A4FhBTk6lpw5j4dxcYFv3 + C/ZsKMf0MCcc3HzwxD6j/uLVpYkJJ9S//vQlWGi96EjbTiSg+luOucG5lOhXJDkprsd33Ct3x4fX + ufkSDWVvj54uow29/phifOeGYTxNVwmI96rH/r3k663uQQmextnHD9VMtZXEAoTv23PvIfhO6uUd + WD5aGv4xU82Z//JLArnIMna/afA2c3iHcHQ5gFVofevVE/gNSkVtY0VbPsP651fMDx/NzOH4irnt + N7XghW4nYhw6lTKc+yrFvd5ILPZL410fQShMXoO1ySppbzzWCBEZFzOM7He8ek5awA9HYqxW4sNb + hs/tCgNbR8GiUglM/uhGUBeTG5bB1Qeb1xkybBL39k/PjyKX2uBl6BGWWe9Tz5luhEDts8u8GfdZ + o0na6rDG5YUY/aUaaEu3Hg7z+AiQiVJvephFBtNz5s/Mno9pZ5kpFK7nhjh5ecr3+ukTbTpWyGmo + hWE52EIIb8W1w+rviWI6XAQBdOR4IQo8XuKZixcX7fVSkipaXCdSu6nolE88PqnfX777MRX4qx/7 + nogB5SRRhfVWMcTrXFyvPyX9n3oUiP/9RMHUOhYJmJVo6wodGTZGq8/82jDxeNLPEjrWuUNcRmHo + gDe/gnkzvuf+NhjaYoxqi5qYx0QP6wks5cSa0GKsmhhhc6BzwDML4L1cnkffYeqlJeUGJVSoM/s6 + +XQbq0MIheR5CtZlNgATvIUN3G+lhc+SzXgDlV4FmJ3kjLPlUXnUcjQVwsOXn6lsGTn7i1wIU2Hr + CV6sTSPbg3Whsgw1lmtPoRu0fwlsTaSS6GJ/6wWeggpql9LDdr688sXktx6B2IQE49+D8mVVm8iY + tRmrYuXkPGvpLizRtyOn3iJ0mU1aAu6yWcTPNnVYFx9cARvzJbEt1sq5IgQ+5H8gwcZJ9j2GpoUg + ibwxk0ddxZSN3iIDnDyXiFomx3i8iVoBfhdGwrfHmwerdv/Mf+tBvHn8DaPnFDrSxm4hUZmuw8J+ + lBI9P7qBlYl9a1vfsSrabuKPyK1uDhwllydkfvVCrOhSegz7USoE7YkNZPa4d+FvoAml1kywXvM/ + uuTuKEB1a1z8UjEZaH7eEcxNV/z3/QvDLVd0OohnjJNSBSx3qgpET+sQHBcr8ngZlQnc42Nm16es + MQwLNhiT7k7OwtUFzBeEDPocvzHBpKgBG9GuRyU3+wQ/cO2Rp251aDwvK1HfL58yYBltWB+CS4AU + VdYYFGgmWmUmx7ktH701yKwUQqp9Zv4iRmCMgyMUnoJqzoe2K3OurbALzW0IiHv8iQMtplGGKult + cgPPyps+/rmDxqGUsXlhyDBp988IF5wa5FH8SEw9LEgwo4qGdfU6guV1IAn01+FNNEYY4/X9c1L4 + Fy+eb2sDc24/FRqokhD9euG1TcH2CI8zysmju7T19jj6CUzfnRF8PyeiLaoKenDXWxNfhvIIWEbZ + MrTvBxLGNM65KDxXqNfbG3kdltJbwcvj4MeIG6zq097lMc8K8fq1a1LoDxSP3LK2aPy5L4J5mau3 + 6FD14DXtc2o7sObb02WeEPfCiwSm6Ncb4zMpmsNZngV0hvWad1GIaowJPoWRq/HzKbwiIW82cknG + OmYuedPAZ9mnAbg8X8P2beQQHp1rhRXrvcbbFNstMh9yQpLKkGuOPJcOnej1uE/elT3GyPJFqllr + w2aomjmdHoEMbwS2+PROHjnbzNsIxYdmEeXVlRp9H40Uyon4xHfiHMBSm3cOvh45h7ViRTX9MPoT + ATnVsd/4Z42tzPKKsOj1GJf0AjiNe5ZwC3QPp7mjDWxweLdoSBaHGJV5HpjQMTto8omOsfww4/l2 + vG9QgScWa6lJ6qWPVhlowNGD4R04Gr1USoqwdXgS5TJH+Va8pwI4+vkSbAu50mUhhgRP/DISu6w3 + b1XLt4r8Qv6S15E6HvOJHAHa3pxhUwrO+cA1gwxoCbmAUa8+WCX1XSIwS7cZNqEWk+Y6tfAIVBn7 + 9y+7d+2TegjLZMKFSfa5p0o6Q63Iv0Q2gtvA+K0EwY7XQf5KTzEbjNIGJVV0iTyNXr3pbcfBWJYQ + fqkeirfRSisYb5dolpZHpY1JKJpQiTxMHNBcc+Zp9DNIn6NGNMUNYjrODIcenPXC2F/u+SolaQLq + qtewO5WDRmdplKRaE1psbQMeuFOePaHWYY1oBpLp0hZsCuknq8nf89q/v4TFCiPsTZ8g34h0zZDP + GcHuCF+GwS1rF8FkyWY+tH41MyhqALOkJP/2+0LN1kR/6x3JqqqxVWWa8HkPf+SxEgasHVs2aOBn + 99/vXapuE9B7bAIckmdC93hioNW9Q3wqp/ewPKuTCvPzUgb7q0PDdkjuGRSVJ4vVqu08PnLGGf7F + z1Wri5jaCdWhoftn/NrjZ9Ehw6B77x/xQzUcjZK7nElwkI7EwqmjzUiZVXA6jcaOXy9tC69hBHPj + qBOZN0VKoiLUofw3FcZRWm8L3NJHezwS455rNX0U4gIXnBkkMOajRkMEG9go7g2787fNqfAqGRTz + qk3SDjs5Szk1QswhaPEJJiElC9uP8DmgN5HZVKXcLfR1+DHODXaS0wNsbJ9H0MoiA6uvScvZHQ/g + kzZ3ogQq8Ja8iEd0ogZHlNa4DXwZMhFSIGZnkZFr0H+kNYI3sKj4PO2KQhnWEgn6ScfPJ/x5jOeG + PUrfvUHcPX+vn6ddgFvHKPhmPEFOj/ekQx1vPObDYr615cdGOhK+nkG8r97U46C4ATSNrt3zheZt + MAttNCSbg1/+IsZrcPg1qFJPNr4NWj8sS1S30P6sCnl9RylfdLSmsP69X/j1XfyB9TexFPb7J5fy + yOWkyFkf6ZCpsI0uAWDv81nez4wa5FG3rcYdg6yD5Ra+yJm9U0CPcAuh79HnzL11cdjuniIhn7MC + crorbs5ePJlDSVc8SFgGw14hZgI0AvY2g6Uy4u381hb0PpYYn19d6dGeHlWYdu83fm2Xrqbf4iqg + iDXeRAuffU19PhjhdY5tosNPUvfQGjfYdVwUVAYz5Muc9lf4OTIFTt3Z11ZkZwyMe1STouOP2nSj + cYBUf7TIRTQ/2uK3G0Rj8/oRm/Kztmnctdq75rr7fi3BEm+PFt7SC5mXwhXo+hFbHy2TcMdZmhk1 + tTKRAYadsoG4uba39g+qorv0tPFJvN0BZWDawhNtbeKV1kQJBYUKS9ED88o173iNx12S3OJyhju+ + rYbg2HBsHj+CG77T5kQsiz++ifHdW8BCmiyFR6me5/1+vO02tql0LicPB4evO6z+tpbIu4CCqFfu + sNMeQYfr6zoHx/bz0uivt3UIT14VvM+aHi9qUj1hKTMuSSnNwZIcTj1qo+VJwkGqPL43qwLwJoqJ + eXbe9XDO4xKJxu9OXi/FzxnrK2fQclVIrJ/w8PrGGxrYgS2euR3f1578GOnFyjM+B2quzR9JjCC5 + lxqO3q+97z04CwDahMU6zyUeeZ8FDjmD4ONo1G3AnF58D1vI6/hUt6bHMp9Uhlpx3yuwxM75C10C + WNHhMZdGwNd9W1kuPH3res78HmrroUE23PGUeAOJ6rGvhwiaiS1hlftE+dZLNQOdQfKJEh4jbZuV + +wZxL72IjR9+viLrGkpLlVhErmQmpxd8leE2Ni+y/35KR3mY4fP7ZognHeFAHmKiw2M4n4m2Ajam + OJgFeBEcRIrbPtXEEBQbrpX2nWtZo952F2MZktP4JWf2YXrcbZxTyJJur/JWikbz0HChdqk8Yugn + sZ4dwfUhfl19YlqDTCcbshK8eNmTmHam5KODhgMcFMbGSl6CgZrDaMIGFCa+Vl5CuzsmJlSRGOOH + eBMBfc+xDaTn+Am+ndtqvbmlEhyv/hOnurDU1M/GDA6SIxI/LUWw8GAyoSfqLbGFNhjm6NB3cK57 + BSva46jNxpokMCwShuSdJMebvL4X5AKOJb7R9XTZ8Qtkg/QgqpDP3nZ9SwK8NWlCfKyK3gbt91Wa + UMnOyyCpGtGWNoL85/UNxP4ZUvasZxBe1DMIKNezw5oVlQ9f5vlMnLjCNXUD5gruEXsm+fKMvTXv + sgjwN78gMWzqgcJ4SOAsF/ye75R6O5CjDU7U4vDJTQtvK2nAAHqiQ1AmFaUr+7pH4A+/sttgeMuR + j1SE3K4g13vB1LQ9Dyl0wXcMaB+HOcvfsxJ+vzaDg1vSePPThQW8vjYmOID5Q/d8FsCrZDbEGkUz + 3k6jP4PpPWXEObynfCWPqYWfcYjmRok1yubfx/zHx8gdksZbrg67IGI2KgnJkwH0ONoqSEK+/cuf + YO0q9G898DP2XwOjBfWMUPOQgvJZKDVFq+qjg/XTyVWee21lSAAh9jMPO81o1fPTOUp/+EvSWYn2 + 9eFdZHzdiHji+TNsF+31hCwobuQm3Vrw736U5Vdjf6m5fBTZ+SpyDqmwW6r9QKeHKUOjehTYVZ5N + TnAwS7CX3X5GMNU83iiFBq49UIid+CNdqi+XQGzB57w+3w1ov7U3/uE3xlok1B+XBzO4gUokylZi + bTMDGsHIXVfsaj7IW0PzfJiViReMf/G681spmg4Aa8c7C6h2sUv4cK4y/ssfq/F5bODeB0eivWYm + pqwtz4hNyyv29TeuqWLHElD92SKBs51j7pCcM5gNwgPHn67K1/FX66j6pNFeiSuHf3x4dq7neW0N + fti0hswQdeA7A+XF5tQW0hGE9VwTPxESOi7KekWk7H7kNkGcL1ZykeHvnZ+IIjfSsHHpWEDV1s5Y + l+jgzddtctGbaChYr+ypZgVOzOCZDv6+3m69sB+nghmNfKyFT7fm+XjZ0FH/jcQKtVvOxbkh//Fz + /GQmSOcp0lUU1mM900KsvFWw6hlmSUWC4x9f6G7Ah/CUDcEm3HKwdTPd4E+UdexCxsqX1FhcVGS3 + DFs3XY75U5s8gfDN/vB2G0ah3xgUv/KQyDH/qddZNmfwtfMbdiDRtUWKTxHY+QnGhc4OC3yyCdpq + YgT3Y2fX/OXtc3DH3/nyhSdvx3MXLJF5CNjuC2LyopcrWNjIxvqut1hJ/VXSn16uDRbGW6i/Cyg+ + ohCf7Dai1CiFFtaa1OJAYJWYSvfehOQeSiRW4a0e9fsrAH/8A7efo7fn0wo6+V0iTnqp4j/8gmHn + hlj1L+w+6LKFECrcfe5Zv4rX72+pkLd3/XXI6azRNXZ6uPtJWAu6gDK7fwR2voT9Xa8uJ+2SgKt9 + N4mZvVBNAw2kMLiWDjZF66ptxutTQveRvPFp0k3AZ8a5g0n3fOA/PcErmToi8aFY+PwNP954OboB + tCNentlk1HI2omUP3cf1TYz1WXpb7GYb7LZhwdbjzVO6+0PgVHYWyc/SCDYtEzY4HD4t+VvP9UVk + G90e6gdbnUnBenqrI1K31p1rQwyGvdJ5ANmLFbB3/UiUWtyrgSI52jNTjUW+uPargmgpXzMnRfWw + qtDcoFG9ipl7lC3Y9lMu0O0EBms2ZwyLaD0CCOqrgS22awdyHG0ZWgEusbfsXbh/x1cF/q53fEtn + b3O17YpY0oMZPgYrn27llwPV/TIFq/u08+1bayOC98AK1olVtE1PJBne/LUg+vX+i9fj8vGR2GTc + vH7UoF6ct9Kjoy6w8++SevGaVh2Es/7bT8/8ymEknOLDdYsWjFl0GniV+6jIErY7lgV4qdniWj9h + v59wuehnj/LNrbfhid9GIjOnSzwSzvGhxqivoN7z1+zDA4QRfn4xziLf40yImb/PRItjSMc//rav + J/Glh67R5+VQ/OV7bJ4UpJHp6TbwXBKP6Epc0+Z1IFdYX7lvcJQfZs7c3w9JRI9Lgn3uvgJKzIcK + +blxcWxzRj3++X+RuNywbN4Nb0wORg/L+K3i4L530T/Y+r4fHxB7c3fLF9mWMrDn+0AQ2nlYJ0tV + 4a5fiKIMF8DXnmTCy+dlBMSwmbjPvmMAH4UsEMUHR7CaSV1COLo29otm9JazUCSgSSuDOL/8S1fy + +LRw8vWS4KhNAH12Pw50/jrNrFYNMf3Tb8x0fBBbMy2PiqdNhp5otvP05D9g5QclQQG2M3yzFt1b + p3ThIP0xIfGun4yu85ktkX+7rWTHH48yvymQFFxc8UMID/VGpGcK+VtQYHeYlXq7e44kvYmCgvoQ + ScO680/U+OOPFPW7oj/1xPaScBUhMU8GiZfpNm0QTc0Dp823icmjWBd0oslx5ieIY06I+wjmHTCI + /VztenWyhIED1ZK5bt5fbTjekx7dtuEVsEb69SjHnRnp0DMBOZfxnJMZvAvkF+oXG6L58X5DN5dw + 9bMpkN6OAZjHtTjA5VoO2GFkDSyF93v++bP//JNtPOMCbk7/DL7GM48HXD5meEscD9s6rMBcXscK + vpaPQp5g/oCVWqsNVovwRHWKJf98/HsPkiPnYTMCLt3q2e9AqE+InITwMKxM/A6Qa5B+PgqSXs8s + dHXYfh5XLH85QhdtU64w088tcZ7bvV7++CrCphNIm3fX/l3vyS8Ix4ob5Ez7ECEwQdURr7roMXMT + tSfa8XH3B0yNphfrCslw3AIGM6eYin2R/MvX8q5fiPBsQhjK9Z0Y570Na/qVRtjfYjz3Ob5RJncb + Af5d/9Scmfh3M1wT7p0vcFQmx3whnBNAdfBzonE9W6+EJBnspeKOH3elz6cdj+AsvR+zSJwnWPn0 + AuHJ0XAwdFKZ04msKTAOET/T3a9b56YU0K81NRLosu2tweHdANFfD1gtlrHeovBeQk6Ybex7zzZf + IexnuP87cYcex/x7jt0//MJWWb217dXlCZTe+ZOYnVEBqi1zCMalCog1y+96+xEhAUt2SLFy8Nt8 + vZWEAc+Egdgngr3v91MPHk4ik92f0FZcyBVEzUsicjQK2tzJVQ9bLyPEF/Cp5oPvWYane5TODKf3 + YLsaaQvLXlpIEMZA25+/D8gYgBncqzP4p6/2/Y1tKb1RKg2pBHc9j7WdLy9SfSzhWTDFnd943j89 + v/NrfD8tZszl223+q2cEixM5gIV3w4SH5/r7z/1ew2jny78I63ER5RMI7i00eRPMyr5fFrs09L/9 + j8/L/KHUcjwZlL2wEPs01F6nrv4C8RYGGOMfosPh1DTg8ry+ZkbsHkMvLn4L1zRw5q1qO21y8nOG + fKkExLPlo9b8LBkic/sF+J8/A9RchrFim3j3l4eVh4uNpjYL5x8Pzzn5EjVCe72IPMCvqsldv+hQ + 5bZ5Brt/vpW3nwl/wcPDMq0sb0l10AJ0XlVidx6v/cufTaOn2M4draaf6tKDnd9heddLf/sVtiYR + yclNoUeHRklQC1mdON+PrzE7P4dn1edJdtp7Li3Pfe7dju/OQ17iEYFXLxVZU+NgwNBbrsXbRqPb + e/jEjANd+CCsYBqc1llSss2b//zl66E6E81gi3y7Hc+bEJz5mBhqvmjL1/fcf+sfeb1Jp+PyCZAv + Py8Bgmmtkcl7BbDwTRwwKf54f/wX7Hpl/kJnydedvyCm6hJ8CZsDmBX4CIEelRTr1KJ0vbh+BLl3 + kv2rj62CNczQ13tx/qbcj1K7/QnSJap18udfroj9pvDm04LgwrbAP//Ne25cwART9G99QNdkn6DN + +0Fbuuuhg8e6j4gFnTBfzUunA/uZR8QNhK1u9bZkoL6GIVFnMfe26twnEoLmGki7HiH8kusguFZO + sDVFDQgOWgn++YWkn7SBZrRaIFPlXrCwaQW2i3Z7QseeVIzzuxWva58F8I9PW0VW1Dtf1eGej4hh + mU+PPqW0g6d7mBKPMVOPZm96gME1loP2IB7jZcfPf/oYz2du50PeCOR7m2DvaL29+XTY+cUSRPPG + fbaYbEtZwVQVOvy6W896oYMdwfvzXmP/fFy8dbJcFWiBXWD91HKARi1W4Rq/Y2K0oq7NL8m1Ib/G + WrAUbkqXOa0SdFFjQHS0qN6q27SAvi6pc8sUtjf/1UfGZtxIVqp9PUlJeEVznQlEfVYsoH9898+f + NLVvUC8IvDrIMVGOrd9ZyRsB/jqo+sWBBHu9a8FjE4Gdr+M/vFqCbVD/4hPLzNmqN3HRW1jdbxN2 + w62sh/jrNbDjrUdwcDit5nJDs9G3eh2IPp3Tf/UN+Fdfscs68tbRNP7lT+zZ8sujz+8Ugg+jKVhv + l/nv+2T4mKGIs/GO6SafrAbGaTTMK/32+XxY6IieH9Mg/s4fiJ+XEOH2uWHL/HD1zCyZvE+5UXDw + FK50188qvM7K8R9+L9m59CFbHDT8p6c4MwAR1E88JAZnALoel8mHCnu9E/cZKto/f8l5sV+87894 + C7etRe0abcQvGt+bQCmnaK+f4YTUb28xsnyD/GdesGuF07CETtBD5coZwfbOs2Hr5KpDbMyWWFsY + ySNrZJgwuVQFVtLhl893TxFgKTrgzz+LF/1+86Guxv0ssX6Vk8YAV6CmUTaLTcfm9IKI+xevf/xE + 6yEnFbBLvCLYVGAO/Hlyr1Bj5Be5vE4+4P7i159hh4PuytfU/nwSeEsbgZgH8ZX/BnCWUBxzBxLE + /rFenckORH4tfSzv+2sR77oMSi0W52NJWUDoyZVgwqjLfGw/R42+p2cD/ocTBdL/c6LA0sOZ1z9K + zAJiV3Crv0rQnWPN25x1zdB2HnOiP+9+vrqh+kRs1n32OXttvMbs1qN6qTiiPa5iTjOrlqF1Ky3y + goUQr10ZJzBSEjbgPsprP5MjJjApuyvW+SzxaMh5DOy5Yz1v99Ly2IyVI3CCekA0WLLalksnHeZu + OOOHIgnezGf3J1TlC0ceI2SH5TCeF8AdJh07RT3V632eN7jNFcEKgTJgcbo/qeOikafyonSSBrCB + +R46ODdeMB/7qVTh/HgHRMu/Xszjm9fDmv1Y2Jg07BE9in1wuPENUW3U0S0UfhmEZ+GFXVWa8/nI + ZhVsv21GTkTQ9/USUxDuGeqiBs3A8Ck14dk5P8ktmcV61cTnAZrBr8GOpD7Bxlz4Cg77mbjAyplh + TrgfB4Xf80SCTHrXy/lrNejthg8Sm7zkrZCrBGBz1SU4rmxG2SNTX9H5WAQkP156j/5+hSnt64kV + yRHyBcz3K1qFy5H4X1fNOeUYPSH7/j6wW+muxtcXNpVib+mJUy/bsPl5uaDvzDQ4u7fywPWLFaHr + 9ZPjR3q/5cyVagJa9LjEyvVNvZWr9QBC183natY+8cpPIIGVTApiey72WD2EDGRPXEhOGXgBjqtB + huy10An+3dacdm4uQOYTK8QPoUgHcvYEmEroRNwHz2tLqQ8mTFT8w0YaI7BWoZKiD1XV+VBZJp1c + sF4hf0558ozUyGOfUm0ieTIrrBwZdeAtqFbw92A2UpgViQerGUfQ6kZBNNq8wbo6aQKvbhORPD38 + 8rlhggy8qWjhIq0Nj+HEjQGZLSjk8SNhzofPWYU3psTYC24D2FatVZEvGk4AjfOgbcY9SCG5sz1R + 76WlLQzRRvQaIjyLtwGB5YXkHqnyjcNOYjQe512sCJ4N+U4uCnMYaPaSOSTGgkZuB+XnUf/uBfDw + eRkzi+yakg8ufUieIiBX+fLzaPiWNshALBB/Y1eN+lMSguv1m2Ofw048WVLjQrYdZoLBIag38WO3 + Uh/K/0falTQtCyvrH8RCJkmyZAaZgoCIO0BEUUSmAPn1p3i/e3dnd5ZWWQqh08/QoXskp+twByti + jBAVXxHjIow9ymNVy1HIXAi2PvIz5g/PUwn1IT9j2RLCeIT++Y38X1ZibUB8s1yG3wsdrMtAQsNO + qJAIZxlVC58HYOCRu2K9hhCWioLlJJ/pEk1PHcnPSxdIcXgbhHo0VXD/vTHZOxc1i2E2I5qrNMeX + S+wOfKRlb/DwuBeu8hoM4y91AjTfmwA717vXcOL9IEG5vR3JySE9WH+HuEbfq/nEJsCdu27+h0HG + A/5w8qH6wDf6SYL+ckn+4leb//LbXdFkonDik+7Xy8DP5vHETmo75jLm5SFd9FqiM6xXCKoj8jBP + LGZe9+8vjNxLUDB+EJ+iU+MKSpJvsLzmKdYffTgIOg4q6eyY/bw6vzWecT+9YPpKUuJ1I4oJjc46 + zCpbJPhbhPHiH50WHgI5wdq98QoexUaLxgtaccrgvFj1LurR5ekLQUVnvuCuj16F97mrcfxe1ma5 + GfcKVolqYO0XAHcNhS1Ax4xJcNJXiruNv81Eh62dcFDzr2HPryYwAl3DqvU4F8LHK0pojBtDzLej + FKMf1hW63fvD3ALcaYRK1JOGoQHz8Rn02mIdWQbszxOfXArddTxPJoQl75OT9voMa27qHYSdkBLv + 0VzdjYTlCG51shH3fGXc8dWKKjrVzgGr+TrEK2uFCwR5l5D9eTWceh9q0AHokWA55GB5st8WBr8D + Inp98eNlLSUPzJ84IvpX7+gYKVIAIw74M+2cH5jpR4Kwys4ODvj+o1EdB+U+tzklxtOd4nk2Pi0K + T+uV3J6Bo7Hq6ucSu1kp1qGxAFbQexXG0fQiZ9uyKXdYqze05orBpn8utelwN2qEH5cnvjub1SxZ + cMikzAsjkktKPSzSoLco+CGEy0cjaFSlOQM5MTLJKf2Ow8pcBhve7t2BKIF2apbwdYRguegj2eNJ + m/C+fJ8wCYnyuL2alS1+LISG6QRHUKfuFjDMGxqP052cD6za9L9h7dGOn9i+vI8DPRQPG15T40Ww + TaqYswIuRL6zv1PVy7LLJt4zkc6HKiDqrBnF1ugnEWSx7xD5s1hAOEo2C6Xbk/u3/qzK7kf4KDVI + AIVymPPq1kH+bLxxIcwVpaGTQKSrnIpVz5ndzZiuFfLfqbtPEXi5q56cWviaU4jNevm586WdM6jx + 6jSvsqQ1gljYHTrYUCPJkr806umbA1eNfokTbKErZN1rQenB7nGyDm3M5yZlUTt2A86GTXVZJpNH + 5DevAzGVm1SM3/KRwXQKdRw+WBPw0+Ung/Tg9KSsHgZYQG6HqFkTFWdXNfybmhAe9++T6Pd7xLwr + MwFktawgRTVdXT4L7xswtn7Ht+vLpc5nlZDIPVJiKGw1kAN/muHliQVito8UbJ1Yp6icrz6WH18K + KO97CwwDrsP3y+3jbkPBBH/xgJPHHQ0rziQZrU4ZkuxZFvFUdLc3/MunDJ3TYs631wx9BxGsGS0H + lqNAI7gG5BLAyjjQUjpcOnhrA0zi/vGIf6mtByio+++eb+7x3iO4hPllTMkVsbxGr6fXG5UH50ZO + aX4cZpVbXjC8vKIAvc83rbuEPEQGejs7P4iGvetmCrhhOgRLNNRgoyjb0I4vAfmgm7sk3jOFcp4v + +DRfF3cMH58AtmM/EPOjHJqteJEQOmEZBev14BRs9xJFqL+nbj66tNTW7XmQYXW/FlgLzltB03wW + wc4PMKa14bKhoMkwdrc+EIIEuysNpRmk3/xLTMoRuoaem0C88XUQ6rrWrOP5Y0rvbSTEpB2kS8R8 + csCJoUmc5vUsNgsUKRpYwyWR0oSDsFpPCFm7HInmy4G2jF08ouF5uxH/JgQDH3UTBD2PGmJIN1sb + hcuuyKPvGmzAc+NF/zYlXMXrAWPlrA4kfG4LPCOOx/6tL4b5cIcjlB+PvSv7PW8W93Dd0LyqfoCY + hzJwBXz1SHoU0cwO5FNsyyFM0MpKJjEhk7rzfHdzuJYJ/8c3taFMvxvI8+2LfV5M46V6Vab0WG+A + GN3nM6yN7EPpedU3YipULnY+noo7HwqOl9Ml3jqxS+EfPxF5bxwW6wgZ+G2MnjifSi1IXZxGeBIs + nuh8qoIxZ28mfH+MIKBw6tzVaswEntRnRczjuuz42UHIlEeOnIezXmzWV43+8T+8qT6Y5L0HiHIS + BaJ/6HvgPKJ2EJasT+QUXuONVeYSoEWcsDkdibsIei/D98cKZkZdRHcaUczABA7VzO//N1wYmkuZ + dPCD386v1nqf+rLjQ8DG20xnv3Ed2Cgywfc1P7ir/4ohOFkfMl8vB0QXyIEZvha/wXn7okNbiUYN + V239Eudhg2L1j+sC9vWcl7/967wmG/7xCyf1unjn1wksr1k60/j0pMurFWWIAF+QP/618m3GwCYe + nwHlzBJwj8iG8N6t2qxXt7HY4IMNIboX8wz6bQCbKzMeyN5qSgK3sxoe334eBJdWC7b7eh/mCuQh + fNTJMxAs6wQ29K16yG35j2gGtcG6PQUV1g/2SyJTkdw/PgpWBpc4CLob3bIvbQGT9Qs+sfXbpeEx + e4P0ePYDWnyHePnTZwc1zokp1uMwdm5sIv5svff7+cVL8LMcwFQvn/inp1ZwSp97YOuGKJCCg6YR + 4h9sKNll8S+/U/kcqOCBoEKCgonoRLvw//A8g7bfTDImPQzex/HvejQ6vq0N+jLnk0pW4DBeQp4B + dmZ1xNKzoFgbLZoh3x4UYpSTRrsdj0BS9ymu/Lc9sH/x+675E9Z2PrT6/CRDqbD9XR9s2pyJgghZ + gaFYv9u2tr0+px4kK1mIc2vrhuoSShG7ia+5Ebgs3hDDVnCbpoZUCVPT7XvnNmgejRyfkFs3Y9be + A0lIH3bApPjjbsy02DD66WLw6nnZ5bZjyUPVM5dgmUcJ0BY+GCk2xxv2sVa5WzRsM/QNUO18TG3W + Jl9e0M6MLkBkOLtrvtIXNJ3s8cc/Yxo6JQMP3g1iA9ka/fs9+G2sPlh3/B/Yux3AlzyV80d7dHs3 + szmB4vyqAuCbVsPmq5eC+xUJwdE8K/HGTKIDV72l2GDvViFkZ535x59+83XRtuIDVfD4ZBve8WMf + K3Ee0eQ5z0Da+c2WfcEbxLfcJXpHebD47/UFPVw+iEdV3Z1+4VWEutYF5Ay+eKDw5PfwgWyOJPtZ + NuotTxmGj44lkTvqBXU+RxGmRq1gfbg8Ckqxl6Koi8fgWCk6Xe/WjYWZFShYXUOnIbUp9lCIPA0n + qOm0NbSZFi7frSKBilyX8x/Yg9X9UuDqVvYxdYLERHLSxDPbn2e3b7KIAYDeN4yLZh5G4XfIJB+a + AVHNYHK3tQpbaIwL87efC3phaAZ3PyRAOFw1ynGzBLkt+xFH50nct9PRgzSDH3KTae8Ol/QWwiVn + Ttj3txvY9A+nA/FO9orAQQHcwjxtdE7FA3Y3760tL6DoxwwsN6K6ox4L8tlUAWpij9hGEBQ8NjIH + dKtc4Ug4mNrG4z6FD5XIgURyTVvUWtGR+2CyQNrxbdn1ElTYdCLmLV3o+pd//vyI4nBxNKo6C48O + PfsLtvbnu8Iff9n32yy1r3jY0rtXweAzijidTVXbVlKKYGIPh2C8K3OzWV8ngvPgEOxKQj0sBzdX + ofxiZCL/HFnj+MbzwN00ztiRFLkRbO0N4SQJJcEG8TS+Gd8yegwhDljU2NrKS70MM2eQgpUbAkq1 + 5lHDP7/BCu+axv3Chwjb02nClyy+0xVaJIK2Yq8kZcqpWAfHGf/pQVwsSrOsE1/DyCxMrEeJ0NA2 + 5ljoeIeYGC6K4hmyfgoR0PDuP321ed2eHfrDb/+o9MV6STYR/lRLxCrwhmK734wElPr4xvrOP8df + qnroQ2UVR2V9K7iF+TnwOHQmSVeTLfrDUymBYAwwYL96BxZDSyEsy/KLzS+ata17Oi9o6k915rBz + bdbayEMYakDC3gvhYuko58FoZi5YlQ57efb67NFDAewsFPfLHu/8G+56hci1cW+mg6d4yOxk+Z// + tSHwHWHSLyIpJKFu6Lt+16j7hXkg7PyP6Ixqou7ePGagPnpt35/dP39NkVK3WPQ4yqHVfZ74sbz5 + Ya3TqoU89hAOuFsNlsGrQ3CMdQGXS2w30wXdVWCVbTZzgxVoVH2cRIB+zg/LZ/83UOu7lkhURgdf + L4c7EGYzWZBSrPf5zx9cf/HYwX6bq2CZCr4Z//jsx+/u2DgVJ8r7waYDKb8qQQcvdbO615GHd1Bx + xH5Nj4HF8iRD1T2/A3bfX8I+vx5NjfTBsmGzdKtwk0M7k9gZVsaDjhVPHBjLLMQVBaeY+6ldgqTN + MrB+AyoQ2s6UYJXFTnBAEhkW4Sfk8PhZfoGQzLeBLbwDD7Kf+cXyg23BsvuJKD3G/ryvV0Oup/4N + 1TMTYD++6+725H0GDKzlYkWewmaDDxiCob33GOvVly4rG6r/+KML27mhR0pSqVFUQjz2/XC3uw4j + 6FBbJ6p3fbrLwTZCxNQZi8+ypA18Q8QNPtNzS9y8lId5cI4iTGwpIuahrprlnoMZ+tJj56NC0LDe + Wnuo14IrDpBOmnG2Ygb++R0aF2Ow8IELofbBT2IdoqkY+ePGwwxst+CrZ3O8rduzR3N/aXAEvCHe + JBnq0LioADuyG1G+8qoeJP7pSszb6U63nc+AbvEj7INmAyPWjhvUtT4Ilt0/olQCAcSblpA/vrwp + g5KhuyneiWpFVUF+Y82is+77M5PXxbBlUO9hG2AmeDuaotF2rjNwIJJENLqf4l4iVweIgwzWOQjB + 9ogWB+nWMcZqWR8LOjJSCvsVRERRmAQs8Nn3UMFbQvRz2MY0EW7y3/7C//w8z1hTwGp5QZRLGIBV + T5QW5v7J/OPLMV3gtkHWGuoAHC69RveD2PC492lidv5AlyTr4b7fgs16rPGMYv8Nl5ElOD3JxrC4 + h8cm+d0nJLc0PzZbEaQBPNZjRsrRvDRLbr886WWsE3YSoGvkGNwg0G77lEv5cnL/6WXd3xaiWefa + XRM/rFFDyG9e5jGn7JXPOlhPpf2Ht/GSeL9UUq/yA9+1jBRbcX4xUK+rhqjnY0H/+dHNmwIcfKyb + u5gnsQLyFxvBwpvU3X7DsYO3rJ2x/pN/YBGWTAK5mNnYPa+6Ox2fegdf18zA9k9rh+1Pv0SNX2A5 + hvtcsttJl6KfqOP4nt4K2g5FB8/HdiWG3EjxZpSxA3e9jY1G6ujfZymINZM462l1FzzGEkhsw5xX + Qnw6LEnWHbtVrbBl81e6POuq/re+sn2R3D89IB00vSXX0j6DiQ+ojeylF2dxO3zjyZ64ChJ7CUmu + /2C8zTOepZ2vzOxL4ABNZ4GBP9hMxE45RNfflxuh+fVYovRy7S7PpgnRWHb53E/Qpeys4wC29NvM + 6+vGgiV8TB6E7PuM01dQx/R66lsIThrEKkl+zRZpWQsLw7sHLDuIAzXZzoEHntvIiXDWP70KwkfP + Yu9najHluFaCfHridrytXXJf8wTKk/7CN+d1BZMVqTKUtNwhp0d00kY2l3o4XtsT1ug1oMvAqTmc + QNfNh5WTwPD4Ljk6vg/3GYTmja4tY6bojw8ZXbYNU6vEDtjkI8Z+2p8KCsVT+C+eRabnXcJxswiX + UC2CLVMJmOs8f8G/+svjzNaU8lDOof55sX/4PHwzSHn4ueRPfNJexvDHZ4930zrPM3eTKZffXxW6 + vBQNm81Db+hpq9s/fon/6jGTflhKRPm3Sa5Fb//zmxH7OSvBuvegW7FSBjAPe4MYV47dT3TOC7QK + 6s0/W8bx9nbfIXBbEM4Sl+ruWqfpG+FBL4NkX//+lxQ8eGtB59OH0gyL6i4yopXME6+I5IFuQlvC + eHUwDtqtiTf9d2KlHR9weQ+Uhv7Qx0F/+nQ43k7xcv7iFp5qxiWGV0B3hdY3hFf3RIlcpWTYnPDY + wxz+fJzSrqTjqlo98FhWx8ruj7HoTAJ4q0oXR2/rPdA/vgan4zdg9vVapetrhrTiVWJ/eKUQTLa2 + 0dxfGyJHlQ+4gTu+4Y63GD/HV7MWIdhgvtUD9hVJ1Db9PMqwHzV2ZhJGBpwke+K//HMWTqa7HrnL + CKroNhF3rzfNyz5FSHFmjag7f1w3f2JAY0refj+4Gf/8zF3f/+mlYZ0fHYSrU4U4uGGhWDOmD8Cb + 4IVgvbIAfQaxDN4Clfd64aFZ71lqw8t4+hD8DVDc81/ZRlgQ12BLuL7ZoJDr0l+864tPNGraawg/ + W8DPMPq+3ZV7HErwZfee4ZoLi/VO5RFNfHsNuMhs4z+/C+34gU1NfzdLIbYzRD/7h69w0Sn900Mv + 4th496+G7cJ4AbivaYwDZWzBErqlDclQqgGS+ZZu7HRP9p60PdHTQ6Jthu3of/ttPnadGq/Hp9fB + O5+/56PzEuhW/OwNemEj7fwNDVurqglwLnVDzJzttFUa6IKKYO9BzK1dvP7lp3AKL396cZg2/wNh + /W0HYnx+KV2CH3bgseelf/jzV++Bq/6mWO/1wt39rwxqiWNhh0vf2si/LjL6ZcuTZOZoDHSB0gJv + n5uGT5fULrilD3VYJ+IRmy5RXG69/3KgdsDArm9+h/EogBCuNJ6xOn4Cl+b0LMJjr83z89V+aTd2 + xQgnz36S7Pmbi2VVre6YBsI1EBnh5xLySSAM3uD/9f/BMnXAP6VlZvf68DLUawr//GD/cvtowpKE + PRp18UqUsv9q46xjDxqodYhRqMeC1FcIYd2uj4DzOqJNnDvn8AGvP6ILX8tdLWm0Qf2cApz1iQOW + Jj+l0GNVMfjITV5s3y3ggXmUWhKoS+ZSrEo7f4MBVsyyipfPsxmhUKYfrFy0gu743YJXWaVEfwkc + /fN3wd7lfEa1rA5CBU8V7MBJxmqGX2CzryYPj290n/NdD62CqpdQ778G9h6L49JPs0aoz2FOLj/+ + B/75qR2vE6z14xavWUQT6ZKHh0CAKQMIcxh6KLpaTVzjURafNnpUx70+hK3LAQHaVSOE1zN7J8aP + P4EBenkOgjtk/8Xn95Z9k/+lRwH47ycK3NGTSXZOVHfTf6IDe/XABGwtv+gyXG88lLfkEQjBd6Xr + 068j5Co6ix/fnxlPcu3wkNd+Cbm0/Glg0d1PQGpsBXGEVo/ZYZJq+JTNgHjQzDRhkrkWQgGQIDrE + QKNDNywwFZoHCbj7zaWfqpeh9DZr4p+lY9zVm7zBI73YOOfjD1g/gx+A+Ds+SdnxSrw8tk+G5kYo + sMHxRrHKtcpL0Zoy2MCdUbDa6QzhW+84UnSIBzNrxAz6Hpw5YESlAeOBKhkcbnQLfu+31QiUt3q0 + SRokhhOoA71sxxR8ivJHzv23jhemrGfU5VyKT0cL0k2XQwmt0bWd2YvlFixxry9AtPmMb/ctKVi2 + nRcwR8yReO99LJV5/+Ww+gYBtkWubpbHKbCP9Sm64UBPSLyMN0eCUWp+iWnuc6kv7ndBg+5+iS5P + HuCNdH0jfRO5QBjSu8aN2xrBb1xZM5BVblif8JAI1ygoiMX4L5dbjpmEKsjf5kPJyYAlLL/BJgZ3 + rFvPKOZgHcuo/LUfjLdNpH3/BBBm2uERrLNQFFx0NmfkgKrD5lg1YPN9dkFelh6DQ/NSXW56SDO0 + jDkjp0ar48244RDw1f1GEg7lgKucfoGFVeckGcsY8PsBMvR6TWzAWEo4LMcklAFProCYL7EHlExR + gkSZH4nDxwblhMxJ4Q3XEKshI7ibtcAM+jgWdsdfjKkyWzpYcpiSh7S0xcQldYAS1n5iI+3jgTSU + DwVLrity+witu1xBH0D+RxbihhdCV+lhvuHpl9+Jd0CKJiS20cHQliccTpE8sOyrltFf/GI5DQdh + nhhmP8RgkDjN7GELodoiZX5csZlL93itnNeCZKG+kfgQn+j39zl1sLccjwQYH92luNilpGrMLTig + 9+jy69XUAfu0RZK0rF2w9f3sIHVpGuI9QrvgE3n1YJ28Fnw6ZRc6BfQ3w8DhLiQRrFezVe+bA67X + 7DL/bOMabykZOgC4OQ/oIf7Rz7DaATq9vYa4Q0PB+2n/XkdGW5J57b910Vno+kav7nrA2sEzGmG1 + LgFq7wMzi+n8psJjkCEivcdg/6McY9pQJoQZG7bEUmwU0/kit4jhLIvg8GU2rOqSDkaKIgRi9K0H + 4dtwntRan1cATbEs+vo3JhCeuRir/U8ZOO34lKE+RT/isEdN42S38KT1ubk4uZeZywv2rYTgaQCc + HEFXLLKZVMjkIitgIo2hy1ylJpppKBMfm7R5D+OphH/xLr6Ll8vzrJkj74gLbDrXzl155dlDkjiv + mXf7tumu7dai7FfWpKDdp9k+OujB9/2g82EQArAEN5qg7jKUJJjwMqxaY6cwu337maNR3mzl8ZMf + mXfHE79NrkDAhzBC/kk/k1vd5tr0vfbqv/3mNBfYbFpfbpJrRjNRfndxIOF4HpGhXc2961YElk8j + brADzh2b06oWvPx9RDC+cgMuoSlqRM02HqbC8xFA0rwG4VKGPTp82AfW4veL8pIIGVhX4hvfsuw8 + 0PfavQHIKw1X6OXsr+U6KlA5W8WKmziFYNrHBTAnJGH31MBhgYsrQouXFGL8ehyvQxm26HUebXwN + vmdKtqL24BucXvNSFaCZNkhVYD5WY16RILhrmwU9nNyRYty/l5gGydhCCw7pzLX8qdkgfOhwXKQ5 + kJJ4/pdf0cibHbY2xdOmTt1yRLC1Ej+pdE3IRP2F1P5hBtIjyOKN6YD093xJaZJPs2ZqZIJmJi9i + ObdTvD2HSwavU7ViLXJkbZtChUVtFFVEr2UVLNhkF/RNtFfAOcGrmbQgnOHf/XvSyGnr3+fTkZ+x + u12+LqtanQRPGiiJ8VvjZt6UlYXSbx4wfgwiXdhXJyOjkhWc3q4a5dWumVHJspAo+q8YVvVTqZBi + fcRxlBfuqFq1hOzIzkikMKDYVrUqJemhEKKVcuIOS15sUvCQjJnPJRSTbfu84OiggvzFw8e/KzIS + zyHa46kcFq51bBiKi0GK1T0Um/5bHKT84jt2F6AXfFbaJUQ01bDLl0O8xDzooJaq71kIFRZsnGXI + UJrFDzF0lDaEdfL5iIW2JEbp5oBbXqIHx9bfsMmOXsGP4VxBwI05sUrfAYJ7/urAU6M7Ph+XiAo5 + uo4IxTLE6eGbulxgn3MEjFtMohQ37npnVRlFhmfgTMpybcl/oo5Y8wGDMitVynrp2UP1I+GIMiUK + IJdtTVFoq1PAQTEehNuReyO1NiJcTmYZs4HhjvDbW7tjxL615Z7HHQL7VKX0dm3AsjcihZ99pog1 + mKrL40ZkwRh+v+RBGKpN8CQlaP1mKY4S+B4o/3pX4PVNY+JeT5eYNfdeZLg7CQTv+CKIX6EHpXS3 + 9hMOVrw1/bKhV3c5kIukT3Q9hEwLYYkt4hbhrdmMdG0RCP0bDk5z5G6N9vXgww97HLcAASqPqQza + r/jD+fnRUbJ3Q0PvE0sCNHwaShzMyFD2bt7Ma58ELKE98YB5yD02JN2nbK+ZPBSwtxJDa+KBZk0i + Qgr8Dyk41W5Ic7+0yKz9K8kF/hCvs6iXyPlkHil1VY955/mAMIXTDVtYWoql+2kQ6lP4w/onN4Yt + PuwdwfYuWoplKmDNV/0Ffs+fS06hpoOFccwNfe3VDZDyfbh0POgVfKjlh1janBdUFOsXLO5vntih + I8WUGZU3+hhsha/aLBUbiRuIHvPJ+8f3Vka6s1A05RZHh7jQyI7v8K0HbxxIAgDTtbdDeDWhThSD + asVyDc+ipNg5IOrFKOIt0QUIx8v6xFgS+maNmrqHm2YyRBM7oK2XQ/ICR/W7kD0etC1K/Qo5QL6S + y0nPgeCKYgeVSjNnxCW06eXwDJH0GweSjT9n4L+01SG6ZwvRz7Ll7tfPwCsC6n6GW3PJpTqN8PVN + Ynz7OvqwzedNhPUj5XDwcSq6RsFRlZRuVrFfSxdt1F1N/Yt3rFGZDOs7GV/wzK019pTqNlBweC/S + 4tQGuUlnfRjEc8rAUXZToiWMQalelgzgvFDHbsdHzXasGw+66XwIartqhq2zjg78w6M/fB6BTt9w + enYmyT6z3LApNHk4guiFT4ozFcT7bCYUuPYyH6XnY1h1EJnQGrFBzPct+tuvpjR5Jz3oBvwdtjCU + IrgYx5aYe77aDv2bge1rJCQaHRvQZJUj8L5YJrHeBY1pRAD7t74BOkvHYnPs5xuejuxMrE0Z3fVn + li/psthh8HoIQbG950hECq1tfALvRqPe2algLDpb0HDspdjWX6dC/VW+ZzH4xe6Wml4Op5aViV6R + tCEiViLYDfEXyzy1Bl5BcQJhK2rE2J/3lsyPAMrjGONKqX/N9jv1HjgdquPMgqAFXXcYe5i8zwNR + 5UKh86qmJXTaQtz581ZMenhaYJoGIzklzklj335mgly0OKJEmI1XbpAlcOPEmjih5xTCIWd1cKU6 + 2vPDcdg6VxChPM7xzg9v7ua/LxU0qk9J9HuZabxwP+mQcI9LQNCwFC1vdhnsLr6LXf10BAt9czPs + M3Um2jqtLpV7OYGy8LoRSz+YA99oxIPaSjwcVJilY7WwCdRFnez7+zPQPJA71OH3Yb9fk/K6nEnS + 8Os68pDTpRk8GG0IXhWfeAGatT3eIDifgpR4F1F016v8UyVu2zas73x/BfT8Rq1adFi5OUyxeNks + HX8wMYmcJ9d4SY9HB7BymJK72qsD/6ef0udD+7ffxkfRmfD2OPjEu5BnsUpm5oHLQgZiIS2Pvzk+ + zlBNeED0i+XGyzE96XDPv/gqQt5d/Uc1//ELEoQ/qI3FxLPAbmSJJC8+pyTs6w7dlvCCcynvtcXP + nAi204slxkoUjb1+Ti9Uk9XEp+TsF7P+W2zIzwkm1a7PtvzRt9LOd4LWd31Kh65ZwFH9LNgoVd1d + 2nDzICi0KuCCiwmE541WkAL8mddRaQqCNJMBgrThGV6IUpDbx4DS8R4OGO94tH6vL1ny7V7GljV+ + wSbQ3wIEsIJgUC6JtjFoeMFdX83MMVHjidMOqfTHX4KwNgrB5oADstunn03u0IN/fOZbcV+M14zX + 1rIQI+gTe8LazQtiWrRRjhwn3LuSb9+4xsU7ByeidsTO8OiuUdP1kKNYw5qk3umYNEiHunW9zmvu + P0EnDvdU9Cyxm6XTWS4EPDIp9JjiQrysfIFVaUYG3kG3zFQfhuHL+yyPvCw5kovVX2mH7PsIgg98 + kfjLPIuRMWUZ3ovLD1t1gjXyFUceHaOy3Xt0GQ3XfAcdTrBy8SkVnHjJR38B54B/Ej3wnvF6010H + DoeIEu+lB3/xKUP/NIzYZlKr2eKDbEql3nCB+D3nLnH2qQlXr8UBw3C/YhvDuQS5mazkfnU/xWiG + VQC25lBjp8zkgp6O9xD88Ze/fMBL/ruDf/GpF2ChlPJWB8DTAkT7Pexm9wtGuOsfrNWWWEyHoBP/ + 4gmnu97d4rG10XJJXWxWa9vQ4AClP75IdFV2Y1Z7SD0som39x4fY1ZFE+J3nO1bGcxWPO/6CqeVl + bP4oC8aInmt0I6xGLCg1wx5fDPSlQp9/4K1p2+2+1tAJryXxj5HibhiS4O9zID2PVsEu97WCWBmc + f3j0wY5awcJ65VjFdwL6R6t6MKjTK5ENKaOL82Nz5N1riq1DXQ/L6kgSBCD9zC8tqzT2BDcGQplZ + sSKmi0aLNs/gVl0TcpLOesNjSDyoR58ZK2vRahuT6NEfXyWqvQTNvPM1CIYHwfqud1d/Xbp//HHH + a7p11Y2H3nTQiXYxepdC5S7CZ/h6EOv+9ZvtVNQ25FArBPW1Xpu1t/0WNL4gEyc6Ds1ioUcLmf6s + EP3Kjs182542SmIux/bXbgridniBYVYyWGk436UJvTNQXN/6DFs3Kja1G0a48D2DnZochu3RqgGk + 3Ccn5q2Whqn5NrrE2DDY9asdU3TbZydZFsZO9hS0dX3/9p4jfIiDHX/mJrdEiJH3xckzVgo+aIMa + 5Nu9xydFO2mcWLwr+IrODd7jDayPF6xB5VlKsK1vt1kkzamh60pHYhv2qG2nTWbRNtMPweHPoYIr + PUdkya+KqA2zDb2pURWddbnEVoJRvFyk+wJlZL//4fGMP08JvLa2x/41Kd213EIGuZsgBotep9rm + PK8QfH9mhM+mxsSLsto2fLD4Ny+Pj1as7WWBIL6G4Xzb8ymXRLEDSnvUcLzH1yY8LRad3yITAKMU + 41+xoTd0HouMY+3cDVOCmhKJm/vA7in/0CXrmg6lDryQy+160dZuVue//EIUJFxdNhHfM3AVk8WG + 2A/uxF48HszMr8byVFyLKRKzCIJPeJrX9hfR9dlWPdz5H745Q1Js5yTJkP7tKnyP9ae7yeGZgW/d + exP7nLzctS85D5qSGBGL+G3B3mJWBpGiCYF6dyZ3cdUggHCdT9h5GM9mvcpPGaqn9j1nl8IsuM+B + ZQFrVGGwaXajTctLDKAkORC7ERjpugT2G/z5q398a/tWTAn3eCfna10XK2VGE0oPjZAg+f6G7X2V + vH/5PFxvOmBJKefwgBeV3IZs0tZDyL/RaW6v2LiE4T+/BOz4ju8f5RYLnSh7iKO+hu9hzQ9T6eIc + xKNY7PntCbbyOGWgncyGeAy9FSuuGweOLd6C49d5N3sFZ0MafJ5xeK3PDW15uvtfTB9saFGaTYm0 + fWJlKJM4cmp3+SLQQvk6NMH1Oa+A8EffgfjruNhnxobOp83mQXYwBxJA296n5MkegkqhBvPOn6mI + lRDdLp/bvKJZLjaRi3jYW+191/+/uBfsWwWu6soTf/VJ8acvoS/ddOwCfnZ7Iz228FB3n0CAUtNQ + 962oyErf0p9+KKj0LVSYm+lKdr4OJla+e+B7sGeszAKI50X+5H/+OHFGwA00Xt8SrExgzX0rD83u + J7ZQ6Ub1L7/TPb55iFQnJ6dWLun25+cYjNjMHO4+u5/GRzAtWAGHXP1zV+QnFdz5J3Eaz6B81T3E + oxVCMotlGhb0aWgsvH1KjLX5OMTbRe1DOB8f+6j1pS+mJbJSKVR5SrymCsGqg1yHjH4lwe+bPxoB + K7MMmoNL//xAukocicDGdAVOE/MKVncZA6AcrsfAUzulGLPIYf75OeWN/wzb4cIl4J4vwTyjVw/o + y2Ql6G6cuL+TL7mr8gQi2J9PwO/6e9r5IXylkjpv6YPR5ugwzvDmXNWAQS8HfJlp6+A//Cp1I6YO + CBKw5z9yBtoaz+DD55A+aoQ9/hPSbc/P8Jt1ComjHLjr5Lo6yEWDI2rx+4BtudxV6H4/T2zW0UTJ + Xz3BXXkPe7cDjEcL7T1lLA8TTWF6d5Sum40M+3Cev0jLi1XI1BS6T5QSwwZvjZXNsgS5YJ6wihZl + YP/qEZyYo33O9i1e+Fh7Q/xqRuLcwQ0sj6LWkXRIeKJ3wRCv9/svhV5HHSLbc9UsF3GfOihfKuwP + XEWnJqIV2v3oQLhE+xSlvf5yN2J/vghh5/7hFTykN4n4ratov7u9jqjP5BnLaF6bMeuaHmruegw2 + T4gAPZBmhNN5DfcTPjydgZglf3qTuO3prC3Jy1Ol6kt/AQQH2lB1elVwuK0bOfmPXFt9TsjhJimQ + BKhoh2F/vkgIjYTo3lGlf/cHQgYN8+vu+O5UPd0S/uGbNu49EhgnWOCdVz9/eNj8ZtGrYJgBPN+d + A4nJ7SVGYMd74pvi4K7vX+n8wzfHC0pA/e/gwFa9dcS+FGZM0aWWYP7FccAWIKRTFKwqNOL1ik+X + 9wD6zlodWJVNRCyn9pv5x241FEIrwf/qLX//F1x0meTMM4ipvnpveDltI3b4shxIeop0aG3nFz49 + FwVsf3p/azJjFh/a2lDpG6vwkTXK7HnhCLbyjRZpGbLvPLEK5/5Ky+6BKFxvAcshiY76XnEMjKHA + p+hY0p1vvdCJyB3xle9Bm5cBzfD6ZduZL1ANNuDbMkxuSoItxb4Xi33ue6iSn4zzbpq0OWcjT2pY + epgFl45uZ4heDs8nL8Vn/Vc0pM7UHFVKBIn7CCyNKpdshuaglNjQSlzwvg8XwHmRTnQ2fxbz7m9L + Wlty2L4UbUwjfXjBFh2MAARs16x//kqmoQd2PKN11+P9Xv/x590/KdzFahkJhPFZn1+vNwZUX/U3 + GEH4Il4f8cP3ftEl2BHJxLh4J810I3EJIXpKwcF+3zSy7284XYsiQLV02XuaHnN45yue+EZlF/Oq + VhWY0oMTrLgIhp1vQbj7wbOw69nVnuZZ+uMrux8as1M5qX/6NlhZ7FJ6VqUSALxaRPtkesEawceB + evSdsXUuhGEDviwj/Kxl8lgz3p2fCZyhHb1/+HZ5D3RdgQJh+5V+2Oe/mzvt+RFelmmYOjUxC8G0 + 5BlcpFrHspsUA4+l80sqGqxO5Ax/xT88ZsP5S8z7fwAAAP//pF3J1qowEn4gFjJJkiUzyBQmFXeA + yqCIMgTI0/fhv73sXe89CEmlvqGSyqPXqHydc+ljU5fIpahq2zZEOtr5OLGt+attsK3mP/9sJotx + HTb/aOiQs75vYvcD0hZcvDK0FO36r/6yXlVmlCKOVsEXvFqX5scQou8numB1Xt5gcMXlh4bk5QV0 + 95PpdjJjeGwkAbtPS6VDPQ+6ZNitjOXdv19CM7PR81i0/8b/m6KhhDv+Ez3OT2BTE9YEZ/7M49Ny + TAv+Y8o6fNQvjWg7nk/sWWdh7xQG9ubw5a6X8QyRoLNoRo2ftHt9VET8d1rIDQxWQdLVDkGjCDG2 + zcIZxktMXn98cdcPcrtI43cDMY9Of37Dv3owUGhjB9TKRsonV775w4tgQk2pbYOY9uCw6jw2GDd2 + fzAWWDj4jEzsCZ609TLeGVgGaUDi3R8crcBuANdcanzqBxeQ92nL//ykeV1saxhfr3MOvdY+kFN9 + FNsxch4i3OvBwcFnCjAubGz/PQ+rF6y5e33oAh91pxHbv6TtejRZFlrGmOHcZZSBPbDSQyqlp4UD + hjsVSysUDLz67g+bs3JPlsDQxj//HV8Hs9EIVM4SUvXsR557fZVlrKMON8ZzcPa3Xk1LHiEUjmQW + vlBKaDQ7G7zUd23m4nJwyV/9KYexNi/r7QXYthlTMFtjQ56sZrj8WW1CYLyG48wkvtlO/L7j/f/Y + UQD/944CTR2ORPaWhq4vOxQRuVmPgA9PARAYHf5gOigxPrmVQOlnIgEULVbCbj+ChOK2CtCkX2Ki + qhfkLrc3m4GVXBE5eaHScu3tKcET1mViaT+r5bRqUoFXPZYA8N8xWdOFT+GlG0OiF0ZTTNH1E4J2 + sj38MAq53UQk6fBdvD/4cXivgN6FlkfM5hJiz5ilBKZ1ioJU+2Jv0G13u1zzGJCb8cAZ39wGcsSL + g0TGGYh3fFyTZRbDDvkCZLE7On4xpmUsQyZ+kFlcJ5PyVh7zkHE5NH8EtxqWF8xU+N36lkSmlGqr + OSQ5KtsHIvJdzIulXH4bkKnqEkM0kkK4uY4Iz0Un43C48wNtxd8IvndWJ9pbzhO6khcDRY/JsDcw + Pzpp2vADi+0csal6qF2H8vuD2cH4EOXVeho3x5uNlu/VI9qCNZel1uIhFLy4+UjOjsYT5djBtHLW + gCxZMyxMZDzgmQEusQy+cQXSZDHilaggxvA8tUtKHwFQFCxiT3yl7dJ+NhuZObxhp7k9iuXntD3S + YjXFJ8hdBr7WOgeZFqLYj1mnWNOEhWiAvI918fIeOFw/NvDo69Pe9V+kVIwfJUSSapB77nuFkGSv + EfJw/JE78ywH7mA6JvKX+REwHX6AIVX6Cn5B7hEvV45gzhEvg/XYScRkJEZbSD9I8L7hDJ8C5gR4 + MWVSCJ0CEkXEMZ2Wztu7ZrohMX7vjvbGhzeRlCEQHG7BCQhyNXRgXn57F1qtGKbkLDPo1IpPUu7x + sJa3SYbg/X6Qx+9gJ/P1tReE5vMDh4djB5YniGwUta+KPMGqF7xz6St4TeNsRuZpAV//8HJQUESH + oJGba8tCkrNwMJQTObv2N1kF9vqCTVG4xNhEt90u9nG/x1nysdWnI2BHiZHA66qfyPV+z4AAgtEB + h+xekUQ4Vy5voVsGI1PNZsBdU7q4Wmgi2TR6ErPFu1jL3piheB7VYNPlU7tFljRL43oeg61xcUG9 + T9ygEbx64mmpARYYRw/4S3h+ht5RLdaFfY2oO453nPLjUVswE82o8+wzSYriTvk8F0q4xw921t5y + t2McyuixLSYp6XWkQ9RHObpWV5/o9SzRaTveRag38X5GOQkKFjO3EQp3CWHDy2ttO7NFD1l0q4Pj + ASQDH2inBl7dZibqzzwkayxNLITLNOPCDMKBfY5LiW41DbBv4FmbdUXbkBC1pwA6k0Unjms7xKex + gk/ehofFuN0zQP0H/Rdfsza+ebSNYYV9q7ALzgxyWTo9qDz3cfhNKLBnBha1fsVeWN7blSZHGfjH + hzXnEjcPNLu1r3/jUTBwdVdDvfb7Hr4bDlJfb9nekEqos4ZETH3xgCByZxVGP5Ynbp3cEmLw9gUC + sdOxHZYt3ZxLVaF+sB9EceMe0GNW9PCmmgEx9SzTaPtTVbSMh+tMQ7V06fER8uiXsDy+PiMmmS/X + OIZmztyI9e1eBVu3MoPyM7RwQGNf4z9TtSHxXajzfOXIQB/EvsCHNg/4ji9isfT9UYWsZS1YFoKp + oBW8yyDeUhCM37AZhPtXs9GRPd1JvvaWtsoiF8Df+3LC8nC/DOs3AjHY80GwCJ3qVty32eB7J8FG + KBzcWX/4Duw85xxsE+y09cseeXi/2C3BS6a2U3t7ilCBg4XVtEtctk8LB3XdXrFkqEKXnz6XUDge + NuJ0a6IJFste0P5+OBB+NZ3yfFXRWxUYcjLWJlm6h9MD/7LvseSb4zBp1VtG5MJ02H6ABowDdMS/ + /EH8OfrQ+TS9RsRJFz1Y9vxNnFT+oTnzXGLzzbFdpf60QD9xLrMYli1gfdUzpY/4EAPmAjzwt36l + bb8X1Ti0hibEbib/G/+4vZt0cWbJgY4ThDO9HyZ3qYPBhNUrsPD5cqzc+YuLHB7Tx50oqfAutkPe + ZZJ/wA7x4l6g9K3d7L/1MzPMQLSt+iwzDJ9zRxyJC1p6Ok4VVCchD6xELAv29oYZXLPsQ56fqQeE + /OiG+hN08bWYNZfrLE+ExcCIODhbn4Tm4GIj/AHNLErSXCx5IAYwarsK4wLGGhV6OwUf6bMQ16p6 + l3sdFxtRIV9mMJU64NxTMqOZ25L591wMwBti78GyLRGOTubc0v5Yv9BdQgq+z/QMBFYqLjC7g5R4 + pc8Wy/kQvtA90gxs9GudUCXkRWR3BxgcRnmiSyUZNnK8qz3T000byIjMBfCz0AQfxpvAkpa5DKvt + xuLk46UtC700RPmHHjBu0LMgQNZK5NSnDwnG6O/MgMzA/tMk2FOqqviBdy3B3u8u5JKVybBeQ6dH + cxa4wSEmziAwfCKCHZ+IIYRVIkyd+oIaY9lYeTB4YKXSZuCFPaKA99+HYRXlWEW+EQUkMuUoWbms + CVD9tnLihexWEEs7m+im7j2X9nj4+z9I8tca8GJkgFWLvRQYllnhx++TDJPf/Bb0ycUFmxZ/olse + FxAOSs4Q/61XBTkPTCcFbabj64uXwJq09Q/+4bf6mFAxoW2xkTIyFrlx7KkY3WjZkEHqE9njTxsf + 8TlDO38j6fSMWnLgLBHt+IdVFxyLX3Ha9/zPXxNrzKxqL1fLdOQc5JnIXP4B6/OVZGjHG/LHx8jn + rc3opM5ecP/Wj2IdjoMKPFs1iVdwz4Ri+fv7t/5vyB8SmmfXF9zjK0CR/AIL59UZarir/bfe/vBN + R3/5Ndrz4zzmpwVeq/N+oveqgMlO3ABaSugSXJ+OgFraXYf7/BG74ze6/p4YQqH7fbBvT5W23lbk + watZhERv5Tjhq484w2rYb6V4R2WxfZ6yyXwxhCQdkqvL+pcaolTtvti2blyyor0ppRlOdxL+2gQI + rieU4I6MI9757SAYdVPCD7S04DvVr3aNriQEw9FTcdKxBzAvm85CmLgdtu9fN1nKZe/6x7tuED6d + FUxQEwL0x8eUR6kMQrRACX6SX45P3xMB8ylsL1Dq00tw+Ptedth+EBz4nJzqZaJbUDkiuNSvbe/Z + o2icUgAWtsdKwYHHqu00HFsVvldw2ytMfDLWrQzR7fA7E6O/eS13huII30eXBDTdjGFrxq8Ed74z + X198DqbNB7y0nQcys0v1AKMZHiVY86+JBMC/a4S+USl9ylyb4afjteUw9A3Y8yPx7HoaFob1ZPCX + X8to5rS1qb46HLX1hd3D4+AuFsAQbP3vjP3p+tbGX/3hwSuf00AUvU5b72rFopCtDaKB7aFNUq9s + MHUvMdb0r+TOH16PQXT6tcS4FrO7+kMQSrUcQhwctBasqX9YJJfTKuyGbg2W42RBeOe4FruYjoCA + YWqgUr0SnHndXVuqYXwA/M25oH2Z12GSDC2AshTX+46Xz0ALDE3YvvIWq8kqux0XGjZ8jG46Az13 + NTaswwuccVmT/ObxlGaNpQNNeqAZHB5Pd+V07wLvkWJgrxaqYfvAUwnp4uTEl2Bd0BGZGxwl60xO + bnWlPBtFDmzeP3ve/PcbrF8tfkEjeZyJ0Xo62CgEnvTHH08HZQHz986aUAqyGwlLRhj+xhee77OI + jRMfUcqXPgPfP0bFtj34rnB79svffBDnM5wpWx60B/x7XuotZUtjN1SRTGUX6zA6t9tFyljYOmU3 + c3fJL7g/vTh8RoJ1316L8SblIQh+vxzbL/s3bBd7fUGAo/GfntnCBy/B39g9AymLU3ez0C2H8XYB + wfEa7Wcys+SH+KvokcjJOncbBdeGEnmNJLJYni5HqzWho32eZM+PgBq8nKLjMQT44c9GwbEMNuH7 + ikqiPa41XTnnVMHTY5VJRo+1O32rAsJ6gw1+fOpLstLmNkN2a2Vy0p8/upkHYUZWZ353fkJ3h/yo + ApJcCmKary1ZW3h5QHJ+JrNgLi9Kx2icYUsSPeBbjkt+d1mG0FZPD6wdGXFYplaaYcnCEFuX0h1+ + vFX3f+t7Bu+T7O74n0MDZVFw8zbSUnWIIHgm8IaNj8XSqf56HaQuqmaYSDihRnLad2W2OdEtpaOz + md578BePZsb2xebi6wWmbhr/44fU0L86CI37DytfTmt3PHfgYzylgYC1b7s6kRDAPjuJ//jHdFcr + HpQsE87Snk+mku9y6M7mCeN5vrib/lJTVKrxj5gm/LiTfxht8Ic/OOIsOh7smkHip5YIfuYvd32K + xxgo8GsRz5k+dHbFxES3ThzJ4xlE7SL96h4yD7YlKquFw5++hHizn7jY19N8EMiGatGUsT6itt3z + 9wjZ7uFiq7UMd2mKwwamWsoD6vonjRfWwwyncQPkbzwWiEAFztoTYk+xh3aJkDFDdWEVoqHSGbj1 + 3ldQ1pglWBlaA3J8ZDx8dFlEbrve5Oiy5Si5urcA8S9vWOPRsCW+s44z9Olb26TnawSGMw/EHCui + baJ576HcDgsxvFzRNk4ZS7D7C3/vP6wHob5AQkaMsy59ATpJfYlyFdUz6LS9S6XVmpJVcGNAhXOl + reK9yKXUmi1y8p1X8ec/gPjelNg5Kidt57cSND9pgJWTueuVXg/RY9vMgLkmSrvVFWnAV1gU4t2f + A91YdSlBXrVfokUpHdZ4cy6AIAADQMJfsV7BawRilbyCgyTU7cwfbQnaZZ5j4+Auxbj7D5Cqrxj7 + kWQNlPnQAHnDNcIW6B/t6tl4hm40noPtmoKERk0ZwKf6NLGzt0feXLHQwTEz+1l8KJW2Nac2RdQM + 7uT0iAGYSte7wD++oYNSHdbPYveQKbwMR6qhJGyUYxvODZ0C/m352lbxUY7sveSW32FVrK/9xMUe + P+SmPx26PDR2/tPbM7vr0+2d3kZIK3ogOLon7tDbzgVydVeQUzgdhnem1BCVB36dN3VeKWFtVkc2 + md7/5m+pW5sB21t2ZrT7GxuzhCEKAz/BeGwA2OYMbbAiZr/zNTlZDvYXQnXhFawsqQqWTQEy0F+Z + FPAVERNyi2wT/uXPM6eb2nZfqwA6Plvge2iqLlfngwgMMaTYul6ChKZK38C03SDGux9B1Xe6QAA9 + TJLxVhb/+OKOb8Q3a6Pl4vv4kKIxPZHHzzBa7pA2C1IDOyd3C12S9eoYD5SeATunXjC3xL4vNuzQ + pSc4t71hQ6+z+Of/BWs8OID1L18GoTnisOyafLEIUf6C6/El4XM6vekKXDuDQVg8giMNpZ0PVBUY + mndEXD4SBmosmgMtvAk40LC5n3mrKgTiMCGp1qZ09x9GaL0+F6xWxyv4jofKQ698TP/7/DNcRoSZ + KiFF4c10XD9PCe75mpwauxl4+xbw8FNmGn4sXugKD0/ZYHyvSnxpwDZMvcuYwFjlO3kKALXL8FQa + JNLYwca1CNzFttQM/cQhxeZtdt111qMHUNnS+/MvB3b/P0D4oSNBvVWAl7/yD5LDDRNdueYDrZ7F + D+z+YQCMsRlI8/JYuPtR2Ep70xXsMePhc1EComo9207hdHOAvXpH7Of+mIwd/kIA6kNClHhw6GIX + fQiDSFqJPj3Xdh3ehgPP+gETzXvKQGgst4Tnp+fiGN+GYqnX+geVp3kgjvpswNPqkh4djzGY1/oA + 6fZaExP67OWLPbv2W+EilxXMBTvHT0vk22HQQxmaZxZg4zvoGhU1tMCZjWTyOF3CgX7CNw8v3pZg + 54Wmdp4fEgPTdoH4KSZ7D7MpspHJRC9i7PO1je7XhB4OS3I3cOCuLGOZ4I9vvr39nua+rAI0QNbH + 3jsv6GKL3x7ufImcmG+537yRQERGUw+afbzW+qu/YL0xzczu/uPu/6lQrKJXQILrGUxpc/7BZzmq + ZNezLv8WwhCGivAM0qymyUhWK4cyksZ5QL9bu9RpFf/FP9Fb6awtEy75v+/f9U0EFjl7VMc7rldi + 7Xi/hg/xAS8sQERlUVJsJWvutwDPzQxHeqe9GhAV7H4xvj/eisYy8NNAkZfOxGYvz4SqdpFDtXvr + xNLimFLc9h7QVpJis1MI6A92DfcelzaxGl9NuNwLLzBdf/r8p3fJ5+2OYNWSU8Cpd7LjVciC4Rio + 5A9/FzNcxT8+Sx7DcXDnk+r0UOZOGlam7EE5aXmHsMCiRqL6UNIt0eUcJm+5xqH8EQqCsXqB5Xx9 + BPZDqdx9vDtwmZgOK9l2LkZv6hyoGvmKXaEl7uhNswN3PRocXodqoBczmoEbYhX7n61vVzHYZukX + BgMJ1LOW/P7Wiy+7C/a/NZOQ5oF/YB/PWTyhV0utUxzCp7gIJDshvV3+/u8Jbyo2vgkuNhXmFTxv + 9ZWUPjXcZQh7VrKDOibBgy//8kWAHNb2//kz9MBhERzvgRmwrRwXg6WddeiWHU+stU6L6bC+bMi4 + Apr5eTwlVNZ+ATw0453kzyzS+D++azFsjDWlH8Ha94cH6M2LTHCZWtofvoDcwcI/vbs9a1rBfafi + 7j+plD2GZgdL33WJf0u+LU1DxYP802mwv8fHkt8zCQLxpQdsnPd0/eNTf/4oW3se4G5t2UAfmTFR + HmXdjtM5luGcvmSc/m5ssvxu3gWGAU7+8fkF5hD+8Rtcaumbbsi/zRKuBo/ogj7RUQZn86ierXXm + Eu2WUGjaPfCVr/YvP237JnzoBmQmBqeb7obOXICC28ki2CnfdPHpkEJ+ufABKiyifQPjm8HvqKwB + XfYTIzMnjvCKnSP2VndrpzADMRzmg0FOym1qv7v/Cq4Dnon5hU9AdWbg4fvNv3Y/9lSsxgU1ELvs + FOx8iBL+iC/wuNTxH94l4+4fgb/6g9kpmNLidWMglx++5J8/WML2hcR6EbBm3g/t+hh5CerVRojN + ZHLC/fl7ux7BmuhQdw4vRQYc7f3Ezv01DPM9qHu0v09Qj/xC1+14lmAe8t+gBdvDXeki5eDqD11w + MJy5mMrZEaGbcD7R36963z5LGvCSihSrQ3ahu78loRF0fbAfIByo3FUbitTcJ8onYgtqSWsD9nw+ + s0MiuOtfvYygI8ThGf2S7c8vHJpPFEgPvtW2HPEqRKgP8d1vNjClvrChMQMqCTy2afd6hgpY9iti + Qwjlgj94/ij1nyoh2YhQ0fnGhwW+DW7YvWXfpHdc0EH/WTnEkL+/ZEsK7wUPSuUQZf7xCZE5pfzH + l9TCUV1uK8tKMttZC447X1rX8eRBzC0/4pqdUfRPcHOkg9I4ePfD6Zp7WQr+1pPMp077z6/484cU + HhUuXXVHgoLCnwK2OLy1gfhb9pdfZ8kdf8nyGZ6ppJ5USNyxUbX1YDo63P0AbChAoWstTDFY2WcZ + LIejCdbaDGSYnxmLGJVpAUHUuOUvvnDAnBSXbXX6QvibccRYO1br26SM4e96V/b6YkTn4TD3ULvF + Pxxs8nlYD6Zqwo48q7/63ECrd8bCKPxIM5gv1sCmL1T98Yng9h3ugHqfvAJX6XUihVXZbs+K2gPc + 0yEMnp4LhsFzIhvtfhEJrhxud37IwOR6upFA0Vy68qBaYN1PbxK8iltCzepSoi8fzwGSoJJwTfU1 + //Qh0RuTDOuf36MdXgVxOfJt16rFORzPoYA99/MbFl0/NECKpjdWr+OiPXRLhvATqhg7pyrUhM9b + G9Fc8BeiXcW9fghJBRJVATPqxDf49r1QwrY5htg9S9dC8DjqQctzxoAPBMElb3bJECEzxmrjkoKM + 0WtEvx6mWFVY4g4ouaew1IuGGM95A18pGUTwteduZsRnSumo6SooYdD9+SHt6zNcU1iKXUHk8/s8 + LMVv6sCe37EH1a7d9d0oPk+kJH94tsZyqwPbiiFR6VC0W0jLBXS97+x+5q2YzYa/QAcfEuKyXEfH + Hd8BfiVXrHsLHKaZgAb+rRf7E3yGv/oZ9I0kmKW1/2ibx4QhsvnLFeuBymqzbtkQXi2xxY6pMu6o + nBcPFlqDSXASmZZGvKSDaLycsHfN6mIU2JsNMvAsg2b3EzbJDXh4aqUn9nd+vx3jTAWK4os4OSkB + pTtfRUiSDVL4B2vX888U7PUQvNfHk7nJdBmW4qvAumm9tHWuYx4F5waS4J6P2uKgmEe191uI9oxe + A3+aXvNf/ZC4QKndlQf9Ap6f5EXU9t6B6U9f/R87CtD/3lGAnl1GzMJUNfbpiSO8LD8JB4i/FdR0 + tQcMX/GDaOx7SOgQxAGqh1DDWcobdBNCbUb+8fointCSgTT1y4G5IT5Jcl0CjcO+rkMreWyzKD0F + unnM4EFqNAh7J8YqaC7iDD6fsU8U7TIBauoVi/rM84mvHcOC+NDwoPy8h/ihpXKxfPSzAy1o05ns + Z08IzTIbpglPgmPDy+2W388N5DXiYkd/xmAznMyBYCbtfgZzoBvpOgax8u2K74bgDgvTHzs4bVlE + rMIb2pVxAgeG7eOE5eglF9z3RQJw6BBPTiH5aYQpDjmsxnc3MxGntMtHvzvwMLwCcgdvXHDbR9Zh + Hx1crBeZVggL8vczX9mZxPM7TDZxvDnQkJ5forGZqQl5pHWIEqHHGs79hIP3xkTye0xImgf3YtPc + /IV+HDuRSwoUuiB1CcDmPsSZGxaN8mBdL2hS+JaozE91N/7nxMBLnSdWa85whWP35ZFixgoxblgF + wySMEiy+DymAbZS16xadQxR5zDOQprBzt97dK8bM+MQ3bvUK1h9XBhE9VXEhhBXYNDd+oTISz8S8 + kQtYpaIIpMlZXRKUn4hSoy9S0DvGg2Tu6ThQ0C8jkpk8Jfr3+BrYmvQbetWvNw7GYw8GGa4jglMm + k+z6lJIlMz0drjVsiGoymsbTV2MiZcNX7LC4GNjn0gbQc8v7vL6hTNdVz2TIALUK+HhRCnZ62xfE + 1JVDtPb4HWh66BsIcRSR22UyXMrbfgXUdT4R6+50dBnG6b/xl32cuiVnb2IhP2t8wL/FuKDaQX6h + /ssKJHsWyzCHTqfD9tNUAZKTwl2O9+NfF/4Ia9Kl1ngPRCISLS4l+T27tZsUxy/0YObnzFdW2dKJ + TTpkVIcE2zpzKwTefbASMLWGaL9HOvDPIvyhbeQBwe5Fdnk9QgHknGdBtINzcTdPtDvY/74euQXr + TOeCj37omDkFcWigaLQMTw4A/CecBbHPhi3l8Q9G574myW840tUu1RiCOwjmVeKsdlO8s4xkIUX4 + IehfbalCVUXDcSDEe5c3jXfVgwyVGapEnTeuHe6XdkFVYoXEztmKTt3TUpGzdPoskMGiwv2GdPgX + L8Z2DgArvC8h+BilhbP994TtuxEZvfoj/gNdB6HIDw38FY1CNAYtdFFYRUeb2oT4fLryBZVMu4e2 + znPYGh5quy7xdUZnrbOxoR7CgRxEnUfXo/TFJ99nio3zIQsHJrQwHhTBnd7wIqF8FTiswqNX8OVN + WeAXLD3J9edGV3o9SHB4nrXguI/ncj3UGRw+wRLQx/nmzqwVe+g99F+iXKLbIIhtWIHc0B0c/GpN + Y6vL2INSfwPi5txc0MG1Jci0Doux8f7tXXQSHn5L7kWUWydrbOGOMTzmnzwQLpui9S0nQ2h5RT5z + UeINyzzDGMpd0weHbCoK+rpnOUISuWE7/VbtKKe/C2zmeJkPERhcNs2bB3JC+YLdj0s0chtUB/0m + e5i5qIroyp7EAPLt44WvVnpwV+IiE36u+orP+mIOgrScN6hf9CNJoi1xucrJR2jrakV0+Bm0+Zwd + NvhqMcZ4upVgQculB4pY3ojznvNim+Spk+q7m2Ibhj4VavGWQfcWG0Rz/V5b08wrwfQ2hBkdtrSd + fmhIIcO9KVa0RSzGwMs2FH+6+A8PEhLVegCFcUxIqaqRJjBrE6DlsJyw3KRD+xqNbwf3fEkUal5a + 2qJ9R9X7edvXDzvsz1ugy+5dTo0GAPI5sQ/IlzeP6Ev0S2j/TTwEgwckPuefwKJ98AbB/RgQ9eca + VHD5SQWn1YhxvgpnwF6/zOsoH+l+b7oUa2vwSyBsY4tibfxy2nrokhfa5x+nIDy0m/f+mFJeYXc+ + X6PPsC5DwsNt8CKs5LIzLJdbpaKHZeXkRLy62OZbPYLUtSdyDo+qtvHp9QGrOAT4GrKdtsyusgAm + 9RfiVyAGywsML4m9ljO556dmWOuuspES3xnyvAlKstzePx7StFTwpQ9Fl/bfwoMi89OxLuYG5RiT + ZdH5ru9ncJjGpZoaMeDNpgN2uFvdjv09CoH00z1SePMVcBlfLYgE+oGYddQMnHYdGziDn4zlpg4B + Z/IyC7ckfeOndGzbv/hAt2S08C1YA8BSvr2gkkNMMLaROPT+eISgiN4Bvgy8n2wvPxWR6O/4hM9r + st7KxkNOP17JWd2GZJs0XgLksirYld9lMeWtx8LBul/xWW/rhJM5nkX3At7weVhaSkWnzODkUJeo + OG3BZk6yh4wKJfh2bg/u2j0tGTohP5A75ZlisR59iD7ig8We0OKBFwSrgey65sS71H7CT5csg+h4 + 1MlDeNBhOyh+BUYTnkjqD9eCd46aito+5UihEDVZv/VZgjsezxxB373ztuYgLnHv2NHHzGXRcvnB + v/FOKhO7vNfcN1hDM8P32s7c5fy6ZOg9/L7EqfaK+Lk4L1D9qtZMoy3RCPuQNqhPdogL87K5NPe3 + GAL3gQOIYZMMf3yqjl8qybmbMghh/u1QvnIcuez4M/WKYCMB344ktKMH3bCVLuiy3nHwJXTVSPOI + eVicSB289/FctsUekdbqHjkvr5u29SJTwjufGDP7fPft7O33eMGe9lhNz49i/cnsDCwauERGb1Hb + nkX22/fMX4lfpprG679jCm9mMQWV9Qvo+AuvEPrXQz2jwB+TzT2HFRKP+IpNzcjdjdbhAsXPawoY + /FNcLhwbFhoH8YKNlp2GtdIWHdUnUQ4ASX97hYov4emydDi07qd25w86VB1LI37qWi3bSbUDB+t5 + JX6+bMX2Pp0cVN/Rl9jPYCg2rzkv4KmcnthiuqTdjryhowC/OlJ8T07BTxojAneY9Ln5UNnlikMe + wuH8MrFqMq1LbOckwyFEM1Yt104ICjYHtKsqYHt4ee6ia+cQ5vJzw8pBHelQcbYJ/U4PsX19SsUm + G1IOY0mLsPLaJLACpfzBdvEa4gXvq8v/1ihH0dT4RP0qBzo2x3KGWYrP8/hW55aqz5qHZ6FwiHk7 + e8XMp88H2MeD3Gtb1Ma29Hiw83Ginco44S8rlQDDfejM8qrsrvQqSNDVhCxgfnXrboBTe8BtZk28 + Pb5W5xE5aOev88/UtGTxFzeGsujGxN71APeVchnqn6kl58j8FP0jvW/Apa+VlMFb0NZmSXWUyGEa + MKu534ri1yV82XqAjc0+DVNkLxcwExcETHMQkpVOvxE2+pIHUzfa7Riz+5mnZVoCpB3DhFzeNfxb + 77PoTb9is+iPh85HVHGic+rAJkp3AXv+xiZ7SwbavtADoj7MieU/D8M0PW8qrFzcYjn0d0e5v/0g + RY76T49MlDUuIL9vLlGXdhj+1gt80+aCrR2/qXSKGHhLZuuPz7Qb/YD8Hz7hIYAJySPtBd45n2Jn + 0mTA2dnCwLnDycxCdkkmt25KwLD0QeTzagzs3acO5PrOISezCDQB710rh4O0BuL1mSeUfZ8z6CrZ + jnfDjU6Td6jg4M83IpvmI1mlx6uBDbEMYkSmVQi500OYugc14MZrQtfX1kAYFl2JtSpJEyr60yjt + 3xdIkSrSLXioMhJvT4bYR9HW1vjMvSRHuir/9M9SBOgFxkHZMK5KrxBK9Omg0FtgBrQmYDZDaYF/ + fG/Xk+6251Op36pjsGalpI2JMqdHeK5S4q7xGyz13QqlWRDQbI73e7IVEMbwas0ROVveGdBDpMgQ + GNgnrpxd9/mJflCRhXTXO422KXhOQcPDDYe4M1w+NvMLnKjXYX/nG6u+VD2yqOeSIgy19jfSPoTR + VPn4tr4+7viH71FvVcT9ZGyyLq9jgAaRy7HVNmOxGM/fDOWu6kmiCp9k0aHUAcUMlWDd+RytSo2B + nhduAS/zi7tehMGD+nJ7EwXSBiyrn/3g4Ek6sSVjSkZVf47At04ecWepcumlpyFc1pydpQMzJmv+ + pM1xG1mA3Vt0B3N8an8wzBeLxEoet4u7GjPsrM+BOJX0LlaxvPfgaQf8Hz8etmRMRmAO2nVePfNY + bFh3S9iunBfkXtBSusiXB9yolMyMEM3Jqj+wA4KvjnZ9BYbxT+8dK/5O/vjyHNRFCmxy4DBGuq7R + Oax1pF/MIzZww7ffZF5HiPo4x7IyUjCfSq6DO98jJ7ubByL60wxPP00i7oSeCSkp/SG5NgCR9/ka + pOW+wEsI6h3/oz++m8HIcxxsyl+H0kf1nFHifsEsTo2VcEYSLvDWoQIrx0hwvwaaRMA3PcZnZ1Za + AZ3LDQLIxsTZ8Y/FF32EP7+IsfUhUUs5pZHQT+vPszi6pbtly7GErMHrf/xnWN+BWEqN0RHi6PHU + dslXzZHGte78i7tTQdUyrKRsmgYcvBY/YXM7ZwB2nvK8Sn1TLGh59JCH9ptkn6tJdz0gQsFjJGyr + NKPjVz7lMH6Ob2KEx8alm7imyPgE59l8VmO7Tp7QwKtddSTOTdYlvz5c4HqLK6KeTtrA38tClzTt + Qne8ZZLx5s0zLB21n49mnxTUPfIp+r6mCLvfC2434fQ1YYPPI/El5QFoFsDxD3+Je5dnQMygXMBX + P7VYOx/EYqULlIEyuId5nWzQTnl3ceDBvz+x/LDM4quyMAfLt7Vw4JhLu/n5MiP2e5yCaP/98oWL + isSHOBDPu/gaqThZh7f3ZcXus2S0/fsquPN9gmXlSOnVsBf01d1d1V98jd7m4gHay+iS5ODwbl9+ + jiOwTrDA5SGWW06IvyOMn/MbG86sDIPpug/ojbJLsu0kuOuL3HqoM5tEfMo/ik0cIxtazHvDvqmH + lHv+gAin4XIkMtEvdNvb3IIdrwkeD1sxf2dZhxio84y041Jsa8+xf3olWFYsFH98SGpPtzUQgu97 + mF4k6tGaHx0sr/hacIk4/f6bL6qELb7K+fyT4uYgBaPuR8MezyP407fuGhuUT+1vAB3rkmL7/svp + Yr9VEdZG0mM5/ex3gRpaCl3arRgfpDuYl9PThO/xwcxaO76K/qSE2R/fDJbfz0vm56OG//i0p3V1 + u5DolYH6rSjkvuMn7czWRuHMLX/8J1miLOlQwzNbQGGttJwuuLGk+7kwS9l6ArQ3qhnG9hBgjWfP + xWKOHwalp0YLYBZrBXuWFE+SDeeNo8kuWvqL3wxsuIuNfZfr2j/9gqLDwQu25dIV/+KRFeUSPzZB + A7wD0xTdnsZM7HgwksmCQwm0UmXxX34c/W/VIyZ/bBgb2Vvb/YwedAesz8e74VA+dGYTosjXyDO2 + w4I/Z8Ii7fiFHaEfhoXPIhU9zZcXMGG0aouqP2eYuAMgJjyf9vGvg3/62oqUIZmT5PpC1SanOOOG + 5k9vdX/5ElvKiQF/+g52vjwSZV/vPK2zDTjAvWD7ducGAkmx36JRzzs/9dvpmC4v5PmGhXGum+0Y + wkCXgso5B8yxDChfVwsLuW/+nFF+atqx3jYdXq5OG6xvWNElJ48F4m+mE5tKI6DRfZPQFbTJDA+x + PHSbssno5vUE3759CFjg3GNobkedeI47DtMRsBkq2+aA1QaTdr0rdQBn8ylhe7uUCac/sA0tRO/Y + j/O1HWf3tMDHOTgR75L22o8GhxDWaFJJWLYcnWvXVGHelT4OVjPQOEmofuh3+Pywt2432ge40RF/ + gn7A5gFKloUtH+AzM3eiooSn871MTFjdLjE2z7bl7hXYGFGvSwMA9Yu23vu3DKPgcyI4NYdh/dOr + qnPMiPWzhGTXzzLa9ToOwJCB7eWXEuTisz0r/l1v5/x+roDegWDe8WDgCbwtgHCzMat83Q273osl + oOfiXzy5W1AXF/jnZ+ZtewPbIT/9/vnBt1z6Fds5Oyxw12Pzxs8DndW+eUCjpj0x1HenTbufIuU3 + lxLjpSTatimSDDLuk2Gv1ZNkkeT6ApX4yQTyWDQJmda9J5L1PmA9KSd3vHZTCHvcuNi4l727JV8n + g33k8OQP/4Q5/Jrwj3+rL9kYSPYReTgpahCAj4vdv/yGmvujIE7cfZNdL/RwVK9r8OLxu9hgWpfo + 1p3MAAzbg7K/E0ql/f2I7g/X5CucqQz//IDbl+PdFZ+kDX5LDROf+7AFOZtuIElOSvFz97NeFPlQ + GkQhn2fHXIblMUUinF/dF58aXh6+aZBVsDv4OnbeNbfzRa4DEZM2JHsqpjY+fnYHq4FGeOeT7vDc + Tygt/bwFYMejLnW3Bh4VSyfWu/q0axrV7D9/fx3nVSP2BCq45xuinp+vYq0sJYPhLCyB9IffrK80 + 0KTT7U9faBsEUwiWUF6xKZfvYZzecgp3/3c+NL6q0R1fkc0H4sxYIHTn+DT84FVNXsE2nmqN3/0Z + +CHbFrzYgQXL7i9DI3SYf3x/2WgsA8GnmMgvo253/8qDu3++10Nwy//FZzj5JsFpstDtL1/s+Q77 + Xsm3C8f1IpgL6RMcuo/XstnpnMFxOz6J0ltKwR7sLIeK+LgRW3b0hGrqDYJk0pigfyqmuw5V8kAb + GsJgZe+quyrZDKHAvmQcaZeJbvfnK4Ns5SxBn7My2D4XMwPqr1/xWc0qMEqftwMt5rMF09OBYGnO + v+CPLxHXPI106R1hQ4z6MYlqZX679cE3hd57PQUPq34P9MB8Svh5OHawze+lIOmXG6EwX9DMm+yg + 7f5TAFkFj8S50Hn457/cW9YkfnaA7SBRYkLlql+wv99dSYQf1x2zs/IjZq9tdOydwwY7L7hiX7nX + GrVXnYfim9GwX6attjW/1wJP0fVItDuYh4m7oBIqV3HAaYItKpwSyMB7wdyIC35DstS8lktEuJCA + 2/XmmN7BApK6v+K4wbj9DcylR77lekTp066ddKl8wcFOxoDMrUqn/Aka4LI3k/z5zUsTZ6L0Zi/D + fkLs5a6tV3ewISDHJ45Z20nc5Af6atcggKq6ahSMqoru3GMhzipwYElrQ4euGsyBuM8PJ5b3H9z1 + Mn4Kr7u2Jcm1g7Jhv7EP3iSheL8F4/a0ZnwKmxfYOtxuiKkbZwZ05jT6lx8GJrawtfJ9sWbHzQQC + 7/3I9X71NLpBp4f7+Wy88yWNNtclgFbc37GMsphup63KkLO8dKwvuuHST6eI/+oPxu8mDm9lbXXJ + Y+4O0e7zt1hEoOh/70tM3vkWdL2cYvDj+CmgbNa5v/n2HUFsepjoMhjd4U/vnC5bt/vpx/bD18oP + sIo/Yr95asnCvKvdTzjFxG1WvRX+9C7tzIH408C4sx7tt0BZ8n8AAAD//6Rdt9qzvBK8IIqXYCOp + xIBJAoQNxrgjmeREEqCr/x/7a093aiiUZmd3Z6XtiDVRt50UXI/Q/4DzLL1gzBbndVrg53w+z7Tc + ZcMG5kpBSRPr3/i7AnP10DvkBGrsn54xzr7x2Ay+eCfkmZohPb7DCGZOK859kdwGKs2GDG/eXSdu + PG/D3AMvhp4RS/OePjEQ1yb0oSfuHtRyZdouceW8YdJEOr3whs7YPdneoPC/hQrRpxrKX77tXF8D + ap85XvvycwovtC6oktg4+1jGNf/pVT6/yRygp66oIE6NlJiPUA1FGdIZJMy5zcD5KuLJXtZhyE4H + 4hbGLlxjGTbQ6Y5/hKQnDH56LVTbV+rLjzkN6Zj4Frw3dUNc+tqH20/fuBz97atHPrVV3XkjbK1H + To87eQ/W+3x04fe8Es9jByB8upcLbV/e+3+N12gs+qsqJM4HkaQJRzV2cR88krvhTq3aNQbm3i4W + nLFxpgpIry1LT7iDfQw+//wjdjE0H875DIktNar21Uv4f/lA9jtfvS3EsKH7dL5Xosik5HKPYVki + 46dnMql+/O1gL35Uaj7CJly/+Jefka6T6NhkbJo/LgcB6l9Uh/ySjVWgKqgO0yv55g/wLC6dDIMu + KGdZeuNhJfa2weVvs+e1fKXZIqqzC396sRfhVzsu0yrC52OeiWNoWrYtKszhaG4lNbSjrK1WsuPA + L7/J7zxlkJT9OwHrnvfn+atfS4+XG8PbmFh+LDp2uPIyhYCc44ng+lUOn/HJGfCnN91xe8xWo33F + 4Ocf4NsJgZpNzQh/8dVx6fZ4su7hE+lRSehxN/HtJIq5D3/5IiErXSA6f9MbiFknUffrrzPp1QU/ + fpubbz5r+vHF/1FRIPD/u6QgLySdHkzNA5v2+D4Lki4Xkv+5h4yXMHcG/iPYqLoRzARJ5jbYWVtC + Sd3+tZSU4Pu4fefQU2zF4XYt3o3ob3pBSzHwNOlIoxxeGmjS686CLUvPloUqc0fo4TFMIb1exg54 + vFjOM6srthrSMstRtwz+eZc6eOGNuw/Eh1QQOyIbXoxD1MEgUSNqxusCGOILC9xlbBLvMIrZClsv + hboS9STd839sjhK7BPeq3ahdmXfQ3813ivL+TclFvnSDcE/JDO5QPRB95Cq2AOskojK5qv72HCNG + zcMoQ/UojrPYWiqWumsmw1o7RcT0bglmzyk1oFFFmCjR7jrw2bKvAHeaferI2hULNWMzpL2mUlsH + 53CUkx0HNYFJlLwQzxbW0h42TxiS03tn4c9fx/XwN17tZKityI5xh5wlkGhQ8ly25uQMQWaJ/Xf9 + 7y0vJwuHTNO2qX1AM1ge4jGARa5WVPsTnt/ylsJHq2/+zdAL1Ez0PS1GXLCtRDnDpd0yrIv7w3UI + qJapdrY88/4NU2A2xDm+NCwyW08RVwJxZni2B+nI1AC90vr9W7+WZYg5ULrpFr1+zjhsfuNlgH/T + u0RbbRUWVUFTzRB19vwdiF3Wq//m46fXPVgxqVwUpvRATfd5ZmsBjhC2i175O+cjss33tEhmdp2S + MtjXA7+fHxt0TQtTAzzakIlo4EB+gJLfd6QZ2KuqNlQ+nInYk3MAj0Nln+FVX1xqJEjH0vXolXDp + 7zG12gsJxYnzXbBd5JVGC7Xb9absEshP7o1k7veuyWkBOSy6+Ub1njXDeqjsAG6V0JCDcnRa9ueE + KSqyo0byvGs0Hn8GC/Yjf6XulogD48p3BLfLbiVG+BjwyqmmA4Jq9EiGJpzx0daN0M+fe6puZYtF + QOMUujsgUxvvL4PEwDajhBBGHc2bhi1oDxUcO0uhZdXstDXc3EQ+JvWNqkF6YmPzakZYcotKfdWm + bOGNqw+pKE700Ew2E2Rz38CQj6+zUOancC0LoUKKX2vkXh58bX1xq4jMEPL0pi6RJtxPuouqMb/Q + VD77IbV2dQTDG+tmYD9njQnOJUaCkNlU4WytFSwjjGVSPyK/l3Y7wObrjZPFh1CQwNRVJol8laP+ + eJfn1+EcYl7C4hmCG3eip/exYWLdwABGXKqQu+ZI2TrPJPkW3uakUPYnbXU+qoh0krnkkDSYsZlw + MaRjKPvrIOKMX2ougff5vCc+zIZsdAOlQfByjWcJdH4meZbgQnUceqprat1uB67p0WuBH2I0+579 + 9gde34lArn+KEvJJy/GwyW/VP3ytzri8IdqrF6qz8jIwRysdYIAzmJfzdMska1hUZJ6P8rwLHgVg + 8S1N4bk/T4SogjLMuXqIUSzBmmbN2RjYmu8VKLbPh78fixzP2sNSoeP2no+e8/AN8QQRxdJBJ6f3 + JwDsHdg93H+Kly8a+eeHxzMEnvMkx+GxMcqonaLmqYjEjZ2xnZr7cwEan+ckLLzXwOJvo92jXx2J + ecNG1ktrCuHXvvoskHC2cKHfQSGnAXU55ODtceV95OBxIQ4s3EHs2SmHp4nD1J6cGqxim8zgfJ9k + ovdMbQVhcZS9xpc5PfB/k8bs6tnBn/2RNldol+TVvaHHwGEWeMcYpEw+xghXFBPbb9RsvST1hu51 + QWhZeXUovDerQV3UP/2tnddwmY2+AYuSa/RsVq9w+Z0fb0Pfort7ipku7ox9hw8BuZNTnq2VpcxI + VnFFTC6UhzHvLBfZj3Ch2k7S2vl07lP4tw632QBmxKTtr6qg9UY+iU3iA4m/Gj1s27Pgi/zXZa08 + cYbOIzZIjoDXstRPI3h33gFJpcgBo4I2FTV/PZ63UNQGEd4+T9Bk+pEUSn8Da5PeDPC+9TU9rq6G + +c45BzAhHqN2NmiMvW5HB81AepIjepdMuEQAQvttmNQ+3hS2ueweAFlXUnqviwpPQCl2QCRKSfEj + cHA36OsZpfa0+jyGGEhrvipIyKeAYlTy2vi8mAqKUrIjX7wOwqJ7PUQSzoj55ef5piQ6HNfLhxxr + XAwC884VrPZnhZq3CuE11iMeNodyIebuibFkh0IAGy9CJH6ESsir4RKgfpDlf9+FfR05qKsqnhpN + z9p1PIMGvLvHmdy7TGvF5tjykAuWlbrmUdfo8ZlAOaAbpeZVsTQWF4qPktd+JEpdYLwU2oFDIDle + STKWJ034Wywdhe/SJt/1A2uvtBZ4t4Pgcy/ctRQNtzc8G2+d3IokHKT4iWd4d/qA3sS3rK3eywiQ + EY8CtX09BLy49DG0Dx+T3hdd1tZgySxQH5ScBgBXeFZlXvzeqnZoZjgQb9dYl1EvrRZNyTEL2bjA + pyy+xis1aWFnY+qrKnoG+oEWeevhRcqLFGanpiWWPi14rZTJgNB7yuQijRJYy8fNQmJXqbS8vqeW + jXyfwKl9S7MwHHQm6bs792+/vJMifvmEOPAd5y6xIyUFbBc2PpxiyaUH7vpg2ykOfNSL0kDNR90M + 03UhPPCVcaSX88UNpav70VFBfMHfpQtgw4fiGHTGvvSFO4lb9qdKC2Jn79t49O+RbS67BtCQr6XP + evPEtrV3evksOsCXpykF2wdGHNRKuKPX7/6PS7LJqHhq/Fx7/qhtXh0EqOof2e9/tqAyUOGgDTH9 + 2aNVBYUDpoamBLvsDbYb7F1oq11HbicBDts+iEagyuk2A6038VrBPx9sf43hswpjvA3xQYUgMa9E + MYShZXtN06GmGcPcG6kVitXwt8F59znRH7///FXU0Vc8f8YC4g3QMoGFf0qIya02ZuBmB6hJrTeN + +g5nIsejHLRw5/z8VbA0y+wD4VOBWURLPSz3wbRgrsqR39fqC1DMWQnkhFgiOoM6Xh+5E/3Dj32q + joC5hWqBizw2JNxnp2H8EzhDDvnoSv3D4wCEU/sRwZd/ZhDjwyDKzHXBdN0+87549Wzzvo0mN4dT + 6FHjV0ap9m1jouJqXpJWAVQ47SDw1DqYd5XFs7eExQCWbs7TcHe7aFLqpzE0z6ZM/aMZZpt42Cwo + s/pE8ONeZ1vavXLA740XNc3RCvm8U3xUsOXb+PN51yb60GT4xRvFIBQY7ZtU/WePk79tGaZ0Obnw + PwAAAP//LJ3J2rI6EIQviAWjJCwZFBkTBVTcET9UQJTBBMjVnwf/s3TjInZ3Vb1gWubBcZ13tTOv + 8wpCpnsUlEo2SOP7M8EgrwjStvBNplO1a6FTWmStV3uYwtssGbXo21R4mu0wZrp3hqEnMDRL0RuM + iAURPPr5jpGXtfAhVMblp+/YQX8nIEVeo0OndAhGsK8GdmIYwa157llk6m3Ic+ivl6jlCAmWrtXT + 04qLnx9nHsoeIXuU1AOrv2N25wZkliA/w94lF2w2/Ysvmyx4wEb3fGZfM7i+JD65xlS+PYaz7pRy + b5hMyJlt4N98lkrl0BpipxkYcWchi5iK0ZpKMPOCQhz4gexaIKgvxvBGt1JVqAUTJkMUMLTW23La + 9xnMYhPjnXd91RwP4wixmBHm8HtfT16zdaGXBAcW1dWOKJjpit6dRAf/8tq3DWsE1nmOg6ccO7KV + v3p4gE2BS+XeEEqQ/oGhUcksMtdF5kS5V/DovinDw7Xly7OSEqC0k463yjCRBfNHDw36VzErRmM4 + naq4gcJxRMxUj+Kw1PNZgbFdHyhc8+QkpG770zscrZ+XggzCP/+WbbTB4S9XWuBIbnu2r/dfh+p8 + i2B8z3Rmx/El5Ip1lYBkd2/mSt/7sHg7but52xzxX0vDUNJffQKGMvKw93Y157XqlVGckMWs+4OF + H6PoI5i/wYjj66urF/HufcB8mxp2VdQ5HVc/YfgtjpEmKx2YjfPNhRVKCiq92z1nzYVounv5JOg1 + R44jHyQTQoUoPlI+9jhMSj3b8Kgad2wthwfg9t/WhS/lNrBYZHo6KlOfgU6vfaQtmkt+8xUK40dg + Zr5b0ln8NCX0SXLDZimVhJd709WlMYEMq4+Qz+b0KcDqhxha8xA9hQfP+OMLRDdfOpAJMTuC8bYI + 0Ou01YbZHJfG0OfDlc757IRLqw8UytJ9xJb0FOqlho9kXQRr423516bDTx+Xs4RYebmeCZu+uQfX + PICvmakDTtMmAOcvodhrqz7klmBnRrOJRoZgbw/L3dEbvd9nEosVO+B0qJYzSK3dyCIgtvUSqbEL + EScas0VfcdjGcVzwbZqM3U7gRKZhwyYYXFoBx0A4DnMSMvfHC3BUMdNRUvmqwyJsBxzQ0E9/fuDf + fI2vkslVYnBPB+Q4YlTIHpFbfDWh2pibFTEp9XRAoQCff4qL1HW+SMC7SpAGfzJ2T5NPZmWt1Mt2 + ifDP3655OjGCc/fEDqcWmKTjo4eptR+xI17ydJbkC4XeR0TY+hTX+t88iSWpZNmT+ID+9ETsdIPW + RVMAjqsuAlHiQmxV3w7MSjI3RvMsj8yOBJFPpxcvjEVoOlZmEeEciXMPj175YUH5WcjyPAcCWPKw + w04AmnRaxAMyODMNZgsvz+H7XTLCjxC9kWwvMvl+WtGFtRjaaAlwWv/L4/NE0FrPb8C2yuQa+33o + IxE/0kESZWELOYMFM33T5zL7zNTA575Gm40sAY65mYFerg4sVOcm/T57DRljPIoYfc8Vnx06SEC/ + jQldVn++2LokgdMJiWjdWJmqXIs0kE39hqpCqtf0ZvuZsVEjhnG+e5PhfVsEEMJXzpxtogIeH3IJ + PhLVxMHqr+QHVCNIxr8Ub13z5UwqHrdAkqDMCtJTMP2lz61RV6cc72++z5cX0uhPj/7Pd9/w/k9/ + GLbbLh3WfA6saLwzUxV2hJ7C3ANL7nd04fYxnH37W4Gwana/fkznP/u5rsFZBvoazqCepGBnwtHc + nPB9qrxBtjJVgT5fr+X3rq+BzdcQ/fIVts/SPlxAW2hwKJGHf/U/zpHVQq5vQrzdaKEz33ABoece + bEzmtiXTrfEQEEsY4eP1NP/jTT/9R8Zn05GJTJsKfsLNlbn38RB+5fEWwUC8XJn7Wv8iML4/C1x5 + GksKy+KLtacRtNK6pIc882u5MczW6JKFMqc/+47y94UZjNTmzu5OcEk59HoPfgT0xrtB6sM5En3b + WPkWUutzQzgueQ680a9xmfUB+T48bwRVaysYH4eZs0drTcbk9DlzKwLJ0vNrCW88++Bz5e4Gaa8d + TSMI6cSsa/NJuaVnD2jseI8etB1r6tbhDcpbK8cXYfad5ZdXn/z7YbtnkXHOdKv88QtWbutdyicl + /cDlOYk4I5a0np+J4EstPMoybay5VOUu5Nk2ZqRZQL30rO6NIksa+qAXOaWJ/m6M+dritT99QMtI + cOFYPPp/fn16Sjf7931Idg41kH+8iqYhxdF+1zjcOSSL0XhTjrfHrOGzkuMDvN3AGwmhIpJRcz8N + 0KhbIXiJOofp3aMA3ZeVzLnIdPhOe1DCQxS4SN2jDZ9n3t7gGzcqMm5LQxbz/hCM/CYABO5ZHY73 + 13wzGtptMDGgPfzjWy86YtSCV03GfChu8LZcQwply0nl7PlNfvqB0eUT1yq4Wgf4nqwIoxZ+0+nX + /yf/PLPAFs7pHObDCL/Vt2A3aYmc3zyA5LA+Ylz9zuhjMYDSF12xdwOFM8XSXYM4dmKqO7eG//ie + Lg2XO4sO6MXn59vOIaFZjMa7enOm6akUxjktCI5vyzblQrE8gDh3V+asvGJ+voP893vjvSTbqUTd + /gG7lIbYX/PAKAThemkUSpAGLc/p1vNW977t4DiQVTA3htfA4RPfVv1xa84H0kLuvQ0k9L4Xtont + JMaPB4br+clKsmnh51kYVHSCC1GZ7t/gyndYgLNkmM9e0IObrWUUnsApHftYWiCsQottH9cv+Zyv + wRa6z6HFoataXD1avQCjjXzBnvxO6lnfzw9oVWhd89TA9Hsv9iPc1W6EfYh7slzZpMGU9CUVd3zr + qFraR5DsQcxCh2DeFdMVgW+oE8q72gHqdegnGF72Mn2mwcK/5L5sYXkhL/o0Mwt8dqfu8c/vOqUZ + cfVlbCJYRA+Pnentycc/79JDQc5UFndgx1f+28PDxVXZzq0Fhw3BJQGarsX4WEErnFW7KHV5Ti8U + mrrrTNdLXMHKrNOVh2shP1+THFqB/ELGGd4AVfZY0Y+qeGe+HbX/eJa++ju8SzYFWdJ53MKV32JH + 3jzWefuqwPuWVyx0+rezDORMgV4aAWqFuuXsl4dX/8biSzEQHupFCxTjpdHJx89w7o5BAtQ8GLB9 + rdN02Vy0DByzTEHAwDfAEQwVmMbgjF0JKnzxGzhCZSoOCKbTxeGlOBbgIwZbpGnSN53H8/MMYnV/ + Zm7TFoO68mZo5e2Xxd1I6zGWLjp83rSOpQd1SKfkclBgWLU75smKz9XnESpAFYwOKU+VpvTzR5KN + 71QWAmmVgc/T68yNThaZSrv5y6eN4ucQhmebmac9ApK7swtDIZLP7s4V1VRrnh8ol23wOy9n3sWv + FuJLY7Kta+5CJY+t3DiwiVEBUSeVjMHPIHgpd+Z19t3hz+Mqln/9Fzty/wELCz0BHvd1iLJdeiez + L8NAL8P68u/5zLJT8+LH53783KFjQqIfz0OTDj98miO/gUydApwwWSHfsU76X70ivdo2Kf8+TqUu + 8kfK3Mi//vLCx7iDbcPiQzzxRn9VCfzCM0PvYOOs+WHXAPdMZYbC56emL6SNYO0XvFt59aifQg/Q + 4C7TzWN/59Nf2m11omw7CpKHRpaBlCNc8xe2hMsOTMFCe7jlYsrQmu9mr/gcYKRtdKSK5iOdZMuv + gCqIHQ4IuQx8zc8/3vR/v0XbcoJfQBLst2I98PpxyUFr3kXavIEwzKejG8BTllZIifLDsBQNK6FV + RQ6CAk34TGcDwQNsC6qa9OV8p7BLYJ/GNto8pNfAFesowb70F7bqJaCfQyfAMPK3OK9PLJ27o50Y + pw3frXkoDqnOIwR46k34svZbv+YnkBejw0giuFzRRfmhn3V8xdvehWDwtOf5p7fscNKlla8d4PoK + SoTjvz1wZo3GNzA3tUkNpo717MuSZyjGW8Pemt/H/nTIjSIhPXOj9xv8/CZ8bqOZhbOnAm4/Nvq/ + +eH9Bce0Ti9PD3qJd2A/vzYLG6cB/wEAAP//pF3J2qqwsn0gB9JJkiF9LwFBxBmgoiDSB8jT349/ + n+E9ozPc3+7+kMqqtVYlVe21JNjOul8+KoHyhMatvxBZ0vtqaw73D9A/DvFP8sxXNMvVFnykd0Tw + dJ3oH/8SawM8iVklWs7qYZT+8QksGTfkLvExUf70Lj4LhRQxq2c3Iv5Ohl/lSpdT/NmvDJp9vevN + d77rnQCxey9JB+sg2q6SPwNzmDCWRRDR7c/fWfKMzk9ZTAd6Ol59pJBCxgbzk1w6OK8LbAZlxRJf + JoCpUSlCqdId4hqtU3HuoZ1PuF47kjP0rU48djfwp6+kVL9Ea8BYEHqRdCfuK1bdNX1a3J9/5y97 + fYRpt5aBUGROe/0oqYbLqi5wr/dh90JwxQJgObDpzRi7rT8Na6QMNWTC2MO2jHxKUiiLiEe9j9Uy + 25v4pgwHpZsHsXWs7hH1Pi9ORLydY+3dyyqTdw8BmpbR/NVT6TIuYwD/+KYtaqO7vAZsARQ/8b96 + LH8LD/9+HuLO7QcsyTUOoJ6lEb4Uy6wuWmO1cMdPIhPVUJdMTkcI8mj0N1cRwKwFVEMim7nY5Vft + H7/9x5/MJv7RTeRECfanVtn97hJQvUkPcJWEG5YjJKhs3Ekikj72RPysCVVu96egYA6fmWMgB/qG + QwIctgXj3LtmOTHlWhDLCyv5y1oH7noNfQt+T500C+VtdLfTbUnQn570ueDsbpZdcqhyakj+/IPN + ia+1uPtDf/kc0PITQZidNhOb9/LhbnciCPDuOjfihLOlst+0KKH77lts2NnkruNrZMTxPB/9iu0t + sCFbef75kTseH1VCwTbC6vqe/Dlj24hWQ8vBQaspufNGW5Hme3iKez7Bvp44w0IHN4AHAx/8pLI+ + 7jgfDQ692fpDkvR1HWamNQzI37Dvcy/wyxcrt2pY+fNM1ArSajx1gQFlZFHiTGntti0Y4//pSgH7 + X64UhN4NS/BWAjLVyoIOvclio3S7aK76kwUNqyyxpex9bD7nUwYlcX/Hi9Pa3X7DMqP3ywtIYjxs + df3keQ9ZpymI//QfKuUbMMLYlh9YUS1EyVmRY/Rwle/MRqw8rP6n96BTGovPN/IXTORyPohec1iw + XK0Xyn5kCUK2zgZ8vr0Wl1bXYwC19OwSWcssl3ksFQNRqTvYv6exu0pjfoDBSsxZbLQX2IoCpNBT + QU/OafAG8+GbH2DHzAdfdJ9GzieaVUC3jDt/feoSZQ7CpUHw9FB8ZjPfw2SYio8k5zL5NGTFYXEM + /wO1Tc0w9m/LQB/lIQbsr96wZyl3l+kbKYCdlmhYn6p98qEfbFDpD5C426Wtlu5ORdgdnMu/X68u + CC8I+7j3md5e6aSXgwEPdPqR+GlldNWNNEWBKQgkexIFMMeTl4JsmkWi0HCNVgDNDW0JaxLZYbSc + poK9QMOdf9h/2vrASHnboDa+tASP+6SnNCFPeHKJ7LPV4RatPzZQED8EA5GQowxsE/gMOFRvB5+D + y8Vl5OKSIYTnfbmjDNj1lDAwGZsrVvpUdtf+wXFwHNqMPDP3CpbX+mhhpoojOX+2AoxKx7SoXYMD + yUjzAaxj2Rs6CScG25rGDasoShq6m+qNWHYiVczxpKUQXFsLv/r3I1oox3+giF0ZW2uTAza7F/tk + 5iUhZl5/1NlxMxGSyNzm1t2mPR6FGb1xNROJY9FAO2gpkKsLk+BFlartrmQQ4mzUSa72ssrmR1qj + jhkPJEqa0aVaFouQinGFU0E9DuuRfDXUOLc38ZYzBGNR34oTNyjbLDabo3KDcfdR9KWifyi+W7Ta + 776F3K33iV9tQT6eumWB6DQCrNxMbVg9CzSQr0cJF3nguxwBnCM+4vZGUjldwFIf+Sc8yRCTGFA6 + rHysGyg784DogqTvk7xwAVSmd8j+/dWllyIFihffmo9zoVNmUbz0VGjpk7wWtdz7MF8zpM9jSLQr + dqqFaVcPtYGoYb/Hsbvk9gWipvdLfAevY7Vi1MzoFnkPctNhUDGv9dqi7PF5EAsrl2EbZt+Cm6P6 + /poPWrT9rb9/vj2SJG8z2jsGzJC3ZIvYtT2AdbvkM5TeKodt5ZGDziXcCH/85YIlcWlcnmtsAbxP + V0Q0R2Zz8ilLCZrK28Oh0HzoVN4FCSqyM+ObaLnRv/1d0fmKjbuUR4vaBQ3a8Wgm9tsc2AdlIJLK + McVh+j4P7Ck+iBBx8Y2EvYtz7q5cDrC9Pcn8u5/PLmG/gQbGoc/8+Z1UdKNXaYFFz1k4i97naikc + b4G0Ek8+70yKy75oIMDnFc/EtI83QP5+/kOvs77R8SBfB7x8kKhnmOjk6OcsP6YaPEeMPqOzTumS + ztsTcYO0EfW6qGo90yVAZ0b6kHOHekBfZXaBjsd7s1imZ0ocVjXQXWWDmf/6mrv6i/eE2D/32DAe + tsse4rIGE9QW/7bGW0VvmnhBnMNccab5b3cTELpA31l7fGEU1WWfzLxBc6tbbLb5JyfV9XgBC966 + uQlQ7C4C9/YQeP1uxH26qrryY2qgTjy4xE8/rsuYQlIA69Nr2Goi4tKQbwJxEkYP38/Dqm54W0vU + ZIpKLHc7q7zNEgsij2/mYyts1bLlVIKv9jNh3ArbbhlcLgipGoNfSnIaxvch4tDEfnTyqMqfyuiP + +wG0k3An+vvT0+Wdzg08WaXoi7OQVkskVgn6UNkld4+13JUEi4EORsngeH9ysb1D5QMv/iL4p9p2 + Ac1CoUDac+WIt4CfOuVH2qArF40zgiVSN7aKObjjP86WwnP5sNcU9LU3Z+Zl41ft673AVDw/cdGa + ftRf5TyF1zEucH5bm2orlHJGRjujmU68NaybPlxA0JpXfxkjpVryauoh0FFDzpt0cRlP75/AODo/ + 4h4NOeffU/oEg5Dy+FyVpsrHWl3Cy4GJSbp/T45erQXYY0vI4+SW7uIItYf0j8aRRwiKneK5HLRx + e8e6H/o5I/mLhVyjzPz6L5/cpduIWqflcGRBJ2eaY54BWs4bds+x6vJ/eHn5SAVxc7XImT0+oQiv + zDy3jZ1zh+3gAeMa33FsMDeXST+KAmn/qHFQYC1a740H4WmGNTmfXMllInFIoNwT1a/oU1UZW49n + eK9mjJ++bubrV6lriPkxx5qyErqF96BFa1OcSR685YjTnTyBD2DIf/wBjM3XmiHZ4IozLXlR/nS+ + MKi2yxCbblmB/rRqDPT7tiNXyPnVMuSvEi546fB1Tqphz18civTvg2Tts8jnHN8ttMfHfNjzF4fv + cIZe1HgkN7cjmGpZ3EBbv3Si1oHvMiZv1TB0E4Gonm0N7O3ub/DYIIc82njvs99oB3i3Rcmfjf0+ + ITyXDawq40f06sBHs4juPmyk9IKL29mg43UeZ8CDh4XvQ9LRyQX3AN1OvUkMbgbR+vBWBs133Zi5 + N8O4P6WJPn/xiLUL67vswzdnGG8fH6vCLRroY0wFcC06kYQOU0drq4gJcorZ9ZvqcQCbkyILhiCL + SfTgwopxnvcSTVa/zMfIUlRWMpr5Hx+DQ+xFyyMJLbSeuBV7eqW721stA9gCecE4ZeJoid/KE6la + M2OZGbJ8CU9rCS/gOBCMttN/1n/O0oU42Rqp862tA5TFuo2N6nGglGtsEfTMgSderAbq1o2fEuIb + W2B5IJ9h2yqsgW0ZEmy8mdilU+0s4Df1JdkHtVTzvh4R9+LHFwxYu7Q42QlMy980l1cTVRsleQHP + N8nDOtYCddnkMkH9cNVJUvvLMIan0wdt0SAR5yOnA3fNGw1l4ofOzWMdKa1AXSKOaWtyv3Yl2Jzv + t4Y9qD5EAfitUre9tOj8aHisfU5evuGzdwGvT1T6UXE26KqUj/gPT8g5vt6j9uVkJXSDV49tkFnu + Hx6AnPFkXBTMe5jDe9rD9X7c5tls1YGym5iA86Pm/dOOb1zGfp4wk8YvsS+FUnHb3PVQ1qK7L+q8 + C7aI9SGUs7LBxi81cypkagAvn5tI9u8Nli0HEqTo05PHVa0qGoTYhwdKfkQWGgUwHePX8BdLhNjZ + q6OTNecGSMvvhG2PtVS27dATQi/A2CXCqNJThgyw803isBfWJYsbjUB+nBzyWschat6h8wEJ2QqC + tYVz5+9J9/7idabKyxhYA/xq+Ie/dAm/7n5eP1ATNOePHw9bNZoSktFiE+ftc4AUJzlBWntvsT7Y + es4f5A7C7888zPS6VO7CFtKG+FzzcXRdVJeXv90Mv7S+43zHs6VEHwV9I5IRCd4kQIf1doFKtryw + YgRaxTqf4nmaXkZGNIPh1elx7y8w0MoYS3xl0nVQJQM+wBf6DV/o1VrL4gJlJmbnPf7UcZLKDfqh + ImCVG97R9nq+JLEJaoKjeQoHduxZBYLjysx/fJqi9qv94Rux0Qm666+IPCiH+gUrreypzOvmQGim + yx3bzLlXp3nvE31Ybio+m8+7OtdhGkCIW0r0z3Ws+mrEEoht9bFbko9hPtVFChav+WBr0Mrh95mV + C7SL4oKDUEsivo67FLyh1+AHp56ipb0rGmpubjbz8U1T6UcUZ/hsA4ZkYFyj7ZrPBgR3aZyXnhCX + OjKtT9dEx9i8Ol+wFAnyoCGAjGiy9XXX00RH8Kc/bJC17j8+9ROLO3Yfj5Supf94QvmWvHzAW2HF + vvPjKDK35DQz51AEvblsGeDku0z0U7fkM/8RNlj80GuWwq3c+flkiH/5PplYd89fkoLQSxx9NjJv + OTXPVoowP+ezcC2ISlrHSOFXdDVf+JZTtFwt/glT2lJSdMN7GPlHnSHDHX8+4+UrWKdh5sD4so/z + ergwYMGXtgXZ7zHhXX+4Y3cHwt958vlHwOQ0fOQN5A/d/ghy3lzC09GCbHj9ElMEZ3VWb4QTevnD + z7zwkgbq5R8P/OE9TuVLNQW1zsB6ce7YHpE28O8478HZ8SIi32McLSSfArBOkvCPL9DsdRkR9w4k + bJ7FMho+ZSvB09BI2K1bg9IvkAqEPLYh2DRdd1UMpQXuCDUcfq9GxGicyIHm7DvYXPeSyZ8ejdn5 + ShT78QFd4o4luBaDiE0dLsMSlswIF+N6xfakQJcmz9yCz2enYD0Hgft3vqBLpzM29/Utl3fzgZeB + +kRSajdaEvFkwHkbBRJoBxNQmCsM4tPzlxTgquZrfgQ1HF2aYz8nnTviS9kffrFCiH/kv9XiGMYH + yvwH4j9+uYDH5vzjX9ccLCpdvnMK3oWy91GdD/mqTMcaTiNesfI7aMO77n77XITzis/jzcz5qVY2 + +JB/4XzSNglwl0wpwJ6/sTTBKSf5yBzg14MhMdG1G+iv+l1gArcN694E1LUJfA7qV63EbnKOIsrT + 2oKkeBsk8Ps7nV/dqYb92kRY+lrxQLizrv3L//f7aLkstGIPrVzfE+N1LdUtzSmH2rf2Jtn32EZ/ + /Byef35HTLK83MUTJAdedXHD9krPw673NyjdtMZnd76zecfYB8oZNhjf+0dEi9L7wFgpNuKKV0Lp + dHgUsLzEEPu+/ouIJcsbApenjDUr7Yelb6QLGP1W+4//EN4eGTxwTeAf1td92F4P24LDvOrE3vnK + 4r1YB13vUTJv8ouq60sCvniLKoCt+4uL1kKaniCyngY2snoe5ijbr4zfsTif+JtEx+96shCdkjOW + n3oJpoO5WPCTSCxJx2UD9KjkC/SZS+5XKeZVOqKqBdMl+OBXzQ4Rrd4hB+37tyOB87sMVJHmAIpH + 70Hcimfd/qbwCoxG2cSX3HHp4mpCC8GlkMmTzT7q2OxdW1KjkIi2VH5Emc0qIc+ID38psJYzMn5D + 2GZshB32clW3DLM+jF3R8I+1Yle8fGQy0Ib3I9a772XXs7IIW81wCf7TMzmRfBS8Pz3WlcO7WjzB + sv7p/yDcpGrBpSKiqusuWF1CXV2bzm0BzwgPfO/b4zBFqyZBTvta2H/6yKUnqU+BfB4nvPOfYczt + ywFedMmYT5b4qkaTl2p0nLwbeRL94Q6cVPTwfbohYokZU9EtDQX4VrJ0Zk2mr5aB9h/wxwfd7WJV + vKe+PLHzDY74jtCp22V/smPM0MHe/adG05/fdO8D0RdJo4CtGrEC2bdi4is5ztHGwrCFm3QpZjKC + O/3jZ2A6Xnj/CCIdsBUTFHDIpyMOzRENO/4XECoMIObuNyzodV3QZVh9rG1iFZFhsdM/vwcXeltU + FP5AAJymy4l/AUFEH0bVQ5/5noijJa/df7q2//DBXN4LWFguL6ASXt35gOdNpcxhOaAsixOi16Rz + +7k89bBOEmfmUqDSpTurxZ8e9EWzTAaqHtME2tU5x2qUNMPPZwMO3c9BPUMkXwf+wgUZnD7PI3bj + uq1GSX000LyQDstLS1X6it7Mf/yNRPGiJTQ+CvSMLfLT+yuJaPbKZsAxfU3OX+8UtWNmt2Kfhkds + Dd4x2lK5s+CfX2AJqw22Byv4sE5iB/uGeIradzrXcBmX459/M2xo8zJIroecaF4egmU6hx760wN/ + +mN52uUI2TodZih/rxH7xydCkMY4TS6CuuZUteBz5l/k7/utAxZK+P6iOymMpKFLS4vl9LdfGof6 + atVnPUH+7aJjS5XzanKvtYS6g3UhkuefI/aPz+zxQwpLOanU8JMADl8lwlr9ueXzlfmISE8fGtbo + 4kVLdTz3//SSdJCGYX2dqxbteDmvZ1+i/ABPDPCz/Qo1YrSB294nHz6Ec+rTh9RW49/3cKDQ4sv9 + LbvcNQt8cffL/KM0zWBSolrkKjimRNHGMF/RZxZh6cMbia3UGdhfkftiV5dvjK+/xOU/ojhCq/Du + 2PqWU/7lTlYJwvjQEy+QWkD/8seLjRsSosKl2/shQkiKysCGGW1qPbePnX82d6yoXKtSH1cNSOxk + mzt0gup8Ovw0ED6nAitWL7tcSKsY5Uz1mo/226y48f0YwastJ5Ls/tJc9lUG8+DwJJ5qWwMX3oMe + 7dM/sb4oLJ1+3FcACI/YR2kgg7mNphH62dMj5iOII659WYFYzBImYfvR/vkTpzbjo5kh09dd2vOv + /OPjxDfEe9Qtl+MCdn6L85c8qsvOF2BdCCzRy1EGGz6+D394vvM5AWzJT3yCnY/Nh7bpInJ/lgLs + 5tdCzKi55OvBXBywJbzp/4aJB3M/FQnY/XGfD+5MvsYry4BJmD0ik62ju56GUPUVxa9FxwIjDEEJ + dv02c1HcuZsE1kJcq1/tr5mgDIv3Qhb4Vz/A/RTRQw5bcCVEmsmhrN31LU8K1C9n/h8/p/rvaMGd + L2BzmG5g2fU4LPMlIMaOBzTu7wtM4LIRm+ViwF63Xw+Jcrru+WKO6LQAHyyq6xPFiQa6nB5WDEvP + nGcwYzni53VS4OsTlj67KCwgyL+kQK2PrY92f2edwsKCS1Jt/tG/LdXG2fQA93wyM/5mgIUwHwP9 + 7d+fftz9cgMejW2aEXt40+XjtBdQR4/3zHbfzV2GRjTA4bTciPQ9PKo9X2TgT0+exn2OxMgUMWBq + Z/NRyOk5kwGhB5+RIl8MbhwgXZJtYqIsJgnj6kEXcLk16E/Pe1sd0lUEDwjt7uJiSb2d1SlGaQpt + jHi/1Fe72vnEAjMKrT997JLOEj4wAE5OsEUf+Sbb+yyqpmOJdKDnfJmkdoPdbWx3ffrcRxtYHPgm + tUpeL3+pFid8a/CbgLefdHwe0bdZeVBxGoVI/vcTTV13T6Gv+Uei/QibL0+7nUHdlD+88wdKGuuz + wR1fiLbe+eiPP51cDsQ+05amuphPd4OdCF182esZ9bdJavS+2yU5F1Hnjns+gs2U/bByqJ/umh6/ + DHQ4XpgZh3/m2zQfIeBurU+0p5WB9bqRHjoPtZvRvl9bjkMHkS4+4rPmVVUZRaWPgvrkY+10+7mL + lJc1utFMnNufLbjbX33q2ywlDuRZiVhg9hk8JFNITH6e6ag/wgMcQnTGRTfIw86Ptn/6wlKJTtmJ + lzNxH65DtMbohybrsSB2WvchknDfryQmZwmysmthDbeYMgedEeFvVXU/iPIhottW12jn0+QZrfec + HnKmF2cB6ru/2qpb8xoT6Pd9R8znKuVMOosFlOMXN8+f41xtj8A5AC3Frj9Nju1uf3pNqbPQ5wJz + cNd31jFgr1f4R1VTwc6HHIiSVzoLjCRESxS1Pjyfs4z4XDCoi8WLKXS5U0xk3J/zVY2f47963+4/ + 5XNv/gy40sD448cqt+tFeFOvT/JXf6PsNzBgtmwyNm3bBxu6nUbIybmMlZHf+VxvitCJAh6n67uO + lusbBqJ1pJnPTEngbn/4epz8G9ZkS1dXq1w4hKow9LnYufz5oT7Qwrb0hWuB1X96SqnTkMSb+a5G + BsMM3CZ7xFZySdXl+FxHpA98S9S9nkh3v/evvkpkzmsiOq9fCeZyBOe/ekR3TWMI//xaUyGXiDXF + cP7nP3uuh0G382XAp/hLnCf50HXU8Qaf80f2T7RIoz9/Ef4Qh7DqphxdWBzE8Fu+I+yyL4MuhaMt + 0JPeBrn37av6O1+oPgEXu024DmtAylR8HJ5frIRVGPXesfCgVM4pOd9egbpimDJQ6h8OvuqPa8U9 + 1KyESyWdSZgJn4qunl4iPRTv/vau23xrTrwE39cNEmP0pIjb4wU6Xanh265ft7IfUkgiffM5p1zy + 7RvdDiD7vSafdfmQsg9v5VAdvd7+sXkmKnOZZA0Kh5NF5IHJ1K3SXQHuehebuBrUkfgMB17pMBMz + uMf5Xq8e0e5n//GnfDtLhiE69BHj86+a8+kkjg58+fOTKKba5OPqp4v49z2F4+EHFucZltCNkgif + k9HPt9F0JVhC2yFKyR3UVS8HDajvn+ZfwmrNtyn+XoCFsxL7ZXsa6CvqODBbQvePL9LJXjT0tRfH + X5TazW9jlR6gs0RngiNQAZKdphTep5eMbf+5UTKvkwSq16zMvRQJwzyUgwhHRemIMmpRvsqTvkHv + HB+wZyQG4KD7TcBe7yWyL38qGj/eMXiyxkQcpT8PI7fPtfpfrhRw//+VAo15t8R6aGtEOoIa8dZc + W6z+yp+7AE6rYSaY9ey1t6SiTGxwkC5PjB0tO6t8we6j6dvbQnxb0wEb4MpAbBNY5JJwVsVhVymg + 5toacbYhpCt7rUrU3aCEjegKwGoLUYE+TGERrRDew0SPkoFgVgDi1b2orvoQMuJPKyscfK8bWGRw + T+GHeVrEEzIzpy8DlODAkAfRRLq562FvMXMan+N8yuzWnayuaKF8HGOSqy8hX4d0naGGxmrmDynj + Tq+zrKHzbQr8dUg0Som730pWwGM+OGtCN7e/G3B5rz3RecECFGaBgSYcbzh6K1213owjAxxbMoji + nwPAKbXJAbM4Sfgp3+yc46JDLf4E5M6ftDrRJXufStA/o3XeiQdlmFtQQATfJlGHc1VtnfTOEBcw + IrmKSVNtgXGvkZ0UP5IeHz1YzesIRfxVamxTC+XtT9MbZPPuZxYO4zPfTveoh+UGGGzNXZ3zwbXs + 0bzpInG7y6AukfM+oIqfP0SeLzGl+enEiKzV1iS+wgysCKwC+Pv/7TL+5DzHoQTGY32fj8PTUGfS + 8gHsfMPGxjpfh9XNFQk+soQSm2PvlIgqLyFBL74kDb3GHfOALZF/EyLyXK/HfLGadEbkNrr4MXJG + znzvW4GwFwMSyLhVaWppJXTOUCbaI6F0pS/9AqHb2USR67vLWMZDgBPYLJ/bby0u7TFu4dnZFKLc + fod8ub71BDF66hDtUM4DVedZAveXHxHPtz/Remu2FsbNM9jXK1Ia2I8S6nIRkrR1G7BpxiED+8+P + jT6GAz2mFwG1240QL0pll7eduQSDiJl50WBdTdH5K6IDMz3wObvytJMSFsLchhYpgp/kzoly8+D0 + dPZXEvZEv9+CS2FjhzmOhFcxMOGPZvCCQ52Ekhy47D08b5A+BEKSx3kbaGEtM8oe5TYzb3XJZyZU + RJBZNU+CpbAiOtfPFhx/AcQWAm3F7/sJPAADgr/rIZ8rOzmgcXnL2FslP1qztJRQYQU/HD6nSV2g + zcSoPV5u+D72R7rqR/uD1M+jII/wXlTs3Nce4m+WQfA8nYahsJYR1UHqEt+TWLp6SveE81H5EGtL + JMr+flqMUCKEJOBEHfD9gcnAFKgzltbzwV1O7lqi2zufsFybGeByh+yPKh8xViq0DX/rgePw1fHV + 02XA6rfegr/0EuC0fvbRevLZJzoPT0Tk/bwSu38nyLEVY99vf2B4uLVo3x98foWRyx3OcIbC+cET + ScaWy9RbLgFHP8ZEESHNu9hiFRhJ7W0G8kul/atiRpSD6wH7e7ySMI1j9Lc+7xWM+bJZYQuQfEiI + 5c9tvozaaYF6HpvzShjeXaxMb9A3IZBIbqa4//Bvj4eZ1QaLcuEkQOg4J+yDfOL3Yu5FQF1FpBmY + HgOWpj1dxO3StcS5+tFAZ8FI4N/6vCkK6aB+gho+uQuHTS+7AH7twChydfObj/N0r7Z3djmgY/pw + sK6SKV8Gu+/ht390uLDPpsvHycsB2lUB2LGTNqd/5/vpCRfi3nGWL6PQ1MiAnk7+8G2ZSkNBjniH + RJmbo0sQOyZAsWsXm+nl4E6qNSTQrfeHnq+PD7bP7LTQ8vKJSLfLOVqSoniC6PytiSI+ZZV3dNRy + +3nCkoqLgT5q7YDCM2CwcWxGd6uFtIEP42uQ2+P4pbw6NxKQ1eBI8uM1iPj6N3LgQYKZGBmJ6SIr + BwVadpIQjd7v6sIMkgMDawnx9WOZbvPi6ww25cKRa1DFKhNvcoy4bF78+veewFplygzr/iITucC4 + 2rTILsWTfTrjyPy6Ls9xbIIGZzKx3JuGS/9GnZdvacBR4O8l8rNg/ctn95euRvSw0CcangXjM1mm + gvbTng0onXILO2/jCdaLISfIwx+RqO7FHPiuah34jhqHOPEiR9s2xzH6lj+K/dx0VZZMNxFiwLZE + P36/+Xo5HSyoiwwzb+Q2DNRKTi30DZmQQHkagN1u5Qjf3trjSBXmgQpTXsPfxzHnn4U1dU0/9oLG + THBwcmw8d6SY88Dve37ic3gvhim13z7yQSuRtH46OW/mnxKF5xNDvIiLcmYrRR+is5Th6+mbuaw7 + r6Ow4y8O1kNfrbwnL+h17O5Y/7LqwJNhEeGWtVfifU8pXS6qOYvm8fnCIVHlfKVHSUMJRBo2aXio + 6BvdRxh2PSbWJa9p14qMAt8HB2B8zyWwbdnYwnuPInyTo0fEJsNQgpWyHA4MeFJ79FxGVBHpRALN + FVx6rkAA5iQWsfOri5yJ7+WGzuHPJHEkOYAfzzRAoNN74jO1PfDRL7Jg/OoW8uRpQVfIvFp4WZkv + iU6B7TK9n5dQE7uVmKciBd0bww3KsPuQIPUtylFFM+B+Hv3FTzZ1X5/xLx/cX3qVT7yQKbCXkxo7 + RTsO48E8WnCLPx72fhZfLXznQHTRzjK5bvBcsRGrB7DdroTI7HimpIDfD7oJvoeNcctyvoDTB1SI + G3b8twbmECsBcOYekqt08KKhvIoJmlopJ+c2Rfn6xsyCslvakZv7vEWcbIEaDvfPad7MsKTLc6YW + JJclmKnwKvZ2QbRF09OqcTyCfPjL59D2moU4hLzB9njbCsrB7eB3IPqAVX4GHvRaryLJVBmU6T+q + /4f3xIiuOZj0lzGiTO0hxs3HoZwYfxaRdSfgd5ymqRusuhpqW//CfsHAfN1u7Qz335/Z41ePtlBu + Z/H3TO8kX47msNyzWwBl9XLEOsGhumnNae8acH0Ru0dztXwNZxP1UVYIjhFPKTTaTLTQciApRj+V + UEH8oMvKfYl6eP0A36nfGJ2Lzp1HITOj9XccBfjIYkou3KoN6x7v0Eh+DMGdZFXruxoLWBw3GXsH + oQbLbE5PeK33LhvWlETLuCoKeH3Xaj+/erRVyyEG5zI7YeuhhfkUPt8iYtWxwy/O3KptuS0H5O+D + RdQDytS1yPMZVm7oYWX9fMF6UkoJMHrmEOdb/ygtr1sMg81i8XkQezpUpRVAQVA0LCesTJn17JRi + 9vhsO8plLn8qXjXI7YNFbmpkVfzzFRvQaIt0x593vjkfjoG9HNdzc2BfVffZqgvc+fw+VNuv2L98 + GYPMxd5j7tVFokcL5L6ozpx2pO4AtqUHP7t4kqxox6p/WuUTXjQsE3XHUyq/fh6kXJ1i7G9ooNj3 + U3jkjJm43+88EPlFfMjDBfpCfPoMy4PWEvpFyXUWEuudU9VvRTiNhwTb7+rt0hO8pOgZkf0V2CUA + HE+sEvy8N4t1IcnzzhbyJ4Qk43ymMFmXSsvg/ds/aaBBNBUKddCRqtCfynJTtwG3HgQbes/HLUvU + KZKDJ5Si7YJ1J73T9XJfFXDyWQsr1avMx0YOBRjw2kLwy1rpJsafDYovxSXO9SipU5UpI7DL+Lnv + jwO41JucP3wmUnbswAasWwtz45QQmfcXd3rqDQd2PuR3WLlUPHhZEGpornwWqjwY0dgz0JiyyIfp + NtNOoJ0FWeCrfh5LebVKTlnCcK4dcmlPvksZxm6h4DYzdgiRwdp+8kJsGS/CZ/IzI3Y/r2iPJywJ + v70RplAy6C9f7vE90BBZCtxcpiKvnU8OL6puUCT7KxBB7vOlsbgnnDdTxFbWRMNCkjqAKXg2RN/g + NNDs0I/gzUQ6UTK9UJeN6QTxR8zYp/6xVtfb/PKhclCov1rIjVY3d6ST4U2IOHq3X0mtOAcIDLqQ + ZN6+wyikxgGuzusyi+OZieaTXkFYjhrBWfLYckpe0wZ3vPIPk7MMU7zEKXK1iBI9ksNhec7AgotQ + 2z7qU6saD7FygdDeJGL+7lFEixZxsDE8DWd7vp4PonYA19PcE+dQIXcVcqmG4/n4xnIFB7DO/egB + I/kyGF/kH5iEXGrQX3wa4ybmyxLSFB7tIZ1F1sYVvd/UBN5YO/O5+jcM660Re9hTQSXByxEq2lWl + Bf/O9zml7LDv5wddwY/65VGPKzrXSQ93PubHFbNG2x7vUGUSCStqXUfTGQklZN/JGdtj/6JTQdwU + lurdJJIk2dFWurElKrWD8Dml14pbh7L9l8/3/DtMQ/m8wD/9HVbQBROUPhwMPWnA8c6XKAtmD8Y6 + Ff13bzbuSr4ghq9TFWDXs5Z8ZbvIE/UFHojpYGFY7mJmQMX/GESRrlnei5OaoEiURBKbJzaaDlUY + o7/9Dd9uD8ZGvgsQiNDHwc4Pl4N5dFBscSPRlbOrcnEUeZBsAcLKxNrRpn+sGJTbicEF/3Qp/+Wq + Fo5vguftIStgm040gLrIMdiVAt2dTF9KoXK9utg937DaCkGfQaG+3HzmKXAubb6kBaHwlInWXLC7 + PZllRr/Jc/1lH0S4450D//SMji774C0vbaBpHvU/vVn9xncPIVfXP3LuxGO1vMu2hOXFO83FrSJ0 + We83CNU18rF85E3AyCBM0dFoZP9QH990u7DXD7quBM3gCT8uNfErg2tlgZmx2ID+/vA6ed1Kf/PP + C6DO9FCgekw1oqvkHNHzEZei34w6dgT9WE3P+/MDvV4f5/d6XHN6dvfBZnaSzMt66IfF6bwM3GdJ + xzaIFEBDbWwBChaLvD7snM/yO7b+/j52W3XMSSutLXqTq03kb4fULZyEA5I+jUZMJL7pZoP95z2X + Gk7WQI24hecMIPs3FVvmfiUmkAoJ3l9eRELmgd3t9ugsZOfpZ2YC3IE9vltwjy1M0jFhqx/2/Uz0 + R8bHjyv/rTbQXgo4ZuI+qjoK1KlxFwP1H/lFMu+bg9FX7h78Tp+cOHZiRcsYmj4UU1HBprmK9J8e + qb32QW5hFw7EZFQNHleL7nxTo7RTpxhyXfmZl1OqDmzZXmN4k+bNZ+XiWfWYz0YIT5fJF1MQ7IPP + dAUsRCj/8X2OTX/F3/ci6W96DNTEtwwofWTM5X7+mK19LvBWFy8c9w9E5wEIzz+8xHKBSTWIgeOA + 89H74+9+tbaf6AmIzv7+6eeNlXwR3G5vDjuqO7ub24cG6tnVw398+e98o+Tt6eS667+NXukG87oU + iR5uG6WmuTHoLz+Zr3erUnBhOXifFX3eZMOgXD2en+AvfwSOq+ajctA1eKufL7w3t8mXt/Sd0fP3 + 0rAa30PQDLj1YXB/eLgArhXxzvSQIJJhgg1wIsPyyN0C8OcAk9ee79fjljowo26HZcdzXPYblBD1 + uZ/NqwjG6rlS3UNXWr/9tzmPEaUjSeHWFSb2WdpG5N31EsT7oJ1Nru8qLa6GA9Vv6f/5h9XKab8P + IJZwx898v0I1qR4DWeCpxBH0V1UdFvCEBN1kooexNIzKR+CAGTIQn9lNHpid/yJTVR3/+GjI0B0P + UwOtHL92f6ACW6tfLjD2IUskdvi51Mw/Hxh6yoAtiQvBulqdCI/py/G5Ua4rypinFG7DNfCFCrs5 + 3zLzE8rbOhL5vq50OZLnB0hTa83xDAxA9a8XQPYdn32WFR7VymRFCk6PViQX5fqtaLOKG9z5znzs + pHb480cBsFKCH+EdDuu34DJQ+i+XGDbwo+ke6gvqAyshDy6gdHuc3s1p90ux2g3v6E+P/+kzfPkE + cvWPv8e39oktjH4u1TpkQb7QZHL98z+5Z7GBS8fyfhNdc7r4Sugjbrn9sJXfSvXffl2eGo/dUBuq + yZfGAP68iiUuM9Zu3UVeKsb+gcVS/B7zVapkA4X1HGO7+H3BCNd8/svfJJHeEuC9QK3RXz7Ws3lW + 128nl7B8BAjrprjfXWOzFkzLTgCCl14t/Uf14J6vsdJbHuDa9fOB8kZHLEXfBcyRVmQAl0bnn17c + XaV3Ne9hSh+neVW21V2Ph28jnrq7SLyejV3KPeMFhuPU+Jx3H1W6JpoE39qSkvxghMM//ytpHIn4 + 8s2OFm6iHAyb78PfuFWrJiN7cCDjvNwHsQQqSlzO+Dsv/mfoibr9HNcBb4/22N1GYSBnvodQoJpN + TGeSK/Zi2DGUSBUS/7vchn73R4CVWha2Ii8bFlv+KpBhRN6fy3dKiYlfqbjzwZnWxHAnX6oD5LKd + RmzZKdS/7wnR+lKJbV/fw/KD7QF+pqH1TzaNaP99LRIyTnZE1J3ftWGsjvDeaQ7+80/Gv3z+8Q4T + VsamrwachR/0NBYbP52VA2uhUAs+1wT42+73zIn/ieHt2k5ETrRrPl86y4Dyw2H/419rp8KHuNUB + sYufTrfGZySIT/BFLl8vrFZjO6R//rLPPc5bRUX+XaIV52filZ4JWEMMWgi24xvjDztHY9OuAdjz + Jb6otRZNadfE4JNwNTaPdln9Qq3uwVisEfHwUA9z0h1q6B3lH8Z8EuZje4x7OFTZMrM58ugmcF8D + XLilJC4LFMBPhWiIE8edZ6QmN3W7sI9/eDNDg3TR8OfH6KOqYDu1IaWPJdjzi0Nn5mmW+Qyg40A7 + LMeZrlygcsB69aeiPAR7/qjoFh+HAohfNtj58JDv+e7550+SP/6yTJIk/NOni/Js6Ha5phlUokIh + vqllOdXDojnt9Rv/AMp7Ti1y8SAKNosY8/atRjCXCvRt443PcvIZtr0+AXY/yj954Tr8/ftw97vI + xSXUJeaPPyDZ8kX/GcDQHc7uZYQ7H8W+qYl5j7Y2hWr6/RJZrrZoHg+jI/qosnf8MNVOH+7cH//y + hVuvDnQqRA1aD0MlxhpU+cRp5APeaan7n9dWgQ25sgL//OXsbRwo/atHWVqW+/xzAcPwu2cedLa0 + J3t9I1rw4/uER/WNd3/jka/pR97gXm/AwQt81YVCo4CMk+jErB1JZV0zi2EuH3PsgpbSUVjI5y8/ + 4YfMxHRRQ8jAV3Jc5+M4fdXWvu5N2d/7FX8xFKvJ+b2eItWkEzlHdxMwm3Vv4aedpnmBMR023U97 + GKviOlP3Yla8cvwo6OwsCnYLV3bXzt5EZNzbxhf2esNSvGAMuYATsSrITs5kdnBAUr7WOH9MasQ4 + wcD988NsJMVqd1+tBna25eOH/xiHDQ+4gSIhBrFCz9j1OO+IykGiMwSulY/MLS3gWNCI6NXtCxbz + cLqAPX9iCaSpyjv3KQGdYRYY3+aUbl/PSkD6WQ4kv34edOGv0eVPnxH5fDIqjk3JEwav73Xe/TZ3 + 4eShhmdJ8LDRxq26iL/zAuC1ksg5r7lqYW7BEw1tPGDfiu75JkpAgyzSPV/0vjndZvkmQnegZ2wL + H5kywJ98IMzO+V+9cMGoDFDRog/R3V9aUaOMUvh58hz2xCwH49mfW9DJ9ptYx4dDmaH7bfAo98Y/ + fJl/XJSBvT7mH74LP0z861TANb7Z2PirH+lMlaAA+Sd/9+cB+aSzIv7lZ3d8P4e/+ELCbJ3xhZGq + YUK462GfXF7Y8mcrZ3GcQ/hNJrjzWxJt6IA4+IK1Qq7dR1LX5Jj7kMd2SnQr/UZ7PNcwnBuH+ELY + uGtqdz4cb2yKjXnTK5pP8wh7AK7E/gUXl6h+KyCOpnfsGcSOOMEpONH0TX9uW7ehez0P/vMv1fAh + U+4WJgocyuvBJ4q45UsjhyIMX0v6zx/aSHsMoFdizp+yKw9a43YXYJqezL3+tLibGy4lYk7c5sMT + tAdqH+4S3OZMwdZhPOQbuuxPanrF+zuP1YwuUw1vr73rqjc/VHJqPyMyjmGB1e+hBtuuz9Em2sqO + X191qKRzAbZXExBT+5gVvypxAiXtUGLjRjVKTzGVYP074/n7YwV3+vOz/Zh9+xsucMVaATRg/C6v + WFa2Vd1ua8oAraRgBucbdokHIw6abNNhyc0+7nTVNOevHkrOcWuAReHKEl0VHOCz/l7zdWFEBxaF + EvvcVanAVJ35BgrUsAm29XFYO0TGv/zsC132rTbHeQagPQY3fBMGVe3kZ+BD+WGxvvirYU61jrXA + Xzzoz1MBNiZUBIA+Q7jHOwDUzPvPnz+860+ibhr1W5j7grrjUUQr+2It4KVrLcFhkEezXw0KDNkc + Y/VxndQ/Pxrs9TofPl9Hui5v5/nPn5SQXrpTx7kOnI/7E9X9Ihe3aPIFQeByxLvuXRCbAXngr57j + 6YeCbmbo/W9dCvj/0qXAs28YC4aqsodzncCXaixzOjRQXRkpv0AUFxK5CG4U0ZPEiTAsiwfGto3o + AqHTI2Y8n8kDbt1AzU/2hOlqjaSI1yBaxUyv//68D7VPWdE+9zbI3VgNy3mEhyXI5RpdpsQi5998 + dvk06SQA51bDzisS3LozegZKqdNgL16DfBPR2iPQMjZxlQtwqdJIBcpXJcYO1+yzAPPgAB96oM/s + FFnq0miWBwztqxGfub3VNUhqDr3enwb7T2nJaWzQC+zem0XkA3ustvIuZpDaLPBP4pOCMcjlBqzc + bBHfdHk6DibjwCF8XIjRnh/VmMAHB8VH+SPaMw9yPoyKEgrAvuIXJkbOvp19tmok2sRNu4FSlm0P + 4NaqB6IMsIo2Ztwa+JZOOZEAp1EORouBtDNnkzvckEtOnjCi/lGxxDO3k7u1U2HAqHR//naQmYry + Jv7AC9dMxNw+rTvrabbA62huvns1o4HFvdX8+17BrT5XJHS3BnJ6G8wH7W6qbDQxJcoHVsHWiX7y + 7q2GBXrcrym+PZ4u+D/SrqRLWVjb/iAH0kmSIQIi0gXphBnYICAiTQLk199FfW/wBnd2h66yqiDJ + 2WfvneQc0Y4XD40LPOGEcTJYfs9YR4DtXa/H1Wwv8/y5Q2GPL0RgXA5m8I1bONnOjwZ7XA98ePQj + lESyTe8FXw5CMucBAlpS43Oeg+I3n/IGrvw9pCf9qTH+5PMjxJZNvDV/NfXcDx8DOZSN2NOMYOCG + +9uDqWwAbETyI5zhi3XQMRxGvafiF/zT9Ffk+X5Jo5IehuX4SHagCNonfYbfN1iM48Kh4gBr+iT7 + bzElHXKAeuZnGk9CbK/79TfK+Sl/YCPum2GNbV9BdWlmNMNxW8z6/q4CuehNeuajgz1F6q9HO8h0 + L90dYMhKl41ImJSYPrprHi6mLaQoYcmR4oapgJ+IKiA2WRPWku/P3uqkGujD3y7UCr98wZqPliNx + 1Hmqnp2mWC/RuIIQph09Pm+lveDjfga3fDToq1aNgQxdtMJr+XxQ1aZLyM7bKVrp2t6pvZ1SXMog + TpCOx44+6t+XkaG7z3L5Td7U0jO5Zr1YzNCis4v9mF1roRIDgm6jLJMdLzxq0ayHGW7P6+28NQ6J + Jq8p+tgrptvza6yqVwO9E32kZjh7jI/LI0TO63Ej/hvdbT7dsRbtGMdT5yWfAJ/MLxOeD7aFTwG8 + FWLpAgL1yl29vTkkGlOnPgUmZ91worsnID5FFMjvyHFxvo3P3P2sJ6LzRLE6NzGbOWlnwQ9MD9hC + 1LP55vdR4O1U8RgrWg1mqhvW3/vj44F+BjGM9xZ4iItH1VPRFb/r1lTxtxoPbB4VwObulFVQWrGB + TV13ij4fRx9emCd6ouf7gBM82ICTMnTeUsmuxg9+VcHX53PwuBMPAEuTnwrvx+hHgzI2bGGguxF8 + 3I5sW3CqNiu0DdB6Ux8UX8tqWFJBiZBXtry3slphXCY+Kwh60mGDP9Ni9ZRfi0pz6w2mes9iFdSD + CrFrbqeESVmI4S+Q0WAkHkmz1g85Yf1UaJhIiI/l8CtImnxkOIpAxE//dAHrs9RU2DanCJuu7A6i + WW+9uJ43z2OnLrDF1ULlv3jSk+sHUCEfAqhd5MlbdkEwMCEffFjZ0hdjuz/ahAgHBwRtkGKz44Zw + 2f1ABGXd1+j9vGbaXD4jBcIdzKhrlEY4PuoiglqP1X/zx1pwWdEF2ga2L4d0mKnuWZCvtAs17VYP + +V3yauHZrEqqD7GqrWo8W2g1JUhx/PkC+q7KFQ159qb+5fe11+9ozRB0wgWrx6S2P6B4SpDsV0qQ + ydF6vUjxCH0THEjTqhNg04PmcAwrTJVqqsPuOXotUBL/jAPuV9j8li/gsu9vVLn8zhq/4RVoO+jj + Z2vJoMfkpqNYijhcPByJsbNhP6HFXQQip3tsM/eFVqjvngwb334u2uLatigORI9qH76x11MarDA4 + cxV9ZNqrWFDs+XC82CM+XsKaEUm3IpneE4OaHWcX/cFBDRw/VMOuADu7+/XFDM/Oh9HoYUJ7cSPL + gw+4e+Fj4x4HblByB27j4x28z0ubK8BFyLyvV+zJQ2+vaX/soX17dt6UafuQvqtuhfWu9egxq8JB + nLdCUU/zVFPt1j2A+PyWBvLzxPHWS+kOwtfMBNDsaxefTseZsXfVzYgrZAlrtw5thYN5FcXUwNSM + 34NNspWTAY39BPtvtD3vfHUQ97k8KFaDb71SSS3R+XifaSpzN23Nzr0CsuPr6DnOONjzgtAdnofj + D9+4sAnFq702yB/fCn4VxcVm9rO5oxzpIn3JUgh+WViOaMuHWH0nOeCcAK3yvSOEPvxDBNZKMRT0 + +Sgczk+jB7j3F/vgjVVAdUnNmHBbhhZeufdMTzXYGqdAfQWKDRqqGx/N5hIYC0gtzIrGQdaH1JHZ + E8XuAAn78I023ytVRclX33u7FEeFwF+OHdK1BeKQzdYgaLKcw0BoJpwemFqscq6MyCbchNXLjg9n + nM0+ekbJh/zUYsv34SWBM30csK5vhXSP5qTKaA9Sb/+mQTE3FYYwvTyvf89TfO7eL0ecMX1p/pxt + e9znXITKs+LQR+PwGlPRvoGnydnR4uGkbPJXIMCfFgVbIbURjIdDqSBrNmqKd/lo0/WZBTJsvqIn + o2tbkHehERS8nC+13CWy2axIEKiW4FLz7U6MptbXgyhJXtTTYaTx3SHf4pO74Ff8ObPhCIkPs3Dg + qQ7FNlwXT4Eois4a1r2+KxZpTBVoSm5Pz+96sRehbTm4P+4SMoDOG5Z2P6SwPbY2vvxed7bcUO+D + H3I3iXIwQyHKQxm8uaoi3FzcQ8bpjfTHD6lmjtWwnIOziuhMKXWOz3lbX/wdvPadQabjcGDUYdcS + Kah3KN7Gl0vgg4N9MSSehIff8Pd9uCND7cFOnweaylcIf/vz3mOjtw7k2HAKwMJ0os73rYVzko4S + zKgk4IAXe23kl8hDvPYNsZojpxAeQCHIvVYxvoCls5ePMFuwtaqcXnh7P8ycLqlQ56uGKl9/DgkQ + fyrEWO2x11oyI80BRfL3BzB1wYu3F0FWG0jjIKGGG77A0uOrA69i6dLQDxVNvO9cGSgvL/EQOLoh + 2SU5kYWnzGGr4v1ineeyBYyrGFVev4DxxWMK4PdGCmpgrtbGeFR6uf/KKTnst8Yngse1sHohmV43 + /BFK4ZKDMSwxVXBea4up2SZ8MZqT1b5eBlZxZx9eKqnC1qoN4fixUg7sD/1Mdq9Qr8mPTCks9SOl + 3va5O85iCyOLe3p8nR6LdTohC1ytneA16Yfa8+nO5fBMFAO/bokGVl2bCDRqRf4Xb2KuPxu5wzlP + 5p+xMMqlZg+1Lqpxsbda1gtWAOE+8AcadYlo/9LeK0H42K0EojaqiXp4G+DS7nl8mgReW/74RvtB + jJ7hwwPUHFwCpNU1yGHNvHr+9cUKU6VSsfJrDgPTbHUGMLszqqZ+CUb+t/r/+NtFcWbACsi1sM05 + St23ZDNBDD0Htuc3wke7OtdrNuQQJjmP8Pmt6NpyfioceJ/3Nr1UnTGwSrlY8DF2Fg0mFQ0rPO98 + aBJ08qqUdDVbg9fzQOqaYDV25mI+J8MMBl7PiBSFd20W2Y6Ds668adKTkq1pf+mhdKcvqoRo1pbK + 0Ql82K5IL/IzBFzD1wRu6807iKMKhP71K6GmcghfMibbExuIBaXpyHn8JMQaW1Q1gP5CAnp8CXAY + R2n2ELkmstdo73WYPh5NoOG+Q/piV6FYp1+rg5LkVzJv+mX57HIDpl/5TM9nUS24Lm5H6ECF4KI1 + rOEff9q9uyMBagC0SfKPCtoXjkzzzu3tYVxsE3JL+iK7z9dl80qWFp32xUo3PhUykaYJ2vQY9s6/ + 19AXJkjgxmept5tpyFI528mH36HEp+dp0oayV8m/+BceV9XmvwqcYS1WT0L9GRRzXQq9XMH6Si+d + ohSLVD4d4KnmnipmErA/vgLmIW2x8+OdYXXqiwOW+9HH2lmMwYrJ6++g4JPMY2dpS3cIVhSS65ma + 7v3DlmURZqg2xZmIBFjDqqlRDumnGUhHTa5g5JbMMOn2T2+XZCub81IapVtODHpyTseQp9I5h1wd + 7Ih82RRu83g/oe6rO49jiAvp9r5QjtoUW+JhrEklBiNSD03qCbr7AazicACYN4rYFXMMliokPYwG + k9FLmNsDiaiygyQPLwTR2KpZzm1VRsRMo8/8Kdn0V7UjrFxm0D9+3r84yQCzs5uxOyev4S++ke05 + 2sYfbE1Y4l4CqO04fEkcOIzxaHbyzWk++BH0Ppt3/Kqi33s2CUht3RZ/FRlh7EOKNz5qsyKze6iF + dkCP+08Llud4HOEzzaR/63WpkTLDUCcePc/LaC/QKRT4p7fyVo6HeQ2qBqqn/RW7G17yWWH18L3/ + fUhfjPJAXyfzDulx31Clj0rG2m+rgstj8rHzvFwYu+zlFdpKLmH1VnA2ucVRg4RzRb1sMuSQeZMi + o6FZdXL42qxmcVEm8DcCm7AoPRX8SelKuPkB2AjarZHV92DBwyUQsNGTEqzzQ+fQIrVf7FjCr57O + 1seAjp5EVHt6Rr2yF73D0+GuUL+E35BxqdLD49MX8OURB4XYXRUPHrGo4HPAHmwSX44MGzmGWD33 + o8bs5/iUg/oKsJ1LtJ5PBy1HcOYOnjRo1B65rtvBGX1z0o2HuZ4/VipA69pP2OWmJlz4B2hgPJ5W + +jd+y3RlHFquXuWJkbGGy+GaJ/LhNo5UsSWLMXmRN7wTEXa39bPMDlfC9xXP+LLxIXJz1x4SbJ5o + nt5VW3AM2YGdWyb/4n31W2TJvneycKIPU83UrXDuVLWUmkY4a+wnqE8QaTMhnHvPiuV67iCMUPzF + 2v4+FWsy5z7MayHEKqtLMN8uPxVKoqHijc9qG9+/Q+F9/nn0qKeDsOVLlODxQA3onzQWuLYM1qfh + /vlFw5LMNwsKKOnxORa9cAbyIfmHH1SWGFtOSlkCF+6nf/izjOOhAntvL3mHLT/QPuyFv/nxOHIO + 7QUUiXRA10eN//CZ2a2Roo9OXtT92opGq3OtQ1+aVZq/lj1j1iM0kStJObYPmcXItAstZBAbYcX4 + JPXSLTMHfRWesX7WejAkStUDWzoHWNMiE4hMC1s4D3nr8cWisfHrsDuIhWeM03uvMLEepwgMOHh4 + /aZ3ueEMTXF4NSs23fsJ8PyDtfCLnCO+RnhfzOBmpsjwOQOr9vVXz7ZllfBVFydsS0YyLOKth396 + 15s2vjoH0QzhfiZv7LVypbGNj8PP1+2o2zixLWbayQfufmfh8yF7FizXkxbEqfrykE6tcN2rwg6m + xXCiZltW9nScZfL3Ptg9zXu7maxkRtl6HvCfv7O4c+bBUxNQ7PECGsjmz8E4VV70FOYvW1wujQcl + 7Q7+rkAUvORfVFB+nio+RXhgva85DpTF05vM7xcXTubWu1rp7sAjxXjWlo9HI8A3LvJAE5N6Qm9T + B3tL/mDVptdwaU6cjn7KjfcGtxTsNYns4M+PoOrjGtdr+OY5JPhLhJ/uogwC2Fs7uB/emBo+n2rf + xBc59KcHL5v+W7SbkcDnOhf4KgyxJr5XBOHf/Jqb3/sPv+J92pDZOhPwT4/+6VP/diyHlUpWCayw + djzJ9wS2vEmTIJGZLQ5ZoLJ5ko8l2lcfhR41YGlCtkIZ2hB69DpU+3AU63cPn7rqYwP6H22yTp8V + 5TuUYIff37TNT/BgYTsp/fNPuX0YpH/4uPHPlS1/eo7eTJ7w8rwdwXA2PzS0dA8p/ccebbVsYfYg + KuHEYwN+b4ul0FavN2wyNxvm52i0QNOnOz55YvDnHz/RJM4KzbzzLaTS/O7AHx/QPaccWHF9j2Bb + n95+l48ac9i1QnbcXGjwuPLDeomaFTbyDdLTmcCQRd7DA59AULDulxc2o7e+gyMZAgKq4yec5xR4 + EKOVJ3teDeo5WjNfhqTXsQEK6Z9+A7vQselrRG4oZuusogeEL/qgbA0n7WZEMKY69hA3NRt/q2fo + hjlHrUi+MCb3mQprLGhYzxPb3uK1gSF7Zh5wdw/GXczFAYI6MGrn+Zmtf/7QNj/4WdWoWOv7QwEl + 3hXU284b/vk18j+8LY4/jeJA86B3S3xqkt4ZhO7q6kBBnYPVnnAFDd8SBBsek6Z0xnrOvsfonz+t + zg3P+ir8rXDTx9jb/KjZbrMSXOzq/s/fbX/q0QGfgFOopxlrzZpp14G3w529pdcqtpL6cf9br1Td + +C3Z+ADc+AH2bP+hfVUktjCUq8um92XG5uYtwW1/gNxa42lvfloDk0PIKO5/jc3k/qqgH2E9QRvf + WDjpEMAv8o7UmMorE2W5deAp3OUEq9sVkM0vAiePmfQ0VDZbEphCALSopubuAAs6/X7OX74iVRkb + mviwMwN+o2lHNUM4aotUJg5U16rDG15p//KJ2oYtVj5Hpxak8ppAPwzuHrtYVsE9ej74x8ft8/k7 + TPJlTqGVkgdWDPUy/B7PXQsDFbVejdzL8C8+tvWCb/LlVIukX0uIDb2gmDqLvXrcd4TPA2f/+/n6 + hxfnQft5ZRb/wOrsUw8GnhR4+xtnD+vuwfWot2/HjS8ZbD2TzIRzkwj4cnv5YBlS/Q4/+vjyuI57 + guWVThBsepXI3/QR/qLn3KM73XqND043kBZ4PYiXUKfKmZ0GBrm4RZu/hN2MtcOy6QUg9bxC9fdL + YcsfPhjZHHvoS1xtHbNUgrOuvjHWW58NVS3rgP86Oc23fEO+3E6BzvEmelyruoxX6C+H4k860jiv + Obaql10Ez9sViDIVK23znzzgdu0Jnz+nXzFZfFqiw8UXvJ2pNEO/xJUERftpE4bc37CqFyEC+s0g + hGffxR4TDkhyNCcnqnrsDabi8Qng54s7fBz5kK061AiC2ZN53EPSwjkezR5O5eh7K/cD9vLiIIRp + A3pstTI/LP6QyXDnbY26/r5/m7Id+C7XGJ+kfc8Whb5zlGZzix9puRbNH9/lrUHz3q7gM7LElQzt + Z5V6tSEcbf5ZYhmkxe9Ej8exLf70Itz8K3xW2KqtnD+08BK0Lr6k6ydkoxs4CLznjJ5fEx7mWNsL + 0qc2Y/rc9AO9SD4Hj5hXiOgfJ5vcLb2D74d5xxe4PuzVOnxVaDz4zhMl4Vr8Nr4nS6USeHKcLgNN + mrCBbakeyW/zM1YxNDwooKjHTlI5xQi0pof8/NyTNqMXJmj2ufv3974PGNaMzzII2W1/9+Y3umvL + atQSuKrqF3u/kWlzXe46wK/xk5qXmzLwX/M5wssgLNjUK87uj+akwKzdfT2WfC82q8cpAWgaGxw0 + uR1OXwWu0GftgrWg0oAwN+0Mhffp9+fHAnY0Swv6rFmIpOtjMRMQCOAucJge5TkPF1qmPqxIMVOb + a57h+uf/nArFIsPlINWbHjahXU7utv+S15s/06J/fmuxiAXz3lgHJE17qt1OM5gm6yr9+X8bn9mF + c2tX1iFbTwPWnqMXjrbatTC+tD9s5O5Ss82fg+jlE3KICcfG+wmokFvyF8bsKoQzbxwjWJtziO9H + pQvXH/mkKPffb2qWtNWmYzSMkP/pB+xUHrbnI1xSmHNmh+2Hk4L54hcm+CkxT5i2zsPffhI04/GN + Y99qazLPZYMeY2/Ri4R9e5Fzt4F/fn3xNx5BZjXQvsRnfJKVV72i+Jj+4T1+8LajsUfAFHhaU9+T + qBkV7NPeCTSzhfP2xWFm4jLFFrQvtzPZzXzH5ttHUiD/EiKPTfd3zdirb8Eff9z0NVj2uzJHi9R8 + N72Z2sx4X0xYSLqOH4N/CokazyY8GvLi+Wf2qaf5FLTgNqgFNYubE85beS/4lW4nrKRiZc8vhY1y + 9mreON34z/hmFx2+NH2mHtl/wzkVTh5IIsn+l8/Ztn8FxGkfYUMf3EEgeeWhtTjN2GCczEjSzREs + fp2HAx2Vw7rEvQyBZ1+pt4TzsIKKQVj10QNry7tlS6+9UmDe5yvdfh901WtKIFVOJ3r52mFN//jQ + 9vmffyfw2ZODzOFS6q5ixYbbZ5RB4MkBPSsssFl3VRz0vxwpkP77kYKof24SRBPt6XftJLiGT0pa + FJga76leBcvqkFILpGNNaGMHUHpgzpPih6aJ/o6z0FV4hzQhOC+WW7WuQM9/OXa53Slk0nxVYC/d + 9mS1JlLPCqqeCLm/nJrxo9bm5JOl4JrWDhHTxLWF5/5xB+vbmenFPk42e2W/Dk7qouE0OFaA6XTZ + oVCt355sng723JqRj5SXu6MeJ19rlnw7FaYqOhOeZ5SRC2vXrfdw7X147GvC93YREJbVAxnATbPF + 5tMTyKRxj88Xca+th5sbQNQMT28NltlefnkpoGeS7ugNPFm4gLpx4Oet770FByJj0iHdwSGdCS0s + iIEQqKED9yNX4WC6xgUXF1cDuW6QUT1TWE0u2bVEf+NtLi0qiPhyOJgnj8bjP8MI2KsX4KFjSkDT + k8APbHAbBdnnVaPJqH4GtmO5CWPOIfgIpkFjopruUPHrnvSmqR1Yp6utwE+Y7AjsQscWrlFoIC23 + PXomns9mQxl6+KoyB9v7SGL//n7WdQZOGersxT5KI9xrlwVfH65hi5zw6UCRoBlbT+VrMzkiHRQU + Wnj7+1QPy9fvKigJcktQE6jDwq39Dr6XJqKxLn5tEYLrKGPPX+h9ek02/826J9qRJsZJuRfBnEpZ + B6dysWj4cayQ//z4FEVX50HNgmThPKrxE52qS49TxYiAqEuLjCgKa8I6vWdLfxZScBMfJwI/hT2I + kQJ3wBQiSJ9WcAmXWyZzUBt/IX0peVgIs6KVUEqpQJbgqtskPUURKqrXSq27tJU+Ge4CdI3zhHEQ + v0OSsiiFF+5zoUYfNMNSHmAJqcYKarYGqL+cmHpomy+MJ7lgAi20CP1gvFDflodw/S4vHbrmXGCX + yz/DCtFphnMCY3J4fvyC+wb3FlYWl9B84H4F2dWzjGDLKTTaK3mxevuvAGnkWWRV4nFYHxdFQTt4 + mcgukK2Bd1svAHFeQ3wiIbG/WdJV8AdvC2Hv37NY9+ZioOuxvhBUG+9irIywhGzyEMbuRQl5R5pH + dHIbD6fywdK4685/otl+hlTl07fG3Y5ZB9Ny72G3qONi6upyRqYjMBIR6VbUqYoINMSmxmr6vmpU + +2Adtmls0udgYJuXuJpAM414rLXX1zBLH/WO9rHwwma8SwdRskgC7TA40eP67m2adl0OBNZjHHa6 + BbiDcFtB/Ut+2Cr5RzFDVTMQjgOeGh9Dq2mvDQTS91LhrD2t9iJsheOaWzR7/afswLL/NDuU5Y/E + k4PyU4hy1HZQPLpfj4aPkq33c2zA+HWUvd5W6nr+kfsKO0894xcXdwW7xZ0JOTyp1JUto+C5tYcw + /MEQ65yth2LhlTuIiHEhVXsI7W08CVxir6I3Y78bWEsHBe6t6OF9TkXL1kcdB0h/D09P1s26aEsx + luCBtxg+ng9Hxi9U7GH0PnL4prVmvXa9aqBc2v+wXpZxyI6T1wF1tM9ElNrQFkoxlhGkv5Bav8eu + XsH71KKvIBb4/IdXAThwsE1vJk54CrWp5coAfcEtw+egdWxRel4thFbOo34ue4AbokcEKtrk9HVR + uXr4hJWALvZZwZ6WfAum7I0SKDj94uep77T5evafqNdiHmO3j9l89f0UiTKl1DzFisbe1T6FO/X3 + oc5pBNoUp+sdocy+U0MQwTASJ17Bl/I3fNQfb8A44dMjxOsuPXY9qmcpf7UQy8rBE7unby+nSl9l + 2kEPK59g0Rhv+RVqhRjTpA/0YXHVRUe3CCnYmhorZNkTyaDhIxGbN0PSyG6cOXRWfoi6QDgNXKpQ + A57WJqOavcjhJJxqH16P7wt+Nqs3iMfikMBTZffeNt6aECcFB2/uDuKLe5gGpkyLg1jiyPiqXt7a + qn0/BBIGdZoNfADYQcuiQ+DWCT4qcgxoIHAtOvAmo9HwOhYsbDgCgPNTadovZ3utvgpE1HWMbb2j + cJH5UQc/UWREWOUP620z3/Jp2WHzc2y3U4dlhGA6+NRMhcwWb07noa/AF/h4PxnD2nSXSN5fo7e3 + XFWJLVd94VDokyu2ZyiAMRJ8GQi1WxJOl+aaeJ+8kWuj4Kn76o82d+2nDuqdsPXy3k5M8189RwXa + M2x/7ry9OFSaQUmKDONP3IBptUUJqePljJ0RTMXse58n/FLxhrXz5WDPkqhGyLspHU54r62ZcjXu + 6FUVDj2eju96FvZHCUFhuFFVcx2wpqcogeVXOHkS9/nVQy7vdFgUD4KzOMOh8LbNGUlD3mDTewqM + lrv3DIFwsenruhyHRdsk7xRfzb/5sYXv6+XIEu7v9CFo+dabXlIQ/+2P+PJ4zGC5i2CF5aKbNKvd + 6zAP0SOBxu3CUTzy7bCEt1WHOddFf/gzcOJFDdDy27XU5l4VWL/lcUXvpY2oLqSZtvBlbaD+fYPY + Wme7oNKcKaiSAcO6ZMo2Ofzc7RZu98VFPXI2M7W1Q2aiC9ipzz2b1+HiA4GXPHpDHlfQSug9+Ikb + hE9Z+aoXY+fsQBZKKn7eiWuv1fEZgOKSpzTnSqdYh8xq0WkSIuqUuK/ZLS5N6EbBw3tra2Iv1+Qw + wneXXOkprn9slhQ1QV9uPtFza+GQe66BAJ+WBGn2PM41k56ZBdbwTrH9gCwc//C49DkVx8FwKdg1 + 53JoXxXDW/jMGkT4uwpo9p2UKtXOD/miYA0S77TzDtOHgXHjf/DdRVcaV2wMGYHj/S8/Yj+XCWO0 + HxuQJ52L/YdHB3bXd3eYqvszEVBg2qtzpncYep6N9eb71pYO8xCuljdidXhHoViOv+1We2h77aM9 + Amou8git6uv/yzcrETQLCXd8Jzz4KwOsDSMU060RlTbScJnCd4cscrrSq+WqAzfpZgoX8my8pgfv + YZbI1uv1BipPao2iXtKdDGEW71PslrbA6IniHNihf8LFyaF2t9+KdXdA/OALmcJwrR/JE7rmWmDF + qUaw3vnYguakH/DxsyJ7SUdggIV/diSumFOItG8aOVhHi95T+6KRKnxach+uDb0oJzucMzVXwJsc + PvTqnmp7OAaHFCLxUtPjh/8Wc0qvI8zyV4JPITOZoLnpE16qk06NIde13/oQVyhnsojPJn0N9BzL + JYBce6SK/rRDQS9TAuuLXmOnHrZGSvylgSD8GRQbFqct1l0qYT8pLX0ABop55Y4eepuGjV1fC8EP + O4dUxo6y0leYC/Vo0kZHjhAk/4fXb7O3oBetP3oZ+BXQe7KLoBD3JrZdbg7XIot20Ko+PjVMzAam + h9iBV/doke1Qf00u71aSW0c0sM5RXWMOZDkYcDnga5jVw3zdpdsNbGhTk5A7WAGWfLgrfyN+CdVD + I24QRPJf/nJ3AwCrxV8JFB35SsB+34RLe0K9/Ew7hM+OsBWK3E0eLH79k3Ri8rRZNqMZOonie4fa + XeoJ1I0n6a/mvOUTMqy7epbQ96vo2NefQzg5lqGDwk/vNPqZwsAaU9zBuygPRMjJMeS062ACPAYS + PjpvrZgHzVTgPIxbIWIur6kYpRzE5SpSXTPpMB+amyz/vd8xnvRCfPW7HWg8Y9r4174mkqtD+Zpu + jaU+/LlY9TVtYOo9TGyFH7lmz+FXgtpt3vTRBGrNJ5zcgRpke4rvl5qR09Iq8FTYLeEyhQ1bfLZw + 4CoBm+efMoicOTVw/xkEbM5PxH63s7z7f+uZ7vrAAc7p+/bAPkoBaV7H/C9fUex8lIHtYGCg3llL + 6j9WXLDVn0ygloPhzbFFNFqmogG3z1T5BFd7dZrcAfG03erNjshmX3sIoGd8D0Q6GC97sYnLQc+d + PzTy0zWc/vjBxo+8lj0/NVWaJUD3ar+jhn1mbJlD8Ql/knrfOoFaw1q2Sou2+aQb/tRTljMHrvFI + cGF2uBiF3SuBFF1rfByO262V2Wu2LfTAAzsnK0TDPveg0F8D1W7Xu73sxJGTrV2p0JxJFhOmXIUo + sZ0Fn0z/oU2PO9vBX8aPHuy/Tk2dE8hBtkgX+jrkqB7P90GA31SM6LGZvmz+1JEOjVF/k/1jTcHo + D74HKleP8SnpXvaWf3TIz7JORD8Y6j9+BrX84hHBPoeMM+lowNW5D9Qu9BUspnDKYTo+MFnjk8/m + 86deoUGSElsR19rE5WkJ8yj18KkQ+2LOoqsMBcsysNu3tb2qS+rJblLdqcPtCZj/8uO92MtkBT+x + oGa3rug0NDG9W8AoOPJTSgjdi0fx+9AMy+vBRVDs5RSfetEFbMNv5Kkn8I/vi0imFvjx8wM7jTiA + 9WUpOajuhxE/U8NmwvB+7eDlcD5jS02rYr1kWQkpvt5xsJ81W5D72UBq+TOw3UX8sJZblYdWuGHs + inZub3wjgrPvpR4zjlndSq6+g4an5/QxPbV6pjezk7yb2mGr6rqQZaDsoaeoH6zvFTkkH/dYok0f + U7s/KzX/CKECeD5XsROHXb2sS9/CxXOeODvxoz0rqL+jP/zUdKiHPLiOCTi6PfdPv23r8Q43vecd + YPcexu3/A5GzCk/S3XJYdpdm/dOzWGnzuSbnNDNh5yln/PJWvqDvvFLhH993SS2Bn/ddHJS7uUVP + vvQMqWYtAqqbAmJDoVcgVvvKlKP8OHuSmx1rYXPr4fy751hvtHe4xBB3QDjLZ6z6D1yzuy48IXU9 + A5s1/9GWtupyQPe7K9lp8ldb1cPUQo9XKdb8rAAfXz7coUeUgRo3oQQ/Vz3osAtWB+OMr7XVLQIf + hv54pSow34Alz06HDj3/sK1KJFzfn0SBubDdQlSJFc6ZnUEY8s+YcD9W2Uy9eBLczgXRsIe2zVvi + nMj9Mg/4FZ3UUHR6UMKHJxEaSMtNW7+vlwcWJeS9fWaJwwpSyYGHoCPYOY2FzWQ5i2DE795Yi7uz + zTV8sYNQ+N3++U9//gE8DNJM0BxmgN+HHwOu/XP1lnLruFD2YQDsxXewk467epruvxaO0QwxTtVU + G2XJIvA+pj+sf2yFLZs/Am3yfXiJ9g1s9uZEH3IVNgl9rCmbzrnrHP70hgkev3/fBzhUBXouPDdc + n8Q0t21p+y+f1J2h1B10+jmir/ehqdcMRx2aD9cF4+rRFBSVVxn9PjHxvvcdCT+XixRAIP8u3uH5 + mYs5mtId4JmlebePisEc+cQBv7HM6QUgnq2+FAjwNCc5DtTfpLHryAVw8bwnvjRo3fTbrYG7cxv8 + 03fLXz7IJlPHf3x2OXq6iXqsvKjnkl9NZ66RoUL7yZOZ1IOln346XJ3n4EVGg5mIKwZBzHmE8E97 + 0OY6K1ZY2P2N7DPrVrNdQlPYqbcHdu0iKNaPe6mg5g0ONl6yGvLfc9kiReiWf/x6RI6dQyoQ7A0l + F4KpcXkHndY281p3HxX/4nvGrMNYyVkxb34ErOrDTLVvtrcXmW8M+CoKE2ugmwAz5jWBVQ1mrGqL + 85c/Goj5CniNZuJ63vwVGGna5O3uxNUW+njvoMLlK2GBY4CRGy/d4ZhZAjbfzzycF+Noop06fP70 + dzhrda3CT64DejGygS1uNEN0vnG/v3gcBO07bf6DU9EwKr/DH56hMjSO2HKchi26obdwnmCKC75J + NXHTz7KEuzvOuEbRhJFXZ5TFKKWWM+vDyiLdhLfsvqPP2LPDKX6vKRSx5WPNXvJQ2PQuEoU8x+eT + fAlpGEAO/unR0+Nrh2zLf/DaaB7FjfsJx/iSef/iA/RDWcwyVVe04QfpBgGGXclCD/7xTdP3RLAY + Ox2iTe/S3Dxl2mKapwj++cWLmYtDd83SBN53jz02W6MY5m9wb6Af1qq3mhmx12OcKvD81kdsUfcV + sse6K+GZ5+aNjzKwKMAW4M2FkKa+d2NzUPsdDNx3Qs35+WD9x72Uf/4JdWRK66UIQA8jRipqcp/f + sMRYLoF4nzoScbTRFrf1fFCtRoltYbzVdPMz5M2/pGc3H7T1dPJltL8m743P+uEYB5kFuco1Pb6p + p+JfvhaupkWdT2HXPO/VHnodCky9unULLg3IDphpwuMLco6F6ECWHv7GY1bAqxiymV+hnkQn7xr3 + sr0E+tmH0fNNqO6nQcjHhtzA8gAK6j4SZVjmQ29BCOwYm2UiA6aKSwndwS9pGh3qcNDbVYVECzF2 + kOaHc9F+G1AQ807PteANwmt6NOB7FK9kH/1UbeSvFw6GYj14uyCxmLj52X/rnboJ4DQG1+sMFS5d + 8UvsYf2Vl2sON3zBp5z2YNPHKhRu05NaF6nRWHusdrBuMoidje+yxyqUoMJJRB37Zdjzzcoa+Odf + nzzlwVZPzJ9g8/OwEghuOL/TIodvp73QP3+CHQtv/nseIrNosJdwxRH82z84Pw0XzDK1tisa0Re7 + jVCFvLJqEjxOkoztOApsXoW6A9qQt6inFJCtJVUqefOL6JX9fgPjw/QOSD8eaawNJ8ae3m6FB1f6 + /YvnpbS5O5TWtPOoc1SY2IhVC3NTcvCfHpq3+Adu+l3pEQciGOWjStBBdffYvGcfm3TuLYH1I5Ow + bs8+4E+Vs0Kdixx6rN/tsJDL3oQ39Qnw334Ib/i6BHWz1QlQJbLx6Y8F9D3n4tC5t9pMnMcK3cd5 + JsKW/8dTMFSwvGUJtY5cVX/aoUxQdeVymv7pW0W3OaC9LwHWNn60omxu0N/+gD6eKzBmjysHaeRY + 9AguQ7HxhR7Gwn2i2fPjh0syrAT0gv/DR/a71GyP8xxeoljD+txJ4TTPvfenx6gm3B0261OdoDfj + QjKw7YqJVWfc3/4T9kprCql6MSRQX4ya/vGTVjl+PNA7c0kOUD8y4Xu67OArOVjYSSCq109yuUPM + fEhP0jD/01Ng2V0B1gr4HVa3/nZo08//AQAA//+kfUvPgszW5fz8ii9nSk4EVKrqm3EXASkURE06 + HUFQQEQuVVCV9H/v4POmB52e9fjxQatqX9Zae7OLYuXr8MnL6xpaW2+krhCO3rCLKgX+9H57yc+T + PnTwl18CadE/pmFW7T++c0wfljfMLDmh++HqEvAUOScLn0T9SZ+pvtYSY+jchwqtRDwSENWzx62H + YsPUay84sp7LRbexfQcqczjWufmM5tC35V89De8FSU5pEb4IWrvdCVvv3WgQeTaHP31w8TeP62Tb + QHm+h9ThyDGYcR5lxd7XGb54dhfxFaPN9mQLepDQy5Z/G/EZotPHd2iU7GE/muKCmEpwpWqvHznP + oSD+xf+FL3By2J+uMPsUMJBPtuqtO6UncOM5Od6D7JkO5OuUsKOnN7YXvY+Fm7v4qw8EvNzsezGL + lwuS5PuOBn1HlikRvau8g2ZLd4RN1SSHlxDq52XKOXFcY/46T4iU4Qqpbr0Olay8Ng7M1LbEd/W9 + jviaNCYwNvtz0DT02o8/PrY+dgl2Ntqe8yL8DuCHHzXrrUXSmL0aJERmvJR0Rz7VUiRAObrvsWXs + Gm/M9vwffHYSRS0VqzIicOHPwZeMPGX5WjXhos8TJR2aiN1e8VM51JZIF7zNyYInQbzR1oHcyXVK + n8J3gj9+fR5pEIm9ouooEYlC8X58epMwG0+4GuSSiGpIU/KZLyY4ylWEVeCKfFaEQYcKT1XsCf42 + 5Z9H7ypL/g14Cj8VzxIhgYteEPAobiLWnBMFLnp7sLEFyZiTK97AdFtIi/4/GBM6GzHa8XqmP73n + x/+hLCnBom94KQcTEYE/qlEAViszlU/bKIAP6ooB19uqGr/3VoT3r3vCp2Af8MlwKh9117Ck+zOi + xvyLn/GZpNjchtt+dMnmCrfbbg7mN3sYUxZ/BvgURAffj9Cq2K8e8vt9h5ZvPOZcSA3rwBypNm4d + Ll6yYw49V9fp7q1jLq4O3gR/er5nTQEXif+YlFXUPAO2HrR+tAVTQE4krqnjvl4es0GTKKcRWrio + 57fHV/YpB0/1FpFKPW74D4+hlwoKgryURXSFT1f0OZxPwSY+G/1aPo0bQJPoib3Vyoz4tiFPeNA+ + lEhovVwMudQHuvda+dXHPTl4nxp4v6NywY9Rur77sg6/EnsEgit/jaU+PcFbp1jYWPzt60VTg+6o + +FD/JgoeK/P9orfsm0BpzyuDnIt0AkJkx/RQ2Voq50R1UQS170/PNOjOVxpQ0S3Ci78C+cqzKxRu + fUvQku+534ESLPyaTEu9ZyLMyf6/phRs/98tBQ9hlZKS3c/VvCHOFe4L3cAWM3eGZCHWoJVb3YMt + GpNqzu8vAX7eQ4R3cSFw1g1GiZRUv1JfPsUe3ynoCSaTXbDlyF4k+S89VAox0YLZ76kxnfbOFTZD + WmI/pQOYztAZ4LnIX4Slyasiut+38LjCQTCmtzIdyUsQNt6lrvApfb0Bm+5FDo/ISeklfbw9Eh13 + NYKRo1PbP1qcJesw3AadawaArG9Vr/mvDh2O1Rv7TNWj9WVjCEh/fOaANKBLubX6CnDt6Bq2tt2H + s8PjViLzJFyorxhbj5/hdEfv1TejebnSDFGygwTq28zDJvn6FX+eXAa1PNGp4fSeN7dybEK85jLG + kfACIj7rCXIPIqS5EtwiEqPJhJf7TqSqQ3d83W5lEc6FUmCz2umVnDYvBdINHqgRHnswVlp7RaP+ + HegjrZt+3imeDGe1nrFJItNYy2oewFyJVzT7HGVA/N6WIdJOHdVU4Qmk6SzpKDTjFw3JV+xn+FIZ + Uo2Nhy9mXxn8eX8IsHKtDJ/l5Ar4Tr6ZMLRGA4eom8GkJbKvlGlYYfsaw55e2VeEoJEZPezvOZiF + wrqj7DT4NE0p4TN8OQwljwuhTnjovfVby65wdfuQ5boJF4j+p7YRuLAc+/IOA06qJoGkiwuapgr3 + KHfrCY1hs6WuvNzt5sxTh7ohveNdoaTemtlqgpb104K1csXzJ5pgbTIaKNVTAlzD9wGGlsroCcVW + JAql7MDqYyKaFcHToNp+U0PpaBTUMWUrlZfRX0i1wpGmxa1JSVYCF3yFlYE951OnXNx/TLRq3FsA + i0D1pmZzy4BahgF1rWidDm/bltGqEbulpF8acmx7MbQ7+U1j/3OK5hI/BmCRacKWlgZ85Htugm64 + 3bHJzL03tPPcoZeAInorAtuThqa6w/vhrdFzSBAYJaDqUD7EJJCsy9Wbs6bfbD/rV0/TMD5V81Z6 + KFDxuzPZspfk0e57zeEL53uqxscZjPqhSpClP17UE4zUm5LNTQBBFqyXt4ppP7yh6iNN7Y74EsdK + xR5P3YbAfZU0nJFRrbWTMYFVI3fY9FnSU2Y7MaytQg2ky2B6E70od+h9Lg4NnGuVTpbWX6Ennc50 + 302hJ0V7cSnZZy8ck7yNJvApa2T7eRlse3vT88/Gy6G1OjT03r2/gGTvOoDS2Oxx2EA7Xdtb3fmd + N9ZMxfV4dcgTGG2KVaBYeVqtb8u4wCOxd9hhr7U3v9VyQPd+ZWJVvt/6SZEKBSJBpcGszUUqtlSt + 4ahFLnk7y+BCb0QbUDknE+shU6r5zkYZ6ul0w7/9H7r2OSE1aqeglM2DJ97kpIZ7Yl8I01JxiX9e + CWVQJNj4GCMYL4qXwEeRH+nDRA4Q0y7u4A2JGS4O3oozmp5dtNgntuVE5uPrLD2hc9Z0Qk32rHi8 + g8Nmf8JCsGlEh09hdpFR7vsXHH2MkU+Hb5GBRz6p2PbtV7/O06hD6RvaVF/it+goloy+4UfDXgfC + iNarVoC9u/Nx+smQNxwqlEA5u+n4aNI3mMcSdsA+uJxa28Lo5T0FIbynz+DPH+jjxXPUZa2EY2FX + AK5G+wxGyUbHt6qHHls9ZQY3pZ/jo+LB6pvfl7utX8aKyNviVTF3Ji2UaAtp3usXMKC0bZHD1Ire + /Qr3U2QdOzjFT4jNdLA98jqZDF3ulkiPRJT5VFlHF4Ulqant9Ja3ni71HRXOyceYnYjxO1/FE2uT + Prp3BSb4PQtwqyEjWKmo5rNXzzoKNuWOms5Z7btafQ5or9lruis3Wj+t9UsDjY3xpXjZf6n7vExQ + v1iIXYJf6ZpcSxtNpmnTYnneZBv5BDqcbgOCLqbHnY2ZwVP6zHGKNjKf3sLS5f6ZbRprlpPyR+Fe + 4fw8JzQk4gaMEU4CsP5UHjXig+Q1QWEqkDd3n/p78xrxtJwdxLzQwA9UupEEiGDCTeS49HwZaoMf + T14I38LKozv/M6RDcN3aUDpqBXV6W+mZKxYE7m7fnHipEnlSbfsi1MKrHCj9UeHctEwHDvZ7jfXw + OPKxG40AoVV6pLg6Tpy4n9aGz/qVURulLhBvQ6jDrdq8qFrt3F5SPs8GXXa1h2/aeO1ZJyYE6im7 + YXc799H4PB5NNL40Sg/I14Aswe19eYuvxqoqlHwSd6YKfEtXsIMaz5Bl04Zo2Q+sh9nJkJZ4Cevz + 4U2TZmNUBD5CHVycMKPZRdgBcSd9rzAJUI+z7m1x8e2sNvD+DVS8Tx+aJw/9QQR4Pcs0MpXGmM7C + PC1jza7Y0ELSr1GTbOBMBE7TvXlPZfQSM0Svjx0+FUIKRPwMGohb40gtEumG6E4iBEpKvtiR1S7l + iRYr8LK2zgHfa7z/88fl+TQozvue/fZvyZfBvNfWFdlIkQLDUo0ICz9y+sVnJ0ZfVL9w+LllEQvD + mwIfRXbEmWLcvL/42J7xA7tMtaP1VTptYMy9E7XZUERT2mUdBFIW4qusuinb0HkCb1WsCZSg11NR + vzWQfLuA3tKHZkz39yZAl13jUexU72rc81oFz9nc0SuZmEeme5HB+Px4Yi3MYoNbcLOB83HckK+y + k/hEnlYN/eqOArGiGZ/D2G2haWwMnFrObFALTgqKLk2EL11epsx0RAVeSxIHTFHldHi8PAaOmJjU + iPWQz8tdp4ApNlz6aX1Pnu6X/Icn6TG2+37peamRotgDkZuN0cuS4sRwiT90v8Qb7o3SBsqHhGCH + JYzP3ej58Flm34BHoDJIouzj7SlYWl7iPUl7/RjY8IefLPbmnF6lkwKTtExoXn1AShugBGh9zdbU + 794WkLdrHEJzavf4xryhmi5Su/mzv4QAJZ3oWYvRVL1DqvbBxpu28qOGW5Q0wRqlLeeymvjQuK7O + 2L0IHpgW/AEWfE3VZtY9PgyrQLlIhAeS/wmjtUt4A5Qu2eFdJ7p8dMAqgBa6BwReuFFNXz7qEOyl + Iw2IV/a8ka8mNN6lSsPQNaNB1F4bqBmeQpDJ6oqfIrmFS/6jwXYcKqr53xYI3STROK5AP0V6W0N5 + e68w3m4CPq+yLIG3IruRvrsbvaRFh3wbvYWE1BF4ebMN9iX8rV9LXzsuXQXK4OfDbewVtyZiz8jN + oDRZnPqpMhs/e0Rf/SvSuBDWYJb1RwOz2CaL/wXVeJPyzW991KhODlgHFY2hpZkrvNzUBeY58gTI + vJNBd8puC/iVbiZIL0zGTrTWI7ZKsw4Ig3cJZpKTdLYceQOOZLqSlWJwbzge00ZZno8NU7E9zmwn + ARIJ7zgnA03Hsc0ZbM+A0EO479JBeB5MsMQrfN1HacpgqTgwf5xzan/cczolwFPBVCNGtevhZrC+ + mmKYsWymuPrcIv7yn3/+QC3ZE6t5t75P0NJ4SIAEx4q/ItlVlniNXf/qpVPHaAwGK5vI9yIcwEyz + TAY/+4uqWun5ljAbLXiOOkU9VVOQqDp8HKMO703ZSNln67nKQagsqjOnAjMYLB1G7vlCrx93Tscf + 3hRSPaQ7zSqjdq34MfT23wRfHcQ4S0wMgTJ4BQ4c4oHJcDwf+nPs4qho52gaSkeAsRnfaEaSZZDk + aAXw4xAWTFYoVsxehwxdbUkh3bbZefJavjXQPMHLX/yaabxv0Qg3Dr4XNyedpiyXYd6HPbYsnwO+ + Q6QG0eaxoodC3qQjepxU+JrOMllZd+XnPwGK3MuFxs1opOs5Qi0Ea7QN2qIOq/nLxgQEj/eN+nuc + VWM7bSEU5+d5wSOZx/yXc4JD/DLJl73OHt8SxYSZVYRU22vrnhu2JqJGiemSf3qDiTuJQLF6O7T4 + +cPrShooAACxLXusKg3zmsMuPUXUrWQh6mLjHoL96pCTTcr7anwGTxUGmb/G+mV6gvEMNBX98KZw + 9VceO+/k4cdXA641a+Nnf6B5XAK608pNNJOXAH94ip4+gWhMzGXTz77xPjx8vTYdLQLtAe9ooH0v + nEvCcrFGdlfoPtRNQPnhIQKtIDXRmGMBpkZeDtXkJmJXlV9ptfBxuOHnN8XNVBu8OG+fPz5B7eJU + 8inIUx/kfnAhsm+/qqE8r0sQ8/2JOtqFGdO9vyfgtx++iVxOvXc3/J237dtaz26UTVA9jxUpSe6n + 3/H9HuAxdAHVFA/2PNlqHdq8ngcaMmfH2V463+GnLS5U8892L47XjQmrfepRVX4jMJQuOMHWyjfk + 6dTVH99Fl088UcvpPwb7UuWkPCjYYVx9tim98+IJa3awccpes8FPQVbCzPZe2CgeB29dW4arYAnm + OHJkGlG4tLBcnFOG3bQ1UlGRLhvoXUMR39LHyxtdik6/eBpw/1xWPAGMoHM+hfSozUXE50hq4UOV + RZJu5280s3iVwAWfklX12UZU4fkTipFzWvLvy+ATvgVgE7ku2RQ3P53xbWBK5AgSdhb74Y+6fqIg + iDp6SG/6Msj3UcK0aS/00IAhnVfH0AQ///eW72ttBGvwQK5B3YvQg2l1VmW05DuCFvxL0mojQn0A + T+yY8juajMOYgMdxBQJOHI1PjToMyg/vazLm6bCZZxftruiAbynf9+KCP4C+zT26385jxI6xnCiv + 6bJMQYiexsStawsX/ol9qgNOfnzxdJcPpNVGpWK1cWPgt/6T7z6j6Rj7Hfg9f+98nJTf+eUJDuo1 + X+xP52w/6j76XPVXsPG3mcHbrSzDexCdCImtY8+X81X+8NL1kBrzIW0diD1twtjvImPxHwH6c+IG + ZZpovaz7VYs0tT2SrcP8XrYs24H9tcB4wQ8GVXovhsnQP6guVWrE1DN8grrMX9hb+BFRIyODsaV3 + 2CCR7Q0qbgfIHXWP72GHPI6yMAGtOX2okb4Mzr02naCxvlzxYX8XwBxctnew8GfsZBaq2KNuStjf + vGfASdR4I2g6BXxVMcMJmRLjKyt6gpyoewadpln9dYq9TglntqFeKK1S1hGxBp5TYmqEWej94jVY + Po/19PXmXwnNJtCOux3VutIy5s20zVBtr0tqWWUWsRbRBgR+dqJnzWoi9uM/u+vqEPzZ60//ujhw + pOa2+/Jpio0OnoXyH7w1SLYdw7Wz3BroIAamyNEJPAolwnrvomiGbXKCi76Hb7LaRrSI1BzIg7Sh + gS/dop4tF4WVQ9qRWiJqL13WpxC6Bxliy0cb3oSRDpHcxZjuzN4wGlkYJpjuagP7Wr6u/vDWqPcD + tjWrjqYEiTYcTDOg+2itRmIrlgQSD2b0jOZzNK2K3RXmXmljx/l06fzz51MWUGz77FTJ/UQFuNuJ + NkHS/I64+z3myFU+OnaL2u8nMJoDipP9jL1QV7k45vcA1mX2wkaz+fbs9/9yl2DCtTDoaWSEJzjw + 9zVQ4jju6fNSK/CH5354t5PNAMLPJQkJ11IGhkcFZYU8ZhHbse7y8XToZCj0hU0X/cLg+7nuwBHJ + w8JnXh6bwnCDlviP92yNKlak5w6eYa6T7Q8/VDv9ibrD2ca2n7meqOInQVejU6lavL6Am3pkopa/ + n9Rx0IlLP31CE9gGm3vtWImfVQP/8p2j+SOfn2nRLq+BPpb43PSTJSgh/PGFSNo2/JevgCHBpYXl + YRldoyV3hNT6hvdpa0XjvTu1yPjUnLpbSCvuNs87LOSrSH/56PPTK1l2c8nsPzuw2GuuWM91jXED + 2mhIP1UD9aB6UD+tnX79yI4x+uU7u4tUQ6odtIGKB00cFD3zhi2vS1S0xYd8kPVMuX4MTHBWr0Gw + kmYtXUtgO8D96SBgZ4kvc1pPT+i1xg2roVunS/4LYIK/kObSXEVcj1YuUBSGSOuwoPp9Hi34hdqX + dRON99EgMPK0lu7lk+LNC9+FjR/sqC9Br+KLvgx+eGrRLwBh+9kEi74TjHF44Wz2SwcufDtAWjSn + pFZbApySiFj/3DbpTO7f/Kcv0L2/VQyWjas7/B6PJ4oXfMvEw0WGssIt7BZB7U1Xqpk/vkndZvU1 + +nj1zaEkqCtqmFQDo6y4MfDruaBuI9Z8XYW7Ozz1q5hi7Svx7rTb63CIRI59VPrp/EZzDd1t8KR6 + +rIAbeSrDVdO9ln06bL67MUHAQh2ByKy6e6xhb+Dt/aUsN6VlTdJau3Dx+o7kOd263NpQ5w7XM6H + BuG+jdie1TFc+B82Y52AZgqvG5ikzwTfU+VscH5aQ0ilvKX4Qhuw1AM66B6cPhDKlebJS76Aqx1L + qEq+cT9H/tuBkPU+9TR/x1mezD48njAJJARv/XBvXyf4PG8FasX6wNubnDcAln5C80iowGwKU4js + b25Ql2AtlcPr2wRgvdoGPzwkLvkSLvkbG1qz9thYNyrsRqunh+rsVeLLmwT4Ph7vNG+WQdM0t5eW + meRCrZAgTh6Pmw4di0bBpBW44t33moEbeY740qyoMTP8CMDt+XkF7+KxN2btLJdA7vUnmaVtB7qd + HNmwrdwu4KH96UdUQle5K+cdVi+TCdaipXfotPVDemUDSNlPf8YtLYNX1Z6jtR489J8+SXfIUlP5 + 8Lg9gbXCDd75x4pP021k8Bg6gOLtpfT48DnboHLnkT6YqkZMFmqGRPvzobYWOhX33uWir+ssQKjs + ItIIYwjauPLJUdtbxuKPT+h6bE9tFNo9Q4Vxhdf6NdBYaySDJ9t9C1GYCMHkP33ei/uPDZXsqgTK + kv8G9XRj8B6sltlnNqvmtxExuFfvDQ2calcx0zbkn39jT/sizvfrYPrDS0naeumUp68OLr/nh++i + 9XqjQ9jca3fRN0RjbKydCz/uLqZ7v5r74XkjJ8W7nsQAoI0M6p30uqKxCBJqVLVS/epBf+vBRCgM + tnoKDFROaFJs3e+AaeH9CsquPgVyFQzewkcV8AidCseKJ/czN68C7ALtQiZyKXn7in0XKMFjJKjo + Y2OubVOEOFFcbFYUghHdnwJSrdOI7c/zAcYhv/pgOzZLfaqoqj88lz8uOT5o3xtg3y50UeqcYuxd + +Lua90TbwGmS/QXPPDx268IW+nH8pcGCp+pf/i7WOwu7oa7ztWi5f/61+Kfzwws1XPIJmapTAGZ3 + 9AM4w41KE0lyvbmBYAP7S+Jgb5ZUg9eW4QDFE0xsa6PQz7p1dZVf/WTR44xfvvzDO47Za95gOIYP + Ijn1KZbmV8oXvAyaVBjIXCRVNb4iM//L7/7v/Gd3mlAoviB1Heb2g2jeIeyvD4wxEVYGP+K7D7fk + dMK+f6gMZphhDovvfkPd3partjwMC1/PZ6wx5w3mhW8CYZBabDHsRUt9wUXWU6pxzk6hIR2v1IaF + fBexy17c4LZ2vsNDuI6xkz4MY1rwGRSCxxsvz++nED99mM02xiaKXxHr5ncHc0Bf9K+et9gHzIOj + G1SKGkeLXr2Bi///4kW06F0NFCfRpOfevaQjbJPwh0+omb408OOX4IjEAS96Cufu4JawLfMb3lfP + NZjM3eG5VR47j8zhEYP2x5/eL22FLQRv1TwfXjIifjcG0HKowXCOfRjO0wbvm5VnyKtrVSP4Kq1f + /upnPcyuv3poAOQ275motj4Ul8rMZtEX/tZvGoqBVRmff/WFf/CD2sylMesuC1Ecc0xE1HCDB7nl + Q/8Edzisdl1PPvLd/PkfWeKNMVlmnENwmXJcOJVV0fJo2fDinjF2ulxN5fVWb9GCd6hV3Vg0ObPS + QbFeddSVpNL4q/fBWQD0V++Qb6NGwMACOZC20mCMn/WNgVF1DLrUf1MxK8Ireo+iiR/d1HosOu4a + WIixRhoHJZy8d3IAI+lkBah4Ys4PLw6Rlsc61VDR97wx8PSrZ2KDTZm37K8Ir/ZawR5LBv6HL15C + pGPVssJ0wT9XBem3mQzSvEsleX07QfbYpXjX5WXU7+SjidzO9bHlZ4PxDRNO0NOc1vSMmtmbslp5 + gkBplD97nIzDO/7Vx+mO2rAfXGIoiEGP/vBwug6qTwJ36+WiLdR4nvTDa6PqGmQVxl01sciC0DzT + lAZXf2UMlaP74J+Wgn/913/9j6VB4N9N+8jfS2PAmM/jf/5Pq8B/1v8Zmvv7/Wss+DcZ7s/83//9 + TwvCv79923zH/zm2df4Zll4DEaF/Jhj8e2zH+/v/+tO/li/8X//63wAAAP//AwAfYHydugUCAA== + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99ac8df5fa8a-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:51 GMT + Server: + - cloudflare + Set-Cookie: + - __cf_bm=TyZMV15EPWqQXIbTEKRZ1Fggt3ZrMRC8fwKSoQemN28-1727451711-1.0.1.1-1IuaZeYu1bHwXOBs1o2p6JiWcDrm8vtRUTauyAVfYSTF8WGo9o_nFHMe2VvmzxUrbomkaIIk65ZyCtLLaSGbrg; + path=/; expires=Fri, 27-Sep-24 16:11:51 GMT; domain=.api.openai.com; HttpOnly; + Secure; SameSite=None + - _cfuvid=8VnZjkM1FzzkfufFNQobFDcIb4Hs0eOF3OZxxZTYh.w-1727451711771-0.0.1.1-604800000; + path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-allow-origin: + - "*" + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-model: + - text-embedding-3-small + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "128" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "10000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "9987878" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 72ms + x-request-id: + - req_db08890591133dd81002e0e3135472fd + status: + code: 200 + message: OK + - request: + body: + '{"input": ["ability and Transparency. 2018; pp 77\u201391.\n(20) Lapuschkin, + S.; W\u00a8aldchen, S.; Binder, A.; Montavon, G.; Samek, W.; M\u00a8uller, K.-R.\nUnmasking + Clever Hans predictors and assessing what machines really learn. Nature\ncommunications + 2019, 10, 1\u20138.\n(21) DeGrave, A. J.; Janizek, J. D.; Lee, S.-I. AI for + radiographic COVID-19 detection\nselects shortcuts over signal. Nature Machine + Intelligence 2021, 3, 610\u2013619.\n(22) Goodman, B.; Flaxman, S. European + Union regulations on algorithmic decision-\nmaking and a \u201cright to explanation\u201d. + AI Magazine 2017, 38, 50\u201357.\n(23) ACT, A. I. European Commission. On Artificial + Intelligence: A European Approach\nto Excellence and Trust. 2021, COM/2021/206.\n(24) + Blueprint for an AI Bill of Rights, The White House. 2022; https://www.whitehouse.\ngov/ostp/ai-bill-of-rights/.\n(25) + Miller, T. Explanation in artificial intelligence: Insights from the social + sciences. Ar-\ntificial intelligence 2019, 267, 1\u201338.\n26\n(26) Murdoch, + W. J.; Singh, C.; Kumbier, K.; Abbasi-Asl, R.; Yu, B. Definitions, meth-\nods, + and applications in interpretable machine learning. Proceedings of the National\nAcademy + of Sciences of the United States of America 2019, 116, 22071\u201322080.\n(27) + Gunning, D.; Aha, D. DARPA\u2019s Explainable Artificial Intelligence (XAI) + Program.\nAI Magazine 2019, 40, 44\u201358.\n(28) Biran, O.; Cotton, C. Explanation + and justification in machine learning: A survey.\nIJCAI-17 workshop on explainable + AI (XAI). 2017; pp 8\u201313.\n(29) Palacio, S.; Lucieri, A.; Munir, M.; Ahmed, + S.; Hees, J.; Dengel, A. Xai handbook:\nTowards a unified framework for explainable + ai. Proceedings of the IEEE/CVF Inter-\nnational Conference on Computer Vision. + 2021; pp 3766\u20133775.\n(30) Kuhn, D. R.; Kacker, R. N.; Lei, Y.; Simos, D. + E. Combinatorial Methods for Ex-\nplainable AI. 2020 IEEE International Conference + on Software Testing, Verification\nand Validation Workshops (ICSTW) 2020, 167\u2013170.\n(31) + Seshadri, A.; Gandhi, H. A.; Wellawatte, G. P.; White, A. D. Why does that molecule\nsmell? + ChemRxiv 2022,\n(32) Das, A.; Rad, P. Opportunities and challenges in explainable + artificial intelligence\n(xai): A survey. arXiv preprint arXiv:2006.11371 2020,\n(33) + Machlev, R.; Heistrene, L.; Perl, M.; Levy, K. Y.; Belikov, J.; Mannor, S.; + Levron, Y.\nExplainable Artificial Intelligence (XAI) techniques for energy + and power systems:\nReview, challenges and opportunities. Energy and AI 2022, + 9, 100169.\n(34) Koh, P. W.; Liang, P. Understanding black-box predictions via + influence functions.\nInternational Conference on Machine Learning. 2017; pp + 1885\u20131894.\n(35) Ribeiro, M. T.; Singh, S.; Guestrin, C. \u201d Why should + i trust you?\u201d Explaining the\npredictions of any classifier. Proceedings + of the 22nd ACM SIGKDD international\n27\nconference on knowledge discovery + and data mining. San Diego, CA, USA, 2016; pp\n1135\u20131144.\n(36) Dhurandhar, + A.; Chen, P.-Y.; Luss, R.; Tu, C.-C.; Ting, P.; Shanmugam, K.; Das, P.\nExplanations + based on the missing: Towards contrastive explanations with pertinent\nnegatives. + Advances i", "anations based on the missing: Towards contrastive explanations + with pertinent\nnegatives. Advances in neural information processing systems + 2018, 31.\n(37) Jin, W.; Li, X.; Hamarneh, G. Evaluating Explainable AI on a + Multi-Modal Medical\nImaging Task: Can Existing Algorithms Fulfill Clinical + Requirements? Proceedings of\nthe AAAI Conference on Artificial Intelligence + 2022, 36, 11945\u201311953.\n(38) Zhang, Y.; Xu, F.; Zou, J.; Petrosian, O. + L.; Krinkin, K. V. XAI Evaluation: Evalu-\nating Black-Box Model Explanations + for Prediction. 2021 II International Conference\non Neural Networks and Neurotechnologies + (NeuroNT). 2021; pp 13\u201316.\n(39) Oviedo, F.; Ferres, J. L.; Buonassisi, + T.; Butler, K. T. Interpretable and Explain-\nable Machine Learning for Materials + Science and Chemistry. Accounts of Materials\nResearch 2022, 3, 597\u2013607.\n(40) + Yalcin, O.; Fan, X.; Liu, S. Evaluating the correctness of explainable AI algorithms\nfor + classification. arXiv preprint arXiv:2105.09740 2021,\n(41) Hoffman, R. R.; + Mueller, S. T.; Klein, G.; Litman, J. Metrics for Explainable AI:\nChallenges + and Prospects. 2018,\n(42) Mohseni, S.; Zarei, N.; Ragan, E. D. A Multidisciplinary + Survey and Framework for\nDesign and Evaluation of Explainable AI Systems. ACM + Transactions on Interactive\nIntelligent Systems 2018, 11, 46.\n(43) Humer, + C.; Heberle, H.; Montanari, F.; Wolf, T.; Huber, F.; Henderson, R.; Hein-\nrich, + J.; Streit, M. ChemInformatics Model Explorer (CIME): exploratory analysis of\nchemical + model explanations. Journal of Cheminformatics 2022, 14, 1\u201314.\n28\n(44) + Lundberg, S. M.; Lee, S.-I. In Advances in Neural Information Processing Systems\n30; + Guyon, I., Luxburg, U. V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, + S.,\nGarnett, R., Eds.; Curran Associates, Inc., 2017; pp 4765\u20134774.\n(45) + \u02c7Strumbelj, E.; Kononenko, I. Explaining prediction models and individual + predictions\nwith feature contributions. Knowledge and information systems 2014, + 41, 647\u2013665.\n(46) Shapley, L. S. A Value for N-Person Games; RAND Corporation: + Santa Monica, CA,\n1952.\n(47) Molnar, C.; Casalicchio, G.; Bischl, B. Interpretable + machine learning\u2013a brief history,\nstate-of-the-art and challenges. Joint + European Conference on Machine Learning and\nKnowledge Discovery in Databases. + 2020; pp 417\u2013431.\n(48) Lou, Y.; Caruana, R.; Gehrke, J. Intelligible models + for classification and regression.\nProceedings of the 18th ACM SIGKDD international + conference on Knowledge dis-\ncovery and data mining. 2012; pp 150\u2013158.\n(49) + Bastani, O.; Kim, C.; Bastani, H. Interpreting blackbox models via model extraction.\narXiv + preprint arXiv:1705.08504 2017,\n(50) Gajewicz, A.; Puzyn, T.; Odziomek, K.; + Urbaszek, P.; Haase, A.; Riebeling, C.;\nLuch, A.; Irfan, M. A.; Landsiedel, + R.; van der Zande, M.; Bouwmeester, H. Deci-\nsion tree models to classify nanomaterials + according to the DF4nanoGrouping scheme.\nNanotoxicology 2018, 12, 1\u201317.\n(51) + Han, L.; Wang, Y.; Bryant, S. H. Developing and validating predictive decision + tree\nmodels from mining che", ".; Wang, Y.; Bryant, S. H. Developing and validating + predictive decision tree\nmodels from mining chemical structural fingerprints + and high\u2013throughput screening\ndata in PubChem. BMC Bioinformatics 2008, + 9, 401.\n(52) Plumb, G.; Al-Shedivat, M.; Cabrera, \u00b4A. A.; Perer, A.; Xing, + E.; Talwalkar, A. Regu-\n29\nlarizing black-box models for improved interpretability. + Advances in Neural Informa-\ntion Processing Systems 2020, 33, 10526\u201310536.\n(53) + Shao, X.; Skryagin, A.; Stammer, W.; Schramowski, P.; Kersting, K. Right for + bet-\nter reasons: Training differentiable models by constraining their influence + functions.\nProceedings of the AAAI Conference on Artificial Intelligence. 2021; + pp 9533\u20139540.\n(54) Ouyang, R.; Curtarolo, S.; Ahmetcik, E.; Scheffler, + M.; Ghiringhelli, L. M. SISSO: A\ncompressed-sensing method for identifying + the best low-dimensional descriptor in an\nimmensity of offered candidates. + Physical Review Materials 2018, 2, 083802.\n(55) Lipton, Z. C. The mythos of + model interpretability: In machine learning, the concept\nof interpretability + is both important and slippery. Queue 2018, 16, 31\u201357.\n(56) Harren, T.; + Matter, H.; Hessler, G.; Rarey, M.; Grebner, C. Interpretation of structure\u2013\nactivity + relationships in real-world drug design data sets using explainable artificial\nintelligence. + Journal of Chemical Information and Modeling 2022, 62, 447\u2013462.\n(57) Sheridan, + R. P. Interpretation of QSAR Models by Coloring Atoms According to\nChanges + in Predicted Activity: How Robust Is It? Journal of Chemical Information\nand + Modeling 2019, 59, 1324\u20131337.\n(58) Rasmussen, M. H.; Christensen, D. S.; + Jensen, J. H. Do machines dream of atoms?\nCrippen\u2019s logP as a quantitative + molecular benchmark for explainable AI heatmaps.\n2022,\n(59) Smilkov, D.; Thorat, + N.; Kim, B.; Vi\u00b4egas, F.; Wattenberg, M. SmoothGrad: removing\nnoise by + adding noise. 2017; https://arxiv.org/abs/1706.03825.\n(60) McCloskey, K.; Taly, + A.; Monti, F.; Brenner, M. P.; Colwell, L. Using Attribution\nto Decode Dataset + Bias in Neural Network Models for Chemistry. Proceedings of the\n30\nNational + Academy of Sciences of the United States of America 2018, 116, 11624\u2013\n11629.\n(61) + Bach, S.; Binder, A.; Montavon, G.; Klauschen, F.; M\u00a8uller, K.-R.; Samek, + W. On\npixel-wise explanations for non-linear classifier decisions by layer-wise + relevance prop-\nagation. PloS one 2015, 10, e0130140.\n(62) Sundararajan, M.; + Taly, A.; Yan, Q. Axiomatic attribution for deep networks. Inter-\nnational + Conference on Machine Learning. 2017; pp 3319\u20133328.\n(63) Zhou, B.; Khosla, + A.; Lapedriza, A.; Oliva, A.; Torralba, A. Learning Deep Features\nfor Discriminative + Localization. 2015; https://arxiv.org/abs/1512.04150.\n(64) Selvaraju, R. R.; + Cogswell, M.; Das, A.; Vedantam, R.; Parikh, D.; Batra, D. Grad-\nCAM: Visual + Explanations from Deep Networks via Gradient-Based Localization. In-\nternational + Journal of Computer Vision 2019, 128, 336\u2013359.\n(65) Smilkov, D.; Thorat, + N.; Kim, B.; Vi\u00b4egas, F.; Wattenberg, M. Smoothgrad: removing\nnoise by + add", "(65) Smilkov, D.; Thorat, N.; Kim, B.; Vi\u00b4egas, F.; Wattenberg, + M. Smoothgrad: removing\nnoise by adding noise. arXiv preprint arXiv:1706.03825 + 2017,\n(66) Pope, P.; Kolouri, S.; Rostrami, M.; Martin, C.; Hoffmann, H. Discovering + Molec-\nular Functional Groups Using Graph Convolutional Neural Networks. 2018; + https:\n//arxiv.org/abs/1812.00265.\n(67) Jim\u00b4enez-Luna, J.; Skalic, M.; + Weskamp, N.; Schneider, G. Coloring molecules with ex-\nplainable artificial + intelligence for preclinical relevance assessment. Journal of Chem-\nical Information + and Modeling 2021, 61, 1083\u20131094.\n(68) Sanchez-Lengeling, B.; Wei, J.; + Lee, B.; Reif, E.; Wang, P. Y.; Qian, W. W.; Mc-\nCloskey, K.; Colwell, L.; + Wiltschko, A. Evaluating Attribution for Graph Neural\nNetworks. Proceedings + of the 34th International Conference on Neural Information\nProcessing Systems. + Red Hook, NY, USA, 2020.\n31\n(69) Ying, R.; Bourgeois, D.; You, J.; Zitnik, + M.; Leskovec, J. GNNExplainer: Generating\nExplanations for Graph Neural Networks. + Advances in neural information processing\nsystems 2019, 32, 9240\u20139251.\n(70) + Rao, J.; Zheng, S.; Yang, Y. Quantitative Evaluation of Explainable Graph Neural\nNetworks + for Molecular Property Prediction. arXiv preprint arXiv:2107.04119 2021,\n(71) + Yuan, H.; Yu, H.; Wang, J.; Li, K.; Ji, S. On Explainability of Graph Neural + Net-\nworks via Subgraph Explorations. Proceedings of the 38th International + Conference\non Machine Learning. 2021; pp 12241\u201312252.\n(72) Shrikumar, + A.; Greenside, P.; Kundaje, A. Learning Important Features Through\nPropagating + Activation Differences. 2017,\n(73) Montavon, G.; Binder, A.; Lapuschkin, S.; + Samek, W.; M\u00a8uller, K. R. Layer-Wise\nRelevance Propagation: An Overview. + Lecture Notes in Computer Science (including\nsubseries Lecture Notes in Artificial + Intelligence and Lecture Notes in Bioinformatics)\n2019, 11700 LNCS, 193\u2013209.\n(74) + Baldassarre, F.; Azizpour, H. Explainability Techniques for Graph Convolutional + Net-\nworks. 2019; https://arxiv.org/abs/1905.13686.\n(75) Hochuli, J.; Helbling, + A.; Skaist, T.; Ragoza, M.; Koes, D. R. Visualizing convolutional\nneural network + protein-ligand scoring. Journal of Molecular Graphics and Modelling\n2018, 84, + 96\u2013108.\n(76) Rodr\u00b4\u0131guez-P\u00b4erez, R.; Bajorath, J. Interpretation + of Compound Activity Predictions\nfrom Complex Machine Learning Models Using + Local Approximations and Shapley\nValues. Journal of Medicinal Chemistry 2020, + 63, 8761\u20138777, PMID: 31512867.\n(77) Wojtuch, A.; Jankowski, R.; Podlewska, + S. How can SHAP values help to shape\n32\nmetabolic stability of chemical compounds? + Journal of Cheminformatics 2021, 13,\n1\u201320.\n(78) Mastropietro, A.; Pasculli, + G.; Feldmann, C.; Rodr\u00b4\u0131guez-P\u00b4erez, R.; Bajorath, J. Edge-\nSHAPer: + Bond-Centric Shapley Value-Based Explanation Method for Graph Neural\nNetworks. + iScience 2022, 25, 105043.\n(79) White, A. D. Deep learning for molecules and + materials. Living Journal of Computa-\ntional Molecular Science 2022, 3.\n(80) + \u02d8Strumbelj, E.; Kononenko, I. Explaining prediction models and individ", + "lecular Science 2022, 3.\n(80) \u02d8Strumbelj, E.; Kononenko, I. Explaining + prediction models and individual predictions\nwith feature contributions. Knowledge + and Information Systems 2014, 41, 647\u2013665.\n(81) Erhan, D.; Bengio, Y.; + Courville, A.; Vincent, P. Visualizing Higher-Layer Features of\na Deep Network. + Technical Report, Univerist\u00b4e de Montr\u00b4eal 2009,\n(82) Weber, J. K.; + Morrone, J. A.; Bagchi, S.; Pabon, J. D.; gu Kang, S.; Zhang, L.;\nCornell, + W. D. Simplified, interpretable graph convolutional neural networks for small\nmolecule + activity prediction. Journal of Computer-Aided Molecular Design 2022, 36,\n391\u2013404.\n(83) + Riniker, S.; Landrum, G. A. Similarity maps - A visualization strategy for molecular\nfingerprints + and machine-learning methods. Journal of Cheminformatics 2013, 5, 1\u20137.\n(84) + Humer, C.; Heberle, H.; Montanari, F.; Wolf, T.; Huber, F.; Henderson, R.; Hein-\nrich, + J.; Streit, M. ChemInformatics Model Explorer (CIME): exploratory analysis of\nchemical + model explanations. Journal of Cheminformatics 2022, 14, 1\u201314.\n(85) McGrath, + T.; Kapishnikov, A.; Toma\u02c7sev, N.; Pearce, A.; Wattenberg, M.; Hass-\nabis, + D.; Kim, B.; Paquet, U.; Kramnik, V. Acquisition of chess knowledge in Al-\nphaZero. + Proceedings of the National Academy of Sciences 2022, 119, e2206625119.\n33\n(86) + Bajusz, D.; R\u00b4acz, A.; H\u00b4eberger, K. Why is Tanimoto index an appropriate + choice for\nfingerprint-based similarity calculations? Journal of Cheminformatics + 2015, 7, 1\u201313.\n(87) Huang, Q.; Yamada, M.; Tian, Y.; Singh, D.; Yin, D.; + Chang, Y. GraphLIME:\nLocal Interpretable Model Explanations for Graph Neural + Networks. CoRR 2020,\nabs/2001.06216.\n(88) Sokol, K.; Flach, P. A. LIMEtree: + Interactively Customisable Explanations Based on\nLocal Surrogate Multi-output + Regression Trees. CoRR 2020, abs/2005.01427.\n(89) Whitmore, L. S.; George, + A.; Hudson, C. M. Mapping chemical performance on molec-\nular structures using + locally interpretable explanations. 2016; https://arxiv.org/\nabs/1611.07443.\n(90) + Mehdi, S.; Tiwary, P. Thermodynamics of Interpretation. 2022,\n(91) H\u00a8ofler, + M. Causal inference based on counterfactuals. BMC Medical Research Method-\nology + 2005, 5, 1\u201312.\n(92) Woodward, J.; Hitchcock, C. Explanatory Generalizations, + Part I: A Counterfactual\nAccount. No\u02c6us 2003, 37, 1\u201324.\n(93) Frisch, + M. F. Theories, models, and explanation; University of California, Berkeley,\n1998.\n(94) + Reutlinger, A. Is There A Monist Theory of Causal and Non-Causal Explanations?\nThe + Counterfactual Theory of Scientific Explanation. Philosophy of Science 2016, + 83,\n733\u2013745.\n(95) Lewis, D. Causation. The journal of philosophy 1974, + 70, 556\u2013567.\n(96) Tanimoto, T. T. Elementary mathematical theory of classification + and prediction.\nInternal IBM Technical Report 1958,\n34\n(97) Rogers, D.; Hahn, + M. Extended-Connectivity Fingerprints. Journal of Chemical In-\nformation and + Modeling 2010, 50, 742\u2013754, PMID: 20426451.\n(98) Mohapatra, S.; An, J.; + G\u00b4omez-Bombarelli, R. Chemistry-informed macromolecule\ngraph representation + for simila", ", S.; An, J.; G\u00b4omez-Bombarelli, R. Chemistry-informed macromolecule\ngraph + representation for similarity computation, unsupervised and supervised learn-\ning. + Machine Learning: Science and Technology 2022, 3, 015028.\n(99) Doshi-Velez, + F.; Kortz, M.; Budish, R.; Bavitz, C.; Gershman, S.; O\u2019Brien, D.;\nScott, + K.; Schieber, S.; Waldo, J.; Weinberger, D.; Weller, A.; Wood, A. Account-\nability + of AI Under the Law: The Role of Explanation. SSRN Electronic Journal\n2017,\n(100) + Wachter, S.; Mittelstadt, B.; Russell, C. Counterfactual explanations without + opening\nthe black box: Automated decisions and the GDPR. Harv. JL & Tech. 2017, + 31, 841.\n(101) Jim\u00b4enez-Luna, J.; Grisoni, F.; Schneider, G. Drug discovery + with explainable artificial\nintelligence. Nature Machine Intelligence 2020 + 2:10 2020, 2, 573\u2013584.\n(102) Fu, T.; Gao, W.; Xiao, C.; Yasonik, J.; Coley, + C. W.; Sun, J. Differentiable Scaffold-\ning Tree for Molecule Optimization. + International Conference on Learning Represen-\ntations. 2022.\n(103) Shen, + C.; Krenn, M.; Eppel, S.; Aspuru-Guzik, A. Deep molecular dreaming: inverse\nmachine + learning for de-novo molecular design and interpretability with surjective\nrepresentations. + Machine Learning: Science and Technology 2021, 2, 03LT02.\n(104) Lucic,\nA.;\nter\nHoeve,\nM.;\nTolomei,\nG.;\nRijke,\nM.;\nSilvestri,\nF.\nCF-\nGNNExplainer:\nCounterfactual + Explanations for Graph Neural Networks. arXiv\npreprint arXiv:2102.03322 2021,\n(105) + Numeroso, D.; Bacciu, D. Explaining Deep Graph Networks with Molecular Counter-\nfactuals. + arXiv preprint arXiv:2011.05134 2020,\n35\n(106) Krenn, M.; H\u00a8ase, F.; + Nigam, A.; Friederich, P.; Aspuru-Guzik, A. Self-Referencing\nEmbedded Strings + (SELFIES): A 100% robust molecular string representation. Ma-\nchine Learning: + Science and Technology 2020, 1, 045024.\n(107) Nigam, A.; Pollice, R.; Krenn, + M.; dos Passos Gomes, G.; Aspuru-Guzik, A. Beyond\ngenerative models: superfast + traversal, optimization, novelty, exploration and discov-\nery (STONED) algorithm + for molecules using SELFIES. Chemical science 2021, 12,\n7079\u20137090.\n(108) + Kim, S.; Chen, J.; Cheng, T.; Gindulyte, A.; He, J.; He, S.; Li, Q.; Shoemaker, + B. A.;\nThiessen, P. A.; Yu, B.; Zaslavsky, L.; Zhang, J.; Bolton, E. E. PubChem + in 2021:\nnew data content and improved web interfaces. Nucleic Acids Research + 2020, 49,\nD1388\u2013D1395.\n(109) Tolomei, G.; Silvestri, F.; Haines, A.; + Lalmas, M. Interpretable predictions of tree-\nbased ensembles via actionable + feature tweaking. Proceedings of the 23rd ACM\nSIGKDD international conference + on knowledge discovery and data mining. 2017;\npp 465\u2013474.\n(110) Freiesleben, + T. The intriguing relation between counterfactual explanations and ad-\nversarial + examples. Minds and Machines 2022, 32, 77\u2013109.\n(111) Grabocka, J.; Schilling, + N.; Wistuba, M.; Schmidt-Thieme, L. Learning time-series\nshapelets. Proceedings + of the 20th ACM SIGKDD international conference on Knowl-\nedge discovery and + data mining. 2014; pp 392\u2013401.\n(112) Kenny, P. W.; Sadowski, J. Structure + modification in ch", "ery and data mining. 2014; pp 392\u2013401.\n(112) Kenny, + P. W.; Sadowski, J. Structure modification in chemical databases. Chemoin-\nformatics + in drug discovery 2005, 271\u2013285.\n(113) Tyrchan, C.; Evertsson, E. Matched + Molecular Pair Analysis in Short: Algorithms,\nApplications and Limitations. + Computational and Structural Biotechnology Journal\n2017, 15, 86\u201390.\n36\n(114) + Griffen, E.; Leach, A. G.; Robb, G. R.; Warner, D. J. Matched Molecular Pairs + as\na Medicinal Chemistry Tool. Journal of Medicinal Chemistry 2011, 54, 7739\u20137750,\nPMID: + 21936582.\n(115) He, J.; Nittinger, E.; Tyrchan, C.; Czechtizky, W.; Patronov, + A.; Bjerrum, E. J.;\nEngkvist, O. Transformer-based molecular optimization beyond + matched molecular\npairs. Journal of cheminformatics 2022, 14, 1\u201314.\n(116) + Park, J.; Sung, G.; Lee, S.; Kang, S.; Park, C. ACGCN: Graph Convolutional Networks\nfor + Activity Cliff Prediction between Matched Molecular Pairs. Journal of Chemical\nInformation + and Modeling 2022,\n(117) Langdon, S. R.; Ertl, P.; Brown, N. Bioisosteric Replacement + and Scaffold Hopping\nin Lead Generation and Optimization. Molecular Informatics + 2010, 29, 366\u2013385.\n(118) Turk, S.; Merget, B.; Rippmann, F.; Fulle, S. + Coupling Matched Molecular Pairs\nwith Machine Learning for Virtual Compound + Optimization. Journal of Chemical\nInformation and Modeling 2017, 57, 3079\u20133085, + PMID: 29131617.\n(119) van Tilborg, D.; Alenicheva, A.; Grisoni, F. Exposing + the limitations of molecular\nmachine learning with activity cliffs. 2022,\n(120) + Fischer, H.; Gottschlich, R.; Seelig, A. Blood-brain barrier permeation: molecular\nparameters + governing passive diffusion. The Journal of membrane biology 1998, 165,\n201\u2013211.\n(121) + Liu, L.; Zhang, L.; Feng, H.; Li, S.; Liu, M.; Zhao, J.; Liu, H. Prediction + of the\nBlood\u2013Brain Barrier (BBB) Permeability of Chemicals Based on Machine-Learning\nand + Ensemble Methods. Chemical Research in Toxicology 2021, 34, 1456\u20131467, + PMID:\n34047182.\n37\n(122) Wu, Z.; Ramsundar, B.; Feinberg, E. N.; Gomes, J.; + Geniesse, C.; Pappu, A. S.;\nLeswing, K.; Pande, V. MoleculeNet: a benchmark + for molecular machine learning.\nChemical science 2018, 9, 513\u2013530.\n(123) + Ho, T. K. Random decision forests. Proceedings of 3rd international conference + on\ndocument analysis and recognition. 1995; pp 278\u2013282.\n(124) Martins, + I. F.; Teixeira, A. L.; Pinheiro, L.; Falcao, A. O. A Bayesian approach to in\nsilico + blood-brain barrier penetration modeling. Journal of chemical information and\nmodeling + 2012, 52, 1686\u20131697.\n(125) Pedregosa, F. et al. Scikit-learn: Machine + Learning in Python. Journal of Machine\nLearning Research 2011, 12, 2825\u20132830.\n(126) + Moriwaki, H.; Tian, Y.-S.; Kawashita, N.; Takagi, T. Mordred: a molecular descriptor\ncalculator. + Journal of cheminformatics 2018, 10, 1\u201314.\n(127) Chollet, F., et al. Keras. + https://keras.io, 2015.\n(128) Wager, T. T.; Chandrasekaran, R. Y.; Hou, X.; + Troutman, M. D.; Verhoest, P. R.; Vil-\nlalobos, A.; Will, Y. Defining Desirable + Central Nervous System Drug Space through\nthe Alignment of Mo", "lobos, A.; + Will, Y. Defining Desirable Central Nervous System Drug Space through\nthe Alignment + of Molecular Properties, in Vitro ADME, and Safety Attributes. ACS\nChemical + Neuroscience 2010, 1, 420\u2013434.\n(129) Ghose, A. K.; Herbertz, T.; Hudkins, + R. L.; Dorsey, B. D.; Mallamo, J. P. Knowledge-\nBased, Central Nervous System + (CNS) Lead Selection and Lead Optimization for CNS\nDrug Discovery. ACS Chemical + Neuroscience 2012, 3, 50\u201368.\n(130) Polishchuk, P.; Tinkov, O.; Khristova, + T.; Ognichenko, L.; Kosinskaya, A.; Varnek, A.;\nKuz\u2019min, V. Structural + and Physico-Chemical Interpretation (SPCI) of QSAR Mod-\nels and Its Comparison + with Matched Molecular Pair Analysis. Journal of Chemical\nInformation and Modeling + 2016, 56, 1455\u20131469.\n38\n(131) Hassan, M.; Brown, R. D.; Varma-O\u2019Brien, + S.; Rogers, D. Cheminformatics analysis\nand learning in a data pipelining environment. + Molecular diversity 2006, 10, 283\u2013299.\n(132) Schomburg, K.; Ehrlich, H. + C.; Stierand, K.; Rarey, M. From structure diagrams to\nvisual chemical patterns. + Journal of Chemical Information and Modeling 2010, 50,\n1529\u20131535.\n(133) + Sheikholeslamzadeh, E.; Rohani, S. Solubility prediction of pharmaceutical and + chem-\nical compounds in pure and mixed solvents using predictive models. Industrial + &\nengineering chemistry research 2012, 51, 464\u2013473.\n(134) Boobier, S.; + Hose, D. R.; Blacker, A. J.; Nguyen, B. N. Machine learning with physic-\nochemical + relationships: solubility prediction in organic solvents and water. Nature\nCommunications + 2020 11:1 2020, 11, 1\u201310.\n(135) Loschen, C.; Klamt, A. Solubility prediction, + solvate and cocrystal screening as tools\nfor rational crystal engineering. + Journal of Pharmacy and Pharmacology 2015, 67,\n803\u2013811.\n(136) Diorazio, + L. J.; Hose, D. R.; Adlington, N. K. Toward a more holistic framework for\nsolvent + selection. Organic Process Research & Development 2016, 20, 760\u2013773.\n(137) + Sorkun, M. C.; Khetan, A.; Er, S. AqSolDB, a curated reference set of aqueous + sol-\nubility and 2D descriptors for a diverse set of compounds. Scientific + data 2019, 6,\n1\u20138.\n(138) Durant, J. L.; Leland, B. A.; Henry, D. R.; + Nourse, J. G. Reoptimization of MDL\nkeys for use in drug discovery. Journal + of chemical information and computer sciences\n2002, 42, 1273\u20131280.\n(139) + National Center for Biotechnology Information, PubChem Compound Summary for\n39\nCID + 1549018, Jasmone. https://pubchem.ncbi.nlm.nih.gov/compound/Jasmone,\nAccessed + September 26, 2022.\n(140) Sell, C. S. On the unpredictability of odor. Angewandte + Chemie International Edition\n2006, 45, 6254\u20136261.\n(141) Genva, M.; Kenne + Kemene, T.; Deleu, M.; Lins, L.; Fauconnier, M.-L. Is It Possible\nto Predict + the Odor of a Molecule on the Basis of its Structure? International journal\nof + molecular sciences 2019, 20, 3018.\n(142) Rowe, D. Aroma chemicals for savory + flavors. Perfumer and Flavorist 1998, 23, 9\u201318.\n(143) Mallia, S.; Escher, + F.; Schlichtherle-Cerny, H. Aroma-active compounds of butter: a\nreview. European + Food Research and Technology 2008, 226, 315", " H. Aroma-active compounds of + butter: a\nreview. European Food Research and Technology 2008, 226, 315\u2013325.\n(144) + Jelen, H.; Gracka, A. Characterization of aroma compounds: Structure, physico-\nchemical + and sensory properties. Flavour: From food to perception 2016, 126\u2013153.\n(145) + Licon, C. C.;\nBosc, G.;\nSabri, M.;\nMantel, M.;\nFournel, A.;\nBushdid, C.;\nGolebiowski, + J.; Robardet, C.; Plantevit, M.; Kaytoue, M., et al. Chemical features\nmining + provides new descriptive structure-odor relationships. PLoS computational bi-\nology + 2019, 15, e1006945.\n(146) Mostafa, S.; Wang, Y.; Zeng, W.; Jin, B. Floral Scents + and Fruit Aromas: Functions,\nCompositions, Biosynthesis, and Regulation. Frontiers + in plant science 2022, 13.\n(147) Tokitomo, Y.; Steinhaus, M.; B\u00a8uttner, + A.; Schieberle, P. Odor-active constituents in\nfresh pineapple (Ananas comosus + [L.] Merr.) by quantitative and sensory evaluation.\nBioscience, Biotechnology, + and Biochemistry 2005, 69, 1323\u20131330.\n(148) Wei, C.-B.; Liu, S.-H.; Liu, + Y.-G.; Lv, L.-L.; Yang, W.-X.; Sun, G.-M. Characteristic\naroma compounds from + different pineapple parts. Molecules 2011, 16, 5104\u20135112.\n40\n(149) Stumpfe, + D.; Bajorath, J. Exploring Activity Cliffs in Medicinal Chemistry. Journal\nof + Medicinal Chemistry 2012, 55, 2932\u20132942, PMID: 22236250.\n41\n"], "model": + "text-embedding-3-small", "encoding_format": "base64"}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "26417" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/embeddings + response: + body: + string: !!binary | + H4sIAAAAAAAAA1RZyw6qzJqd91Ps/FM6kZtU1ZlxExGQQkDEHgGigiJyqQKqX76D+6Q7PXGARFOw + vnX7/vs//vz5p83rshj/+deff97VMP7zn+u1WzZm//zrz3/9x58/f/789+/z/91ZNnl5u1Wfx+/2 + 35fV51bO//zrD/+/V/7vpn/9+UfuU5n6X95gjNyhBU/PEWG7lhLGbwKmw2NyD6k3zHLNZL324eHj + 7XF52mFD3B2lBdVOq1JX7oJ+sREiYFHTB3ZQAKI5dIwS5Kyz8XE8sHo+uF4DLrEzYFu8dIAJ+PRA + cFBe/qJXCZhvEVfAa8YlPlfsPsZLFoYY8Edzxpf7u8oZ3CUJPI9njQY3wwXdzYggGkx/8fnFV3Mh + GkEHorT+YO8RiWy0QW2h0FdmamhfkbX6uwthn+IQ273zctle/ihIZW5NgGeYNX+/miF6HgZEJPUz + gIVoVwLGj7OjybnsamYphwoabzhi4/EZ+hYesgYKW2bQQ/7SchFxZQz28zHFRXJxXf7SqyVisrGj + 5hsdI8nfqimUn6cH9njzzZbZXmwEsIqwUSdztMSalsJPQhvq9n7LWM75JoLzR6KafS2AyFV+A6L5 + 3mPs+l+39XO/glu/iOl9T0Z3+S7BAswjzcnsENOVXsjjYPo1MQ2efAXo5rYVYV0+LljjBM8Qmenx + cEPPMTbiQ8fmnX0qkKcOPC4TIwbSW753MM+Iif1D+oqWzxlDCO+UEglqustjjiPQYPCGjwN8gc7o + vw6qo5fps1d8yYUX3Znodx694pJcCMvMR7cGayRzPsilnb14ENV3jnT6AeQsx5qH0u/WoZ5Q0Xp+ + WZ8G9cZDxWXER0B6taIPH+Hs0JhtUyapyVaGK7586Tpt3CXZ2gnYZK1Ob9/CZvyg2jZ8Tu1Mg+x5 + 7ifz9g3g27IRxTGyXAmOgIceJBG9grGvacyWCsjv6kSdrH/mM+7VB9L3gYv1hb6jxcSnDKoepj5K + qOWK3D4tkbCXoQ83ZW4IyB09aDuHnJbLnY/YJ4UeNE/cEesK1vIFNUCB8nBk2NJG0LPbXvURvb/f + NDOO+349jwK5uHaotvdbl2SAFXBDqitVN2Qw+lN+0WEo8G8aVlwSTcn8EZH+8DE1rvEm745fK4PD + 5If0mH3v9cDP9wYGl9Aj8i3fR6zs5BRedrGGT8trayzXIVPhHro91ud+a7DOXnz4m+dbnYn11LuF + CInAf3z51ooutU1hgFV7veN9aMR5G1QNBzS6QT5ohxCMY/J8oMGPHXosceWKIGscODSAx0drbvO5 + CnQPfZN7TODzUruCFAATpC7IKN6To7GkBv+Atw3a4DibrF6gJ95D11rJsKdEVs/fp4agRpIB3huH + ijE1mRX0TBUda7bnu6LXtSoyOyXA2nJxckFJowq5qL9gY8Ujf94HAXJbuMPW9xC56/vxgUojF+/L + +BPNMMgqqCeWiHcqMFx+oEcV0rb+kjYanFxqtykBanKxCdAMr+b327SE67zho6nXPdMdt4EIcyq9 + p9KmX+Jz4KAtCCm25+vbYHlTd4h9vQzrk/rol6aTE6iMtohPOzQac6B8OQim7Q7vg3dX00solIgt + myv2wU0xiBodW8h/gYddzRh66YK/C3wq0wEflV6rJbB8HWS3B9d/Lq+r++H8IIMX2TBwsodFtMRp + zwN41Y/Y0/J9z2tjHaLZ6TR61uwZLLvjZoI7W1ZpwdRrNBVfK0Rlv7njw76p6zHw5QnBmBUEniOO + DZfeLuA9BBG2WBqB7poaIRC3rUKxFdTRX3yV6c6m4dxfXZqRV4ucEgo0zb0mvx/TWkHXhe6omm82 + eZfSkUCfnjyaAS2L+HxXWtASr0e8FzYYCMfKKtEVijd6KKpPNF63DxE2361LL/yzcecx+Vbwk8EL + Dr1TxKZDmmdAZFNEPbv1I7LkNIDpokZ0vzdgPhkPyMG6Sin1b4lu8BFXOnDTRS02UJABQdROHUTZ + U6HXPQujxfe4FuA8HMh8dCKwDL6tQrR8DXw/3RxDjM+BDbe3YEOTDD97grgkgR/af6lqGDmYr+/M + giMWJHo4CmY/nFzvoaSWzqglfGI2V59Xi/iX/sYWf3CNucOaidhpH658iPuhNfgU3S/chmytc9sz + /dnoUDh9AY0xuuUCazkLxuPdIfPlGvcC9R8PFI83h4At6Ou5tg8e/OmbLAlxP4/E5WD/iHPqhZ26 + Pu/EhMn9aNPjzT0DRqSyAttD2tJdp0cu39hfgjoFV0S+gj5nJ/Uhw7frf6naLHa9sDR6IUZ3GlYV + dHYFqT1X8GTPW39pyiXqOPGWwPtk2jgjyRfQ6HU14cUzasKOyoEJ2UGP0dZ4aPT6Vb8GqTlLh0av + XqkuFpM7IXMMYNDbNS5L3mWiWDxi9DioLo42b93o/EPHQVp9dRpeOC4fFORN0LmCA/Zlcs6l3XuZ + 0N7MEI1tw3SXSzG9UJGcHVxchBjweitw0InVit5cWDCpmjY+YLnfkcU7McAuH88Dqx5SZ9Uvtos3 + BLT8+0ivMteAz+bQWzCRkhP1mLqNiCNAAvWtHeJcGrp+IPj5Qrbj5tjgnmfA4K5MoNC9OXwLXQSG + wch5QHyFYYPF1JiRfa3A++WM9KhW+5p3odvA930LyMa6ha5IUt6BV/RNscepc0SkQZtgk5sPehEW + zRAvrlCh6MnvaKzWrTG/6NGEZ/G0owEJX/UyBcwGt214I7M5LO7ilFkI1eRs+yB9WpEY3U42RDux + XfmSuYPx4CF0FuHoc8s9zpmzfXBo6Q4lvs7zw2VwBCKQOmLhw84c+3mjBhV4PDlMDzdWRbPHJT4k + g1LTg9Bfe3ZK3RhKxZDQ6/R4u+wV0wKiK9mQdsXXHJahD0VPvFOsXPp8TuTPBKtP81r9qQjo/jZP + aBTgRDbKxY2G4q28oKuqOs6CQq2XBMY6oqliYH+7POtl2qYtOBMXUn1nju7MTlGHWo8vaaxrM5hT + 7SmDdT4x9q6IzaueAf1oLzQXHL4nU3t7QWwcXmQafTkf1nlBYsdjap6XB5uEbvKQY+kL1oD9jMjF + /orwVaY+2a74mL26tcBejExsLAZvzIo6W4jf+44vZOY5X8hUNtBxzRL7jbbUU8QlDpSNzsF4D3Kw + zDdbAcJZHlY+WqI5uW1baF2aitrpTfvx8wKcq+X7jTaCeh67KYD6Rt8SyJs7RjWEFhAJQ4FvB/UQ + CdF1UGCryS7O1LIC7GOOJSwjrcAO5aN+3ke7BqL4G9HgMSW16AuFAwfs7vz5aZ/qCU1lATSvVKkT + pbwxvWEkIzOcbRo0m0suTMKQwVsfjtgXw3c9QY8E0KuXgB41VvXT+bhp4eoXfN64g34+K4cMbMN4 + oG7hzy6bP/EDXFWeozhpufq7JCqE3+IC/NeSU4OwNG9g+lJr8mmDyGXVztcVQ9+e6D40+HzpTzmB + tmgDfDEOFZgvVy/evjx3pPbG1sDMtrMDH2L/xD+/RJznyP3wS03RvoJFqPcEyM3Z83mMUNRd+FyF + c93syBxLXD0Ls2xDyxhcWnzNkc1c7reKZN/f1FMdr5b4/hlCONI79qV61fvk2cA6akxq0I+Zj0fx + 7AEEp5zeDkfZJaNHMpifu5RqGz1jkznLOhiA8/Y5ky/cxUuuAYzAx/C3XbkAktI3ARF4GzTUTjmb + a7eQFfMEjzS0Zwmwju9TKDC5wCeHvAwaFFOLuB25Y1fMcS/276qA7iwfKD4g35jNet8Ao5lbWvbh + WC8NMVWQyqPgf+x9yxi/kXUYH16MCLmt9dLDbzulSC6OL20HpV+IdhqQd54eOPzQMSIW0xpU9vbL + lwULg/Hdzyn8svaw5s0asGqSPCA/yrf/wR+zXlD98BFXD5g+bY+4My9fHvAcZ5yPNFbVC419ESqH + 4E5A4Z/cqbGfA5qXdMJn/eoBMb/XBHV8DKi3xVk+aYnKo/DCCzT3l0O+jA9kw5UvsPbetj0xw9GC + rgPP9Lozj/XEz5cGJaMvY+v9vRiTvt3qkHx4iYbuo6gnXw0yQK6qR41Z3OWDBV8OWP2pD+AF5H/9 + QOxcCfnlWZGH7gTOZZWQyUkOxtKNfQmL/eNOy3Ue/s6n2ImYGodDCAhXWQ18SXGIj7uxcOfLmJsg + Oy9PQosuYKyFQwZnt4D0tNEV1tXKkgDh1AMiGcmpZ6v/gyu+scYzxR3naWNCLBg9ocSSo9nJGhUS + sa+oo5YVm5qH58BZrb7+6mf72b2ECci7S0em+/vJpu1uJ0Jd9+744tdpxAL7K8Nuc/J9bta6fhr6 + OUG7k2gSMcNaPVuWJMKfXh2yyMr5MBs5eL6WHbVX/zzVk+XANuP9lU/EFS/XAeTbivpCXm3caZcy + CC53c0P9iGeA8f03RHeBo1QPzJ0h7O5MBt2jeOGSTTRfbqoaoHbbbnF8ILAme722UOzkxJfuy8md + lrAYoJYHPVbPydNl3NevYLNI2JfVzR6w+wnGUKMI4WNCLYM5J7eUf/4GyYkSEc31OiC9u4yq56ms + J2S+A6DSk4v3xxBFoxQAC8aHhlGNCbE7v/ttCle809A6t+7CB1kDri20COrGGiwHXxahoOtn7I6d + 3C9JuwuQzoMn+fUd85avQyTsLQ3bzfXokvEhOEBLvA3pPcPsicpx/N9+Ye/Xaf5XPyJb0db50OsX + c6IBjjPd4D3mWTRag9ZBZfIjqi2vqzEXmpbAEQgXarzTup6e70yHZk9d7HNIdJcPepcQn9yY0FTa + 1Eu3O3JASXcOPjpl5C7HU88B9fw44pt8cOtVr0v4m1e3Jh7jq8dA4Ewe6V895BfuqEDP7fdYc8AC + 2jW/wTUvEy11esayPoBouNmCD5XIqtf3Rf6t721LDOZA2IGfP/v57+muPXn07owYa7oD2HQ+Sh0M + tk5EFCynrlCYZxUdrv7Fnx71BzA/qGSwPxRfHBhGzuYHHTtx45aIWi/3GM1H56WD4/uEsB/Iar7y + bwrN55v5G8eBgNRN/4IHNYyxW057V/LbY7wtNtGRbGmR1nT1g9DYqgcCOfWUi2UoZPDzGhOs+e2t + H58bbCnGnkP0IJQdI+fNhwDhoe8wVrWqHqnfVrA4nUrSbL4WY+VVSGDw8ALCZ0iI6IpXEGztiHyP + wqtfpAvQ4ZgyQP7yoQpqBxTGiazPfwDzl7weyD/mO3/tr9h8a4/eX3/gx+/emK7LQ0bnobKI+OO3 + SsYO/PHdry9b7rVcQhjBJ1Wzp9BPevzxgb7dHLF5CZxe2jZRBkbT+mINb+V6cF9fG9bXosdFV0n1 + dBgq59dfYHN2eGPUbuqEkl3h+C/Ja1y2eysLeElJiPdn8d3Pt0gs0JpXRsVMi+gWfIkMOavO8P43 + vzG+BrC+n0OsIi8xGKdoDRxHdcBHQXFq6RV/SrjHl4Bshtbv6RCRRBEyLqV4513cpVaUBD4OkoX1 + Y+Uz8fheCLpK00wkfSfW34fBl0j9Zi+fT/2pHjz3baLTxTEwnps4mnEyT9BoWOv/8hoPtlsLYkHr + f30p4O/nqYFxdFl88Yen33wl1+T5tz8YUMqrCEmEUucWb9iwcEcZ9I8kp9bVPUfU2T4gnNzPzq/5 + 8OySU3wxgb4Pf33TLl9i82ZCUad7fAzdG2Ppvl1gNN96ssmv5/6nnyiFsMHh5aUCMR6UBt5vUkPk + 3LPyhXNeNmq+wCXtyeGjRWeVgy67RMNuoemRNEQkhuXDO+FI0q6MSewwwLsAKd1jPsoX4xKmP7z8 + zavNr59Z/SO9of0czZjEPsrte43397ceTT//DYCQ+sLzotdCMSgVVIDd0CyTJ2O4lpkDH4oSktnP + fSCa+JpBxdln2OwAMwaxcRQ4O61Gw4euA/GjnCb0F9/+coh6fiOrMDzoDt3fN3c2Jdkxg6vf9hkK + FPZGUpkC5MgxtoIj7el7noLfPGHtAXQgxZVUKFhPvhjfQM/Ys9StX/9F1dao2XIp5Bd0a/lK76C1 + 8sWwLjEcmi1PvVCixpzc5g791c8o5d15KjUOfrlXgQ8XrmQTnGUT2KMn4d/zWbTqYwFw2Jzo2nez + 5RB6FVz9IQGKnUUs00Gx/eHloBRTP6iTCOGp1zPq3c8BE1f9gb++6XnoJDDv1hY4VJQTxuP20M/i + qPMoMvgzvauaXrPXdhvC1X//zSeMZY0HtYuh+JvJ7mphGesMvCX/QF0evRmjuTXBzZLOOJVZ1s+T + 8EqBh2qf+okRszUPKWh7yVKsfZy6pw/lKoJImWx6uSbPfvLpOQC3jwF90N5fLpOTLQSppTJqv+3F + mIPQLH59A/3hczrtZV5Z3//a1xGX3sqqhTPncj++rNnZiHVocMpE90ZyqqdFWXz0OSs9qadla1BN + 6Uz41RZ91ReV0dd2GwDB41rqTPLDWLJAD3/+ZtWTph7k8qDALP/a1JefTb/2by1ckK9Q55u+XHKv + 5b94IO+FOa54eZghXLRjTg1keIb0m4cVv1i3X193UUQkw/jzqumBw0/GiH1t4QTbEz33EXFn6X3g + wc9fG9d9UE9rXwKf/vLF+O41jJ7UVoaBvdvjfbe5sPleTBXaFqWJXS9gEXN51im/38PVUIF55SfY + R25DUIQdd9hvgwKu/gZbB7UF0/jxdPhi+yeBp4Z3pxPrVNj7RYRPv3wCg+wBTOUpUlsPKjB74VsF + 024ocbQrbIMVTzFWgMgGsiA4uNOgqg4YRGtLnTIdDCoAnaDjO0LU51BizKMeTvAJrxt/IqFZz4V2 + iOFGu7zJrDv5ry96wFNoXbDp31m/vPaqg359WxJkZTRtr/FrS4cpo9a5dPrJkC8ZbLnhjTPTHWvy + 80f6yXD8RxzybKnDQoTz/RxhfHZixrQmT6A1Vc2/9X9RFg9WQxus/b7QL42eiSCJAwH79TLXi0Rv + JXjL0sYXHtb+3/70KQw3uoP0DAbwrmL4BB/Th/fqXi94UQuUB6/Bv9+Ay3jvzkQIBG72e3/5RjNB + 7wyWjSlTZ/eV8hnKkw7X/IvdNY+QkiMvMGJJwuourfNpHywcCnqnJkIfEWMA7y5WVv9DjVTR2Jw9 + hgnuBJL/zffzz0/KfnPDvh2n7t+8JfD0gr1fvpynjQWDo8Pw7tiwfmqKqfybV/YqD+pRmCcbrucl + G8iA+9unKD8+vVfbuV7Kj+nAdV+AXR7t2MJMT4RrvsRGIx6j5cFOMtr1bx3/8ghbXHOBz2G60ogk + 33xxrwOEzdcMqN4XsjHYjziDzSJg7Eh7FQiWmpbKfnjsVvyV9WKVCg/OUWP4m6Qt+1FMeVOhT35P + L8cd6ad1Pwb3VTZi39pd3CWScQhdi76IkJ+naL4/TA/WatVS89vJxjLzfQLmJZuIIOAtm4/s8ADa + MQD/3l+t+0UYGrCka9/uLqeTkQEHcu+f3uWiwjoP/vY5Hng966XTlhAV9jnD6dqfSGXy7eBP//3B + eLEve3cZWP0i3okPHjC/PSY/PfDj1vNzMTjLROkv6eW3/6znY+WXv30CdT4Hw5jna9zCaugCIurz + B8zTkg6/fEh3jiXV07o/gbtiuVArK4Kcj4FeoNQ179i6Hz/RiEnhwYDfVjjj3w4TUf3w4DVQDeoP + AesXBJwOivuF4Z17Ceu//IfUCeJVr9z+aJ4d5Mqf3t9Oj53Bs+Mnhc+pm8njEEK2nNqRhyebbYmC + +28+ftiVh72gjdhe/QMN+MiEalA+fZpK936W4BDC/p3ucdSIY0Rvy5yib/9Uf/0p+15PRozWvEOt + 7Ltxh/M+DRTjAbS/+7FFBrUPc5onpLq8HmwWR4eXf35HffI6WPszHz7FOsX+a5L6ie4fL7D263Tv + J4dakLZBg268N1PD+cCaAcdtYa43kKoncWdI3Nd6wDUPY2wFRj6PnRzCSe0WetxmTT+zmMZwze84 + crIjmJPFzn7nw6Z8e0bzuk9A6/8Tya/laA6XB0RSX2u+/N1+jam/9w1YrNMehwvkDRZ+XhP8vS+n + UKpoXs8PJmdycKS6R3cSz3se7mecYnvt28XRazK47luwzh0URnNlSWHXnF1cFruPO6v2osPHIFf0 + fwAAAP//nJtL84KwFcW/S7d2RkAgYYmAyDsoiLgDRXmIvEwgmel37+C/3XXVJaPj4CX3nHN/Can1 + 1MFu9XOZa9Q33q79Owd58PrLD+/ZOORUz7AJf34egEfmTjHQCujjJsNykAU5z70UA27du0I0nS1s + 5XUFkNAJEfvY6vVXD7MC/ON3KuBf//w/ThTw//tEgfRoGDkP0bFeJL5+gUkuVLx8asbY1roXEGRP + SEwzb2qW2VICe0/MUOhkB31H2qRQCsl5EtX0e7CcDBYAlWuCYJztMOLLdrFgt9M+wYKffi2MWjDD + Ut1WKLjePEbP/SWDT3vjBVuHzfk0RWIMl4u+xfrlhRiNZO8Osw1U0IW9YjCLzs0D2r7eEkPIpZzW + d+0OtX25JW7zltkUglMBi0cmIo0etHyJjReG6uVeEz9/U8ZIKgRwFFUJXdDQMkpuUaH8Pt/r53e0 + 6Cc2KG8NSwHPAJ/TA+UqpZSthRTSUuiLQHwZCgdtCvg5EXSSFTaGVZcdiTkq8cjLWa2CD/5OKI24 + K+D9VkgUGqCAuNXpzqg1SjI4ia2OWWkcohk9HAsot+KK/IycGIudIgbvtJ2Ivh3yelo/VzrnEZH4 + mVU5tUMphVLLjSj/8jRa4p2aKabZZ8QD8OUuh2PlwaU4CXi74dEo4DtHlfkB9yQp98Ad9zvRhIGH + n2h/fKo152N3AroWGRjq8gvM22/QQSczGUKwPIFd7BQJhCPaoyOV9fFr5HIDT2KjE/WVbsESK3Ol + JO9xIRpJdMYPo0GVBNQNCdJdD/gXOkGl0rOeoDyrXL7H2FQW8H0ik0pBNG1qNEH+sO2Di530Lp3r + LIV5dc6Ia0BL55C1UAXXbor0zHJHThIsCnl8dQkK3q6+7O++AcOURPjtbwzAGwIzFSUxIpLUl5R9 + MbYN6JrdkYSHaweWfaV4srUsEgmkZeMyTpU1WeqqLynaas9ok/YWDKVaxacKiC5d9uWsrPUMFm1f + s++xCj0oytGV7DnnADjZ7zTFWpiEtDi46YK4bO8wteQ3yfnPV1/r84L9YzqioD2W+bx+H8rI81FO + xWDclTLSwPZj+yShhQ+WY96ZsNx6O+J3n0PEbjQzIdqMGC+Q7uvldacBfOg5JPqH1eOv35TI+chY + Tt4vhsM2aCHupRexpe1NnwVSO4qnjHfi6bLKxkQ1RJiLjo9p1D70xchpo+DdxUbpuU4iYcuVoSy9 + rZ7sU35bL/5ekGE3h0/kRP3HJapnvaD5ZgfkTBZw8WLdZLjbNypJvupJ54cm7uARCT3xvV0xUoi4 + BI6n6w2ZB6fWGVNGC2rcLscGhOdxx7Ebln797PKfr9sWTyGEdnDj0KMqp5ExAagQKnhG3sTcfPIe + 8aBU7XRAwatvataXc6Lo4UcLxOxycrnwFUKFnDIcLPr5kAubJSugYWY+lr+c4tJNctuAK96pWJ7n + BFB6szbwjrCG9E5QdV6OZBmeDuFn7Y/X+M0/hQWrb+HjJbcfOaftSwFOY3xBXt97ESeVYgZ1GtyJ + quXI5SYzNaEynA1ys+PY5THem1CtaRmIL7WP2CLHifIqd7uAZdY4Mnu8d8BRMoTlqP/otJrPAuTr + 2xn5qcrq7r7IiULbZkGXrsnzSY5TC77eCGDZMz5AOAv8oEwFvqEDFm+68O7DQAkelUd0u7yxWXzf + MNzO+QGFkucBahe3AS6NOyDzm59GIX66G1hwxo1c86vozv4namAucYg8JRrl9Or0pkLezROZqvIB + /eexVeFP783iq+hsMxwMoHJtgPyHFALibrMXWGbWEseTLbbI5rmD2vK8ksPa3+wtRJ4StZeS5Dvg + gHFzbwQlnYI9fncNyEezCGT4GS8dSfuxjISnoVCYqkWOtIshu8tH3FZQ+XYWKbwHi0juqxbkpMuX + pEXlRXP24AoovLkbSp/syxaEvA74m6wkjpSe62/lih0sAceIKi5eviBkDEp+T1sSX7tjvcs7a4Cn + qHmj5N0ngAuWc6XclrdCimdr6YukuwJ4Iu0ccKDgx+Xp6ZaS5lmM1lFLp+f+kcLx2SekqJmed23u + F/IoYRDw4Vljc+I/ZXiTdxwWDo6uTyg5q5Ads4L49vdZM3ehoVIFSRXwSavWHBf4Kgza2SCPMnT1 + xXQWVfnpo+pkb5d5j1KEFXEQyQXNH/kcm2elOldxAOTBzoVqPnOQ2JxInKu6j3Zp7Mzwennbgezu + x5weS3kD35egW/WwcCcJiiJUr9Yb+Z+tnfMSzDswbSsT6UGTrCew2kz5nI9L8Fp4Oi7ES+8wlpoL + ue9iI5rv23OjXBU4/q0vnvM/FHZ5pGKJK+8R3tT0v/5rAaue9vX5rlzP/gFvHRbmk5sdOCVNkoDc + uBJGnR2aMmwdzySqF2kj90gWTgGvy4xi9Zvq/M+P4AR0dOgzmJPE5FUoZ86RhF8/AXTeDTPM9/cv + enwaCwiRCjvYbOojOaX8c1wMpfaU0CgrdMsWEwjRcZrB+zhvSSySayS8XVWGwy7isACXPRDeRjoD + uuU6vPExB2Z3659hppKOeJK/B+xuhpriDOIL79JuPaSmNZpyDzQVXbB4c+ef3+n61kTWtaUj+633 + n/7cTiXJZ54Zd8VrD59gWNerkKiDBZXzhIj/RXudy7PIUcwX3gaAS9xxR9qiANr39EaqZX5yAr84 + BV7oc+Rc6qFLjepQKF0SyIGw9muXYkEDx7C7kuL0IjXzPy8OlOHRJ0ax3QNu6lSs6PkLBFDb1/Xc + m1WjrPkKcw4L6958Ly/wwk+KtGUowXxaVE1hapegv3oGBx/Dket4dPzuPvncTEcN6IW86rs31cuq + H2B93sR/8cM4JdttA9UxTkh4/qg5P3kDhJx0/SIVHAZ3Vm/5C27J5xpkV6HKGT2xVi77q4FUN2z1 + pQB3FQZvQ0OGw8Jx5k8RBy57y0XmYwf17/3wraCQyU90sF6dvpx2mfind4mZN+N092RZuUbSkcRC + fsux2FwDsOY9ZKShXnOCzhr4GJeCPNc8tYCmhAqiexawV2+A5VLNqrKNxyIA9raJKLqjAVS38x3p + TdDkH6EKHfiMRZP8rlnnnCyg9aBEOnxBtubtStECVw42r8Bw+UslatCL3iLm1K+o9/y1xcB1NJdo + aR7ni1xGDvTZvSP3DN9reuxGA4pFORMXaa96wdziwdG8PcmRXRp3WsMNXPuRmIe3Pi75q9XgIusP + pB5oH9HYcAsoWr2CHHc35UtmmwIEXFliwZXKnJaMswBxXzIJhQeO5hQmZ6hmrz5ol25ik0glCMze + tUnQViVb5G11V9hN9skTCd6442RbgHaQcwEswA5gbDWpomVMICodo3GtVwf5yzFG2qsjIz2lOAb+ + ng+CWeG6etCLTQdf+EGDXXcb6kVvPRWQd/sM5oU/16ufBPB0OH8wr4dhzvo4NcGaH4i9fVzAUt5C + GZanBaFL0qojD3EnQ1H8XIh+wryOJSBksAriChVbJDP6qK8NFHl0Io7EmItPxgWCbLNRMG9EE6Dh + HlnwxV1a5IlXlS3mqfy7JnZV72v2uhyb3/MKvqenHC3ly9ZAUKcqOhxFz+VWP5Z/eXLeO/647DK+ + kiRT3ZC4aPb5oj5xBuBHUIgePliNDeEiwNxtJGSHoc+ww4scePsDJQ53EXKyyPdYJtpJJ7muD/Ws + 32wNUEO5EhVuPiMNW7OBBU9ztC+ie0QrN08hvbkW0sbGdrnP20ihuJd44uyMuF7uou+ByfMGcjIu + JaODiCgIjPBCbqoBa2qNiwwdJUVEXfOEsCBNA9ImeyHT1dVcWKyTqLRWZSBHVKhOvv2WwvLTawFw + sgbMTYFnwA7TnbhUxOOiG1H6ly/NNT/OPtaxAsPIIo8xqcflwG4zeKr4QlzD/LjEyr4tPEwfPpAD + d8ixJZ0tZb1/lDzrwt29jp9kVa4tfp+sGizQLu/K009Ccqs2lT633XeCOtYGpCq3Tsf1kZoK9JeU + ZPGk1/xgHhJl1XukqsTX55ZcBAiVaSa5hSswudFjAg9eT/Cu+RDwW6/yhFqM9vePnFPiMk+Od3KE + W5633F0ezSF8J+KAmXOV2eR8WhM2wzUhTjzVI37PagJcAx6It3e1cd7KWIbH7/mMHCml45RnkQUh + VDHRk+UU0WEgFYStA4grbj75u/8oG8gP2yPmbmYzdsVrprD63v0A6IWrz3VMOMg/yoIYRVNG9FjS + DXQd1SXms97oVNnIGPaenBE72U76dB81DyA+3+GNy18j+lt/8KQ46FFf/Zopr2C9Hw0jddVrQcKJ + BSSg3rF0H5p86XFrwDQePGLOdATEHIkDp/15WdfjsV7OPR2UN74/yXnmVLAzi0CEQPBrglZ9ZW8h + 92RB6EzydIMZzB86DYr61VKC0rCOWKIaMkAgGpAqJiOgR8jusEs8GSWi3eq/9akEaWFieRg27hK7 + IacYvnVBRc3qfInkHkMpTmAgQXwd5+F66qADJh8FD9nUKTnXMSwcyf75IdttZiWAwePlEWf1H7Kp + 4w1Y64WC9/RidPJWv6t3FgZrnl2qeTaUHy/ZM8XJic/vDSjZN43sIS3HBWu3Cq7z2uqfe0C3n72q + mH5eBZvNTWK02jkbGN2+72DH5mb8hpxnyOZr2uItR3budBvvGBiWiTAvIJZ/wyfA8EQSFzdx0TEa + vTkTFA6w136HOVkusQxVKntYiZioz9r3loK9zx1IIR7uOiUvvVWaZT2hvN4/5zUelenY7YjW8A7b + Cb11hrIqh8iLswIMz3OiwRJFT+QyjOqfPsDpbBSr/pX1mocdGKs3c+Uph0iwr2UF7WM/Ied0NHT6 + yzPbL7hgFmyTeplw1CjEDyA6nKyaLfHOysCad5FVO5Y+7eJhgBx7ykiv5mPEX/qXpeRfL0JxXFiM + /+X7iiV7vH1g1eV8jnHKMNcTBkhTR1zIGwqmy71Cya4c9OX7yNTf/Ev2vGoCtrlPnHzbIBGPLCL6 + tzy/ZrjmAeSvPTSLRd9AZ6YxUgFwXa7jQSEHTy8g2ssnOsvsJVa25H0le23XgN3sXjZw7wsH4jit + wwiX0wpW1yYLBCk71it/8oDwTQN0y71j9P3l4/fF6/7mIzx9TA9sRLDD4Jst4Hu+ZxlYeQA6fhK8 + ntjEGeR3Hz1gOfetZ9YfKrjqDbFuBhpZ7luWYr+PFwz3ra5TcsvvsA+aO7kDqOorz4mhVM5vgjJb + rXmlVAclf3cF+ZunBSJmcBv3BY7y2/dv3gKTxbUo+ekZDnahPPVDhEz12AK6r54b4GivDXo8dMKY + F9oDMP1bhXxvr9R4PugGfI9GjvRW0cblMqeh5MxzjOIHVnMeXPfTj08EXBrW9Z9ekdQRkFsfH2zy + Gm+GQS44yI5FuxZe+/sd0O7MguXgamx3vU4mOJCtRFa+4M73RY5BLLUXYlmPB2A5nqlih1KDDv6m + Ydzj0W/gdvIjYpI4YKw+3gWoF+IW/a7piD6qcirVANmoOzGiHA7Djwchd94uOY2mJoBZenqsvNMG + O5/fm0CQNQfpV28BlLqvSV7riZKNpkfCIX57yttIAfEEvQJ0ehUmQJseY1zMH/07PJoKLsqtxJwb + LHp/FLoGvk1TQeY+7aLXLs9mqKbcB/m9aAJaGeMZrnwNj/a2ydf88FLW+QhlonJ2OQ05JixCqhBV + +vg5vQRaqvz833oJQ8QO0jGDgjtzKAQvp+Y99LQgSS3hb16m9eUcQ9ccjsHp8QD59MsHv/nmgFAb + fUvulgEnisf1/2iMV8+mo/z8IYo+54hY5sODppLIgbTV+GheeawSuOqDJFL2qVmevTK4/UqXQITy + Y1xGO3agjAIfOV9O0efthRTrG0wNsZ7MB7u2o1DZunlLdNcHbM1rKdwGy4CBbdc5W/kZvGSZhoJj + B3UcC7UjS3Aqgo0IFHdGw5CCn/+KV5e6f/ytPwox5jWegLnIrAFoHJ+Tm5sn4zxs5RRGj2NP1MO1 + Yy05jzF0tuaJuHZTs+Xkjgl8qLqIwQg5QNkuDP/0z/GMD/v+5ovwCFpkTk6os+x6a2E1UpdY8Y6L + Zj51N3950+WWSf8q3hzCY1s56CC135FOXHuHv3nM3dNep5q0c36/RzxBqgERhNcLpqQ9IHWiSU4q + R7BgPNwHcln/Dy4ZdGBhuHusVKbn/niyLLXCGCjfoayF5nkK4ZuLIxIpyYORhDYNmM3MQ44QsHGW + BJWCq9x+iH6KuHoBY2bC5kuyv/l6eguHDbwyjRI7yryRk3MrBD/+7S35x/3xbLjyrIC/7tcT7U5v + ADuOOmLfwIGxz1VKgV8N4Me7AU7efQNLq7yi4nqbQH++n1NgVvmeBMnGi/Cr9k0YlVGOiWHQ/8wz + x1jLiQmTMic0HSc4feAJubZdR/RRPxuQ6w5B3vD8RvPT6zLokE2LDiuvH3+8KDTqCv/xt+XIddC/ + kj1xScnli2SzDeDTR4oZfN2ZsLWWCY72UuPOtLfr5F+ewTs7Y3JMO4ktAsQO+PndZQiVfJE6NwS+ + 9UkC/gOf0RI9+wAKjXNFtpZm+XIk4h0et3SPHLoXI+YlEoXxx2yRVZ4ctmsKqIIDffZYoPo3X2bR + 0iDfnw7IOL/d/OvutqLUm+sbCvbWyPmz5Wi//EJCOrJxqdy5UxRBVNEzub7ZJDqnAC7z0hI/zWOX + /njomq/Q2T8/oxKMmQHGd+0QRzwuv7wkAJNokLjgNUTLyyoFZeW3JLDNmz6fPKuBuVkxFIgH6H7X + eQzqtbEg3fVzMKfZu1MQ1VlwOrga4A+yLUPmzBVJC2Vi1MgfCYhPAQtmXRoBtfyhhalmPdAz1C2X + vZ1x+PO/hU4dmyW9N6FCEgcZlkEjPA9OBXVx0yCTEBLRn7/89N+hdxHQRr7NkEhziH78kfvl9/Gx + jYl/embRstvCDWiGS0IQyif3az73KRTE2EJ6qDG37/jEU+KP0SKveO0iOt2+G/Crr6l4WvTnj9o3 + emMei5JLqzkTgF91YPXnwO0P8l6WtVc/oOByQO6yiw+ZYsenDnk3Mf/NZ6pcPFJxzR8tm99wY8Bp + CWxkrH4ogGXawCmiL2Rt2u04NdIn/Pl/sDmBWac/HoXtI0N7VtlglnP1rIwGeaBj8yFsKbhRlM17 + 7QfycVjc3zwvp19RJyGsP/qbRGoLDd+5IKOwVbbux1G48ldky3ymz1u5FaH9KJ94oScrp2/XkuVa + zjd/9aPFN3lBXVdM4r3vJ53RiTOhKL4vKEj3FhAwl2/ggPcQHQjO9GmXn2cYVcoDOYcNcb+GAEwg + HNSJhCs/nHW3m0HVpUf02y9aefgAVx4XyNb5qy+GMnpAGgJCVv6r/+l5M/tOsLWPQcQi9Zr96b/L + MKkJPYEWwr7aBry99OOkWTcIX7vvgOzyrLq7Y+dzcJjLCTnes4voZjkXShkefHTRx5kt6/3BRHRf + wTzRZH0DLxLgtHg2WvOTy4vj1ELzCU4BvasMzG8dh2Dd/8HYxzFrLBNtQCfi/o+HYHHZ3aHStVrA + RcLEyMoHoO3q89/+327e+RU45U5GbEti43zy1EZRL0WNWX9S69Wf/ng1umg8Yru8UwfoqAIMNtaY + jn96Z8STi45X586G336M8dhoyOEuSbTyoEDRbd0L5rlZ9DmzJwPKrmAHy3Cm0fJ57FTlisiFHL/7 + NqIr//z1d1B+NlzOfJxC+Lj3DB1LLczX/TEHlu+tiqxgAWN3EyGEHetqlD7Tb1QMIpqheF064j+K + OcLVWX/9eEsgfNVFn3JcVz8ehGl9/YLhYT0NmWy4BGXzLDBCmDDt3nQ4IWej4nzeD07ztz8WhNqZ + CayfNQUIqCZrf+ncrz6VnvbEFTS/5rS2DyFPX8pPLyOC91cNepTUxF39evnND/qht9Dtyekuu439 + Bg7DKAZKUgk1/g4PDb6PdIs36a5nzDIvnvJMGgv5n6YDeM2HQAqhiM7jtq/p9/BogVTO/wYAAP// + pF1Lt3M+uP9ABkVVYqgo6ha3qs5QVVTVJUE+/Vn2+z+zMzvDrrV3Szz53Z4kPuSsLXW0vRi3FSXZ + q5CqnUp782nHQyvN1p3f3fFf/+h+D2xPDKSPOvtdw8ILd5zR+XdcwZ7/llA+VAaJs7tuH8GKsCh5 + 1dNrz6bQ0H184F5vxFOsQz5LumxJf/3YCGVHmx7zSoERRLEHj6QFm27lC+RY74Dh5xg0S9FmGpSP + TOjdb2UGSK++ldP/Y0UB/3+vKChSQSXx8aXtHf8tkdxHU3msdSfqVug+ljxID+Sc5gud3lWUSMp5 + yJFd5Txd9HRNYKIlF6ItpxOdvcOxB/4Udd4aKknEr46awsq0f3jtDLnh3OPJBFEpVB4+vKyRRt5r + gufslBPlyu5rwk1ZgZG/SESZYrYZWfvUw08jb+h271mwRN+6g2vQH8nTd9h8GaUxhjObnoi5Sa+8 + x0XqQe31lpAcgm8+vy/uAF7jeiPXINdUVr3cLXg7WgTJ53Ecl04gIjzFi4f0YK3pdotaUcJPpvZA + YVX5rJ+LGl6V+URi/zCrk1MrGSQJ2ddcv7Rx9Q6HATyhWRNnUfdEeLxiiAxzRHF6Ce2j+fFD6bPc + YpJ9Lm20zBtXwjw+V+hsCK9ou5+fPmgnq0MKTfV8qUjCwyTovyROm5P6e0sjK70E+UniOFRG9p2e + WmAE3oCMST2rnFV1ujR/3YbcJFDYKxtwiWhapxe52sp15CdBy6SEOZgY4M88DqclzWB2evfogroz + PZ75Zwz8p8si7SjtCuu9mZLnpSrS1MCyeeNuh/D+eMrostHWXgatSiUSs5jsv2+zrbLq0vv3nsjZ + k5to7Y9rJsXU8Mg1cKrxeN6mAqaqMhBd0AabDYmTSCCsPvikuSCnkuCa8Cy4EnGYmQXUnDZZonO1 + Iq1DJj3+tnBfH+YjFAeeP/J0EDz4Nx53Pxvoyh1MQcRizCPPJnebc0IjgcqsVSS5UY6SffsMTCOb + J9bVtcZpOwoC5GN1wSALDmDlgC9LqezJ5O4LAZhfUZuC+7dt0W1r5ZFt9DeUYJYwxMaJZX/TKDKh + QE4DcRgo5IupCLykF46I7m78ofzZOsdSpPwe5AFe32ah7LTAW7Y7ui9n5JuNgQ7F0+mKnFuSA9Zi + MwXm+sSQXFNVm5eiUARUPEbEKxbSTPHD5v/mF9Ksbmuo+uU7eP74BnncPNoMARuF0nrLiUfm5Alm + A/c6bEcnJefhSpv5uXoDjJ5MiKnwc6K1kgUZbkUkovNNN/PtdNxMKZG0CRU1NVU++n0y6eh+H0Q+ + i0jl86rb91hKPtFe1wYsB8ODp+dr8ZDv3OA4i7meSSkuCIm0PKBboJYs/F5uKhaQr9qsW38tSCKx + QqaUPOhWLVCA5D6mHmMz2chXn9qHj6vIkfPtZtNVjKYeeOrFQtlUHMDaVucWamXRoPtyYe3t93h1 + sIrKBhn6U4u4wF4s6f6ZTJRyjZXTRbin0sqoFlKfMm9vxS324KNNZ2Tdj3nO+yD14SVO38T7Xq/7 + nsza2s8seHm8HKoN/cprKO31gexbXeckk8JWil4sJkhPHJtLgzQRfeEnelL28ZoFV2cNngUkEUOZ + DMDNd7JBYLgjybfradysM6lh4b3vCOksVddiWhWplfEbWfo1bbB1/lbgsY1PD5wQ0yzVODMQRCX1 + +FDW83UdOyzZz/GDrsBv7e0qBSxU722K9OKUgePYAAzdaxojQ8cZ5Z1vkEiaggqifKakobP+KCE3 + oQTdpbWhU1TJvPQq2jsKmuVhc5+CK2HrH2eiPsZ5XJ5SykIFOBq5f06DTdX2su+z9047PtjjFrC5 + D7PXeUHofaj3eloUKSfJAyl5/Ropz60atHs5+IeXy2WcJ4ifsCahiAy6HcyPLJ29t0ECdl3H+qrR + QWro28bsaj7zWa2/e73CkGQ8NVT2dLkPkFeF1vvteEIvt46BsGt18mLMrcF9VXTA9oovSQTNUuns + Ag+uzCFF2uuqUq6zrho4ZyDHA9+qIyVLKsC93rH8NtaIhHrgSdq4XsiDzTm6DGYQgpd895HhZwPg + zuwxlC7N50Vuvg5Vkp3tCm4fnCJ9HcmIU1yxUhq6PUqjkFenIuJNeNF9iURJKtjreRI2eEzFCJ3Z + et9zFcmDdDnwHfGKt2MvenqK4adRNi9Jey1al62ZpPBnEOLFkqwex8pi/54fZj2rbKj5u5r/+PK6 + 9F2+mmMOgXk3v8SvflbOKaueSa+BM7zVK62cc4+rCcXyqRL7/BTUhS9bH/bfX4yMPJhsDH5R9ocf + uAmTi8132+TApk1cTE3Hy9mNNWUoDtydmG0d2vzpEImSJH2fCPmVatPm49fgtJo10tKYa/AkaKnU + JN4PebJwAywufE96lryMzkGLoinPgAmONLNROcBT3rOzKkswDxykrs03osethdJXqR8kOHJqtLhZ + Nfybf+m5jXLOE54CNGkFUGE8bhH7XXVTehiajWIQxvZiL28TTu+gI49gMEa+ZPfEnlta5IdT83c9 + DtQLTyQB8zzn7OvrxxI3BibK9O9hXN9Sw8OCv62kYOOsYZvbXMIaTCsxE0MbWfa9ZFBHh807p7kP + 2MUyO7jjIfGjfgWUP/0Y6AU3EZ+Ky83+xRKjg6m6peiue/sejc0vJPGov5BScjd1uRpeCTeOa9Fd + x+K+Qo/XJOG1lEQu1VfE/9WXN/WUXMozibbgMIvQupMec+UryY+aW5VSW6oKMkp6sY/J7FQgEz4Z + ida4oWvNW7r0WGeE+b1nsApzIsPPfRbIHWRFtNLA70RRkGdk/LxtnF7bwMCJKQFS7sFPXfIoSKXm + oUb4NNM7oEpTwVNJZYzudceOM6esvHQY/AwFE1Pby2s8e9JPsR0Mb4aQL/MmlSAFa0Dsasii7XSI + BKn8HFNiUXKzt7jMFvDizTMp9r+fkP+qoHWZZJRHWZmvu16EY18zHuPdFPsoc0sH+5YdkDYdOzov + fNFCc+tmD/6cuqHtfJdhfgMQGbon0yllpxr6wih67BDkFP/xa9U2EkFThsCuHxOJbLZP1AM8j5xa + ZjLwCzMj+neBKu1ylIJUlQeSXnTnv+uxv5pKbqe8treoVnspFY+qxxzOTYMPGjRBVho1uqbhZ/w5 + 1qeDrhXXyMrAnNNQfzhQHI53ZFB+iagjybqUMIqI3Izd7B0fFfi8PBoin912pDpIF8imVw6vKxLG + 5bGAHpiPs0ry4mhR+gaBDE/XV0y8blToOkWZCa9PvSW6/P7a8+x7MlSvBUv0wVBH7lNIJQCSUiCv + ag4qDS1Lh9vj5ePTjicU6CMDza2dyaX0N7CdVI0HqG5XEqSnI93SEyj/8Nnrn+61GcOg4UGNU0SK + 66GKFnFrHOnNzwp5kqOUb+XxJMLf2WGQvHVLMwfoLUufNaqJnOAEHPlNdqQnOWXo2s0bWPqsF2EE + XcVb4/6d0/qAeyg8RQV3oKvpdnY3D6aKViD3PUbj/AgkHSiSO2Eh+dgjvQSMDnf9gqXo9RvXAL0V + qfb9CRW3BIBtOmALUFweiPWJDYqf3t2D7elpomfkd+qShDoLiv5yJq6CZjA9pZSHxPx0RJt6NSJf + SY4l7K8usu6dRo+OZGpixwcB+fMPJF7SHorPisNTcLlEx68ZyfAU8TJSd3xlX6IzgdAhOT7G9jdf + rEPbw6+WciT+tWu+vX8HEULX7olCrzXdSjAsoLxNjkcH8IvWrw4h1wq1uOvRwV5ty2TF5sreSLK6 + fUMQ8XQIWZYntnnhbXpdUQ8i4LMoWUfUUOZ37YHJ5XcPfCU0rtbDieHI4Rg5das1PP6wC5wimSf3 + aRkb6kotD7dropHrrjcXpPUmEIvui09YAfl2+YUi3PUqXv7wL9PYVLpboeLBwpXyed3PHCjcz9Vj + bidsb3mFeXDw2Jig3KzAYh2mAZSrGaKr4ITRuvBFB4/WRkh4WmVAT9UtAYqEJmLjjztu51vOwJN9 + W3D30MdomsW7A0kyW6ismpfNbamY/NOX22fix61krzJ0mEtG4tWUIrKPB4RdpyMNhKxK19nZ/vD6 + H79vecEx0u4/0f57DQ2VKoFN4vzIi+ke4/anN3SWT5A9C41N7nXSw7PbnZBareG4MJ4Kofasbsg9 + 12p+FF5CC+XKT5HGfK7NjM9KLWlFZxDvgD21n4nTQqXiHh4ru0KOA8Gx4LBMH2Lc50rtf/d4gzsf + YgY6t2i9FmkJh3U9/PNbf/oSUEgIMsdjlq8cSBXYnl4mbhubB7MLalna/bMHsuBFlzx67HimDJ4o + dnyzHJQ1AUkwfDEEVxNgMGsh3PmEKPrZjWhZtB4M75qBXkCZm3n8HVLYlmfFg9eDnHOfh9ZC/fcu + iM7Jgj0Pv6yHkVBaRNvrefkhsO+BXy+kPLyGsX98fgrssxdD5MiyACuXzQI5vlXIpXzx+czqNYYa + SldyuYihvVqF5En9NW09TmNalYZKH8Pj96oiRwc42vlVA4U9zShWyapu7FgwUDWrDrnnVQcrvLgC + 3P8fhbu/3sJprSVZ40Okp8cvWB96XYIkbDlkc67dLK+vn8Cyo51Xu0aYk/exsqRj53HkatHOJi+3 + x3/zEykZrdXl6cSZdOoov69r+uTTcoucf/70D29I+8MmeNzdO7IU1re3aDVMYF2wjKyjYo38CTSy + RMH6JpbJGCrbHmRL8nPPQTv/57SZxBCSmgmROjyVnAYVSQCVPR8febHe9RS1RDpMBnFq2qvrfmAj + YEzhjjQjktTZtkwe7vqIyJcSqJt16iE8Hfw/vXsDy56XSLaonna/3YOtBPUCFaX7EXsMBZXes9SC + P9lf/uUla2UGG+Q1dkPPtlLUVa3JAHf+9Eaozeoyj/kk7v4SMzYjNotyvtbQOag9cnnMgF+N+1jc + 9Sy5fIw4Op6+R03yhu3+50ejtVjeIqRnq/Qq6Vyq8yn7DUDPd/zWgRdx+WFOod5/PWQn5RLR0/eo + w+clb4h6uR/Aam7SBC9tM+Nl28+MSEZcAhDWH6TFb13d86oYvFm2I67o3+mmd94Cdr/uzeZDo3/+ + Eb7m34psgbTjElUmC+fH6Uyuc8vue34lEySXMCI7X6skvMUTDEh6QkbNpTY7Op0IDXGLiOYO7bgZ + 6F2DdBwu6AqwbW/fpsKANeEROa+AqKtx2rAEJLlAHmspOf2gVAG3som8owhvOZE5oYOmzc1IeZ++ + 9tIt70x6Sggiq/Y5Srp7UsEzf38T1H/lfLvM6iK51yzGPXp1zfpCoBaX4meQ5/W20ZU+Xtq/POb6 + 4tdoz98w3OUVuWIlj9a3882g7tomOTPAzFnTCjswMKyJypWKYBs8jfnzK8gkfULH5zfU4Nu9puhh + MMcGP3WugAK0NmSJd6puf/q1swfXO9TeByxxWxZgzydQjo1u3B4bX0iX9j17R2CnFLfOc8fT9uQ1 + w/ShSyd8Bfi5YB9drihrNhLJKdz1hbf5Yxnt/hIDR6mu5LrXc+Xd4xr8TvwDv22bBZsX+LzkoSRA + 2hGb6nHHcym/nSCSL2VuTwF/3lf4zxFyyfEZUZkTWmmvf6TJbpp/gVXwwPDDAl3gwbOPh1+XQe/H + Xglyi8Bm4ZeLwZ9+k+FijPTPTwYqZJEze7dxbJbGlI535oKh0aeg/rmVBzPhm2GOMFwzn8qwhvRs + luRqBK667fmWGJ3SFMVNluVkrzdYLfTiSc6pBHNbpBiqDZyI+kSP8XfbmkIyxCVCN18vbP4FjRqC + 2x0R43j9ABoft1Q6/ySXuIpsRYs4Xksgq4KGnhf6oMepYCa457v/1SOQBR8I+OwRpX9/7K1v3QIW + bHwlma85KpXUFwsmcVKJ/9h6uhY/05ISLb6gku1/0Vo5Vg/HIc88Ekdbsx2tKw8vYX0hl+TZRGOx + tfgPX/F2D372IooaA/vsySCTP6/R+qcXbKe1yU25ciNl7bWXHibzxQdlq2yiuX3xl396pw4mOZbK + uYZ7Ho3f8bGlc6VVwp8/8mrCz3/8I8PwrhvkegOlStDGQPBvPjk/O193vww+RjuRFL30cWlucwH2 + 6yFXIOFmTsg9hreT9Nj5LVWn9XaAwC6/9E9vAy64WxvwvYOG4S0BlDCfdPnT03seI1L8Hb8eDJhy + xEwsVeqSzd8aqg0zeYfr793QZeoVIDHNiM5HrsnptdZMkCtQJ8lPkNVj2boM/Pucj2FqD76y6fBP + f7jPXzUui2W2YHzEMglWlI70fbiK0EycB/HW9h5R4ZvF4NRIDjnH/Tv652/aIbSIvhaOOgNZCOEp + GQZk3GfZ5i5a1cP+swlEuTRdtJm+F0InIMa+omwGVE33M3latcDHwODyvZ4K4MvcGzdr880p1hkH + csfKR54xayr7+Pxk4LKHjji7v+unzZ/gjgdI5S5VTmmJZcC08RM9igtnk2G8lfCKrwwxXtFL3aLa + HqAzQ89b1/kwkr/89WTlNbFv6wes6BemsK5yAU/W8xtNXu6mINLeHLpk+EFXg7/y4Nx5t3/8M9+L + mvnveaX2kO/5Qi/t9Uj++gVUCx4xuJaSTM5Dele5dewmyK7zC5+KWBjXi2Ol4iliZZI00tSsz1Xv + wVe6FqS0GXFcn/NzA1LgWsTy60hd//JA1tBy5O5+hKyzs0D9OMro3ISbuibklUCYljWSR4JUTsJ9 + Cy+MmBJLYRd7RW9hgLu/8phCvdjbX17IPA4rUqtoHNfQdze4qqlBEtDtZxpovQCNV+Ii97T5lBL2 + ukGGexzJpf25+RSOUQmtt6mjFElus7y2GoLW52biNIo9bmdX9MBfnhHys0fpunCDtAiGS86HfqOT + nInCn19HhdEL4PecbwvshOuMZ8jXYIsj2ZQepvzAazrE+R/fQRoEA9nzDbp21lmD8FvFyHyjYqSc + FPIQXUKE3Ohc/10PA79q+SV7vqT+zW9AbgtEbmYY//w9AIwoeFTLV7q1zq3/x89udFbG4z1LTcg8 + zg06z+6l4dKngQGK4/MfP0bkojcl6L56RKz6HObYTbcOaPJNxkIUJvbud1kpAiGLuXsf03VY2Aw2 + 3OPrsb82iP78NfjLp3T7O4JFMRsfdD1DkLq9uHzio6AHu35HqfY4g+2ysIN0G05f4s11G62slXfQ + 7pkXUaF9ybeJjSporUG4X+9nXMdOr+Gej6DzKnP2FDCIF3d9gH8qbBsSuiIDLtF+BsXDBPmi4ZWX + 9vmILMX4NOv6DTToqOX9L2/L6b0ueyApMCOZaJLxx96wA8oWWGjPs1X2rtoJkCKjQ4YCxohIbOH/ + 41+o3uucerytwYZ9PkhsdeGIi1vhnV6KOxHrKnI2dsidB69n9UG6vFgjX4JhA+8CWcjjdD9a//KG + P/2mhp9+XI/T6sPb83shXmgGzeSWjxokRgCIvfdjJufXWpBNbc6ruEFpjtfV6GF8/5mYjcO6WdX6 + 28Orzz3Jhe+iZiuCJoanpB+8I1PiiP62rAR/9SATRRtX3WRCsPtXj3tuI1iyOhZguTgaZuTfFi0f + wCqSfvzJqBRygS7m7ePB6LSfybLzyT8/WvxUhDw5w2B9eJonXeLsjRRH2M/YEcoUTKAPMNz9M4uL + 1AH78yD6jd7oBn55Cv/4zFKMy0gNwFVwql0W8wtzUff7G2CLlpDkUcbkSzaT6s9f4KPuVXTXe6kk + iQtFKQZOxAdqwgMR5QG67P0p/u4QARTW9vLW5s2N+MMtiVRw4d07qfNZ5c3oYEFbuD6Rlx4idZKj + xz4/pBV5B99QuVP27qXMhQnJTOarYnYsIHS5E+tRQSC73nMg2Pt9yMqtBSxeNA4wb4wQ6QPrqfT5 + WmKp0gMXBet8aH7vi9vDK9oUojBVGS189Ojhb9i+O359bJpGuSUa0UXHkhRfAXUv+gCW5+eGcpTa + ETUOXQqL7DSi61CSfHm3mgMfTfEiTj6rKkcfL/2PX0hkOjhaPxWO4d7/2fsN7p9/MuGJKnvexMXq + 9pfvXMuDTNRq3cZtMw8+bNaeoEIc32B5fzkdRkTa9VtYjcsHQAWeO+eGgbbMDY2PYgrjxW+RkZNI + pWdRd/75D/nRXCLqyKsiOlgsPHbvN7Cm/pvgPt+RinFK8948LqJr+wu+tdkVcM/5tsFLWF3IJZOk + fBCFUoC/Z3lBbq5y42y1JQMN7RyT++4nlj9//F//4zSqiyVYAty6oSP6N7zZ29/z3vufJJKPHzou + U6UI+pgLSN7us71abQnBKx+umMffZ4Pzx0WHWxGIRE2ucUOfzc+D/48VBcf/e0XB7C8RSV4TH9G5 + bXi4wcMNC2y60mWezBKCNsmJHR21iF6iKpHEebl54jeQG05HXw+mxX46eBCjZrEfmg/d2jthOL4b + dVUL2MPomiJ0taR3RLZxUERJdG1iJYIYkY/LFdLkJysWi+vVns+Lt8CL++Sw1ISB3bAPrYK3h1mg + MHfWfJve4gBiznsQJBpqzrfzA8IrNnTiPWirLiLAOmR+8YTKJe8iehJaAX442SW+lPyi9dqSEuoK + Eb1Rb5p8BabCS8+fSrzaFZZoa5uulIIGUGTQMwuWQ5V0cHpzkDyV6yei0ytaoNIeVOLcvHdO3w6K + RWmMFYJWUxm3r5L58GM4M/Kl8Tqyr+5awgINTxLTjqG0bGQfaiMcyPUjqpT37LiC2sgMyEjHaly+ + UltKVnS3iREEwJ4/IXWkLmlS4pWmlnPz79fCa6xUxMXPPMLRojFSKGgeid4uBvSNnwp0HvJKEmFD + gEORXcLRLAev3vLzSJUiTCQVHq9Iz38NXS5BUEgqjztkODCjuCmTFN4T/4uMpZHAKn7MElwcm8XM + 1/jY23mzY8ithy9C5ifIuS9wOxhE0w/ZzvaKlq/hKxJ/l98kXskNHA9fWZB8snyQa9EPOPrPayi9 + LBpiln8k+WjnkSmpbISx+EjWcX7pEwTW7HyRoe2h+4/dE42JG9DDZIeR5rW+iXEjpKRMMENnQPoW + HC6sjZzSrW02a6MEnrITIOhb03EmBPtQP0UHoi7zKfq1Zq6dhJVxCErvvwZ/tlqWQoPjye04yw0v + o6sMQ35x0d3KA5tzTmSCHehCYnuAV7cVHjuR1z4/b/3cDvYqDzyWuEYr/40n96bZAoXozJF75uf0 + CMkmSFmrlshtwkClr/PXA64olciXf7PN5zrI4HiWdVIstgpYK4IejF4zJEUMo3HVrkP9V2/YxWau + UndUE7hZ5pXczrwXkUo/hFCwAoTMVjCijXffCdyye4RZQ87tFZqHXaFfdMxcM4lSs+9beFgd6jWA + 88Egfk89HKxCRuEY/ehqC0CHl3LlSXDj4obP8cOC6Pb64g+c3uM2hkcdrLG4vzUgvOSLqq+JpPJT + RwJF/tEFBboAR3TyyH2v51UVsSjKVRqj/Gl/GnpmYQUX/L5i1r7lOXd4HidwJJeVPPCpbZb7hbQg + mFeEHlPujvwYHnRgWe0HZVV5jWgXuTXMzIHH+2kDzSKjJpGaCbfIW/TOXjBSWCnVdAPpMmHtrXyr + WJrLe7YvYKoAZzYiC8ORnjxdBF2z9EOYSFnEx//wafu7X9X4XFCOWZ2y5FFDqX/WB6JUdphzDmOb + kPE7AxnvNGrWP7z6ZW7v/QT3AvgpNSyosgEmj7RTAHv96gkk3U9F5yJholU6+BtIXP2FVA03DW77 + qwg+7yD2lmvvUg6oJINW+3OQdXl3AP9UXZT+Pkffh6CORz+sgXCe2b2eHzk/PV8efA5mhxe6ZBEf + GFsnUY6JiNsHobqd568IEy+rUPKTj81WjQ0PtbbWkfwwhZwStpDhq7NFoimJOC6ZcukhIw4zXr25 + AFPbt57ESuwDxfyDzzfCbwJ8uY6FPKNSVH55SCHULl+CadS6OauHui7et0UnTgzpSA71pwCHu+UQ + 1Lze6taUZQqG6/1KtMtS2dOGG0+KvAf4wx9A9VIupSxiYxIFcxuxJA5FeLVsFRMYTWCZuiWT1rkf + SOK/nYi74pCBp3bfY/iqRbCK1UcGF5g5yFYXm7JfaSogg/wL+qvHTUivBdS7UfLmbQPjEqgODy/6 + OSCIszl1sXm+hlw1dshIHyHgJbiKkvoep7/7abp9fMVeLR7oPp6eKmFB6EhdlXHIvRyPzbjjLwAP + UyIPWOv2kY3YAqrj20TFiwx0TUNnf+vOyBL1fprH2c5lFpZDa5JgevJgPf5eMnR1PvAkTH7qggJP + gAM7r+T6MpyItfizIG08u+GTZg0qmZ1zD6V0kkkQKZK9Cj7LSyysU6Q6K28vBkMFKJSySe5Z/Wso + GAYWHvpyf4uEW+brUhQW1C/WRq7f1QH8dJgEsUyOCJ1nTQO8nlBWqi56gLf6d83/4ffXESNcj9GV + bnxnV6c59Y5e9XwK4/TM37E0RoOJ7NPgAzaFsigxztH1DtvBbubxdoPgFxsXdH7pjUrvBl9Ik9HZ + RNUeasQqR26S4jGGJEZDPtLrvAzSOXIrckXfMziS5L1IvMi0KORLVl0opb7EyoqItCKt8+Wl0g2u + iz+T14njxsXzQx823NNCvpN/8r96Ape3R0haXwhY2+4SS4bIpijzp8Dm1nfrwM2YAxKox5Dyl2Bc + wGgWAzEM6QRW73pn4EHNI+KePpHNPe5FAuT8/iDXDR7tRVhdBr6eeUzOQzDnKxuXG5ib7IZUV+ej + 3/OhFJJ8sk5I8Vx/3KIww1DIPz6KejUFKxsni1QzQMPs/XGO2Evyq6BWGx7RL4mi8mnMlfBr0wIZ + 6BKMvIbfmRQ8uwvS/BjmWG3PvvSHvw8cJs0qRfdOejlGhIVHjiJKvTaFuotdLIbeiVJ49yeJvLwS + r3dmzBc99HTgiocSXazKo5z5swT4PZAX0ouoHRe1bVr4UtwjQpofqJPk4wxWec4g49sWYH3ZciaF + wRwhO8DuyMJ5Uv7xi/K3ZxNFzwWqrd4Sq30DQI/YWSCQs5TI9isB6/Hm+9JBqy10fUccWPmSF+A6 + D4MnKddLzq66z/7DS4b/kqZ53HEP94OJvWNv2+r8szgePnCHEFLBMk7jHfKQ174/7/eObntCHaci + StqS5NPo2z+Wn0qJyhPCW8dcR85/Np0oKOxI3AVszcYoiiyd4+ROHqLmAVrCuwZ5Ner/9MNYGe9z + Lw04UzGjn1i7FS+PHrYB80NX92rRNb+GoeT1G4M0mQ3Acl1OLGj9oUeGGYrRBrfHBumk80iLDgAs + mjEzf+PtSR6f5ksNrhM4y58SXbXXkNPECURYSx+VxEqSNdjmCwfs+oLo6yCoay3UrNQEjUH0RzXb + q8IlLLCbbfZAyl9yGmc5hHs9o+zanqJNbtAGzx18kNfCsxSb75yBfvw1MbT7Lac25RJoKr5E7kdJ + VbfqWlrQ4SofGdKxpL9evk2Qe91bT5Taelw2X4AwMoOUJHZt56zcXypRPx4YLCaCmOOfs6WwybSa + 5O/qAthEflVwZZjRYx5OYq/Ti9fh0b5i4havB5jWVtWk04QgloIfon1ZnTIQ5t2evt0flAiSmYDi + xX09Qedf49JaKwT23F9RYUSbTS9r1kETbzlSk7sGWA9rlbjzP96qTY728eLBxbmyKFWcl7o48qCD + vb7Jrmfz9Y+PLyXl8WKv7bgwuV5B+ihYZJ0dvVnSuDElm9MxcRkUNUtTOhqk6HMnl+lzVY+1wFkQ + /5iMXFb0G6ePr8hQGS8dZudJzTk8lA74vuobuf4kPl+uXy+B39twwoLiHNRlFK0KtoaokPNbLsFk + 3EtTLEK/InGkne2Va8IECoJGSHbIx2alUW391ZMn7vNrGVixgOQpPFH4Mapmuh1qHs6vMUconX6A + Zt3gwD/972ITqIvY2AKgtZuR1N5cSrOCwTBYPJNc9f6sstsvMmEWkqfbHT6GzcYL8eHO3x5r30D+ + M+LdsWd6vdcHQ0dfpBW4L3cPfwrq07nt3ASwIH8Tcz729pTyAg/3mNoDBmUimrJTCLVfs5Eiz4bx + t01ODL17KREZvGSbDYw+BH4fn4n5q78UT2+xh3nAMsjvikfOfyXFgrof++iM9z1/5juHUIu8C3mm + hZFTvpl6eOl/kTd+uE5dNOec/eN3Y7Vsul7bbykdRf1CNOtS5cMZSQO8gaUkeaLOgAi3ugUaS3x0 + 3vXRqhbsIEoe26Bb0Ovj8gsWE3agDVF4IPVI93OPIU9WnVyE+zunAo15+IsvFyzs9z+zpeXD760/ + IZdBdJz8ZD8SUBJFfFRFI8LVOPKgmsiBmAp1VYqDB4bKY/2Qq6hhOu96FgjdgtGZw598CKp4AVl7 + Lsl115+cN356qFy7nCg/i6pbfDCrEzs6d3xKY05dLkTFcMcPry+8jg5i/etE24g8vFqHoVkcqHfg + 1NIUOZ7+G9c//+mt79Zjf3fV5r8ZW8BdbxFbMHt7/fu9/XrxQe2WfNWPGgRjy1yJpYkp2Cj1Gelv + PIvx3djbcKocaD6LjNio1mx2nuRSTE6r6bUDZPLfJwSe9Idf687Hxz9/y3qTgPQw5fIpmXwMVXT+ + /ekNemRB6Ekh90swYOIgP1pCukClcJ4oqfoWkP4bWFI0xyGKd76cNjx6oLwzOjJFvqezPPAT9G2+ + RuZzQc30eNkh/B3C+u/6KcaR0MH1WxkoeIdTs1LkYXg4bZ43mmYJqDzE/+HbM12/zfCAkQMlPTCR + lw+uioEFFZApVYSCXV+xj5fqS6cMALzIzByt43EcoMbOPj5Ecw+23FHLf/mHW7GZupYXIYFVdMi8 + IRGyaC6boRP/8c/+/cspbRa46wtv1dUiWsnpsYHugu7I3PFkK9/2JO7+k7z0hwE2OQ50qXxlA9L+ + 9D2Olla6PayCnHc/t25a18GnH1neOmsanXb+laLQ4vDEcm2+XZwcw+Oz65BJ8Grjr2RZQOAPAt7a + Kmw4VvZl4EJ9X2HDXtRVwESDbD1ALFlvQV3M7jHB6+FhoItue81+fzF4dVcRacoogtUo6wUK7Z1H + 6FtHzTFxHgIUTumCkstY/5fHlENnEo9d9PHIN20vOa2DiBEIV3sVhKyDMYIxenGbqq7m+WNBy+o+ + 5HJnEaD87FvS872fvsd7wbhUV5WFUw0psqNjmy9/+JZ1ReIVh8QFNP8mLHR/vye+BX3XbFv7TKFL + KxGdo/szP4pGVQMYqBq5Zs8RUId5lmDHH7zM7Wqvj6LNxLyxDkg3XFtdV3Vl4Ko8XST7e3t8hHSC + EDrvPf/4gr/6FLVH3O8rlI75nr8k8GH1DvLqpzauhOAQnDvmgY+fOQJLdE8Lyay0pyfezkG0/vld + yh4LIq+nNdqeydj+6ROy+0tK/82Xv/yoJMeRiEZfA/sBfHKNbqy9RvHaAcZlNKRJjguOpSx2sGUK + A0XOyqvr1rE+2PU6Su6x2uz4a8LAYlRMgzwZsdNY2j/8VN1UsxemFWWomfIX+Uah5Xxra74UhN/t + b/wbPGucDPd6Jheq7zsoOrEAux9CSv37ReufHm2vjkqCC4Yq/vhFAjV/dTG7uLNKpaLapGnI78iM + Vw/we14I518loHTHt1WtHQxxuZ8K7OVX0M/fjwKPpfBF3ioh9e/7Je51a4lFGTrOgFQdqIqKQ9fz + tAH66tMEvgt4Jy7UIaCJceXB+rkgpJB+sH/ikNZwFViTXN8fdaTAWVopDsIEryPl7Y1XORakcRQQ + c9T8ZmXq3AS7viWOLVbq9kyaFj4UbUXW4p8bChyhBe+CuWPC2Td12f01LFixJjLJM5vaVIphsb0G + dE2Bah/za9vBomnf5Nld+pG8HSOGWgIWoklvT91eDAdBw7dn5B/cMDp6njvBoxN+kFEer5QPlKGC + MtFCPOqNmh/ntmGlXi0fRMuNqNlY9tJCYYUOiV7nS8RqxoeRqpMco1cnD3RSJXeBRu8lxAo5g/J5 + Ow/w00wOuZXQGCdBXAtYpA8V84tJ8i1iq0z6VvwduVAvKKmDVwb2fIj4XXHKN3mivnQZddvj10Gw + p7AzTHgZNZt4SGDBNk/hJq3f2vCkG+PblHk+U1ieXgHmf/J9XNWuMqVe2zJ8VPtTM01PTYTyyTwh + tb91DSb8JopaIGqkDMcp+gVWUELp/r1i2OIR0O9z7kG3Yh6zRmhSEs1Yg6dC3oibWqeR7nwC/eI8 + Icufgp1P7Qn88cv5/MjpSkgXwrLCEzGCpqeLUWAP0rYP0Z9+mTTlJoLSWXrkP0iXL8+AY2GqaQbJ + vxNS2bW1dajWhxnT/taNK7lrOvyUY+Dxf3xWTFgAJ954EOfZxfZ6a38bqCxDIvbT/owrU0cmyAOe + wRu/vCl/DNcU3n1L85IX09prl9qh+Dq+AFJf50vOCVkfw2GUbIIsbKobdzcYGNX3AzJYgbEXB3rd + v7zyz1/R8mgoYmxkJ485BItNmffCwl651Vhs6bU5/qyogg2jfIg7ZCbF3s2UwV9eGiSF0izmmw9B + aMIZXSNY23u9dxCRu4qHP3/CSqsurWqn4TjH53FJLFsG+J0eSMF6fD6q7zqFOQAtMfl33/z5X/jt + fYDikMhgy9tPDziH1Mj0C9yQ49Z7UJFIT6yehyqJN96EfqFOxP3RDFDxEy5wa52FyPa6r7CNhHbf + 85njb+YDsK2vSyI2wdsg+nw7UlrKWwf2fJF4qv0Gy+McyPDJJZq3vABqVieXSrD7XaSPVy/fvJss + S4/pekc6+0TNrI2LJyXQ21cY5Gp0/Jvvf3mMbQXcuGT6T4f+fkYt62XySOvglQq9MD1RtvjvcffT + G4SDUJG4XW721l6gAx8xpsiss0VdEk1kwOnAcMRixVole94igoclIf2KWYrriAnFrNvD9H0+Db6U + ZfAuegNRiqesbv2+J3jIlzPx8tsEsNygBVqfRUU3rr3SbX25MXSCzsX0DBx7Hra8hc0nlP/lrf/0 + vJzfHmTXX3Td81Mx+1g3pO78sI1iwYK7b2oo66AcLSf53MGzqiRYNFJCF+X8DmGARR256e7oNMPu + T6Nxp8jl/JAuzH4KuH8aFGKcambcsOVqf/yGizyzmuP+vP/yZKIrgkU5sZplGET4h+HuD6cuGyxA + myZHejioNvuZWV/a8z8s7vjL+c+xA2Vv+8QNfoTOYW9PsOsZn+zPK+ec03cSbwab//lVuu16HYbD + S0C7Pgcb2e6lBLKSIbs+j1bgZiIsT8/AO6Hvmf7LIw+JevOEzenzLXSt4a//hnSp8PN/et950zuG + J0tr9vxVPF3utxBdix+Xb4+yZ+DO/2jvh0V//gJM7knY82W12f1kDBixn0mg1lszZt3gAeFWzkiR + WxQdCVso8P4u/jfvLpLAl2Zoin/9pXwRuHcGAyXtyeO+rxCpxpGF9FGyRGanHuBL0CyilqLVk0AB + 7e2zDcrJpbWILtn3G63Ph1WKf3jnnCxtPOb7W+G6l0axyLDPcbsEzQZHXd283Q+pS8ovvORh8Cb2 + sfxG26uLFvhM+Q857/km9l+9Cd1KEIhGhakZJ31t4QUEN++JTF09lo0cQnnFN2ISHNjr/PFMcBB+ + d6T77ymnk35qIethwZPWuvrz75b0168rHgGlK5OSHriXOcOrRt/5h9F6EV4ei4UeH7EBy95fkvb5 + 5HGs8rbpl0csVGXzjVf1uNF1eJxScJKa/a1a0kVdPr4lgz3Pwqdj81GH6flyYPQiELNKIjb/+O+1 + 3lKi7vnfdn78TPjoFJa4Hp9GC54kDTKpPKJgxwcW3YYE9IutINSXis3yalNBZ0g2j1HtM92mQyvC + VrJ0T5oEWeUKXtb/8nQUXdWGzu6wyrDYngPS8iKibP9RfGnPw8kZxweA9zwXiibJPTo4ZNzOMxGg + 3+9vSZrGxV7uXy6E1RE8kdqrAl2Nclj+fV/6XEizVF4kwLozX0iFr7UhUij5cM8Tve/wU9RpOcsL + XM+2g+SYCmrv+WEImGQaUBjWPqWOskD4geW657lKzp7ScfkfAAAA//+kXcvWqrwSfCAHck8y5C73 + IKjgDFAREJUAAfL0//Lb0zM7L8DK6q6qLtKdBPrtMlFfkxvGX76rAAfR+dLjsebS1XxXCzT4/EXP + sGDsh7cKRp9lh6M45FOGMzmCRNNtGlDMl+uLnHM0HjofZ9hayfq1pg0a+1jDmhmk6UqPnI7aV6zS + iFSvkh6uLxURW9si3vxabBmScQEziEt8vzLTmPRnbkKnD1V60Nal/fG9gvDaPebMxAcwCZ8m/6u/ + EbLvFhFrP5nhIHrfv35TuQ1ZUsN6nPbY+Onn1oubjayHNNGsUWDJpsD9QH28vufZXFC6Yck7weFm + 6vTYLi359dsTVF7ThlrOrven8m3W4Ld/Eokiav0lJVkAw0cnUPUBKJldwtnQdPQ3tf2GpPTXn1T+ + j4kC6X9PFIhxvqfFIZ9+iEQRXK9cHpVCem2X4/EUS0/p2lH9clSN7evLPVLommGzJHLLxGrRUSQ7 + Lg4SfSDr9WCcIH6ezpHQPpNWlFwJQslLpGgPnqHBP5pkhBe76WmgfBp/GsaggrL4DrD1/HzJpl51 + G85Fg2YZHwJ/zQsvh3YcWfi3PsACHQrQIo5Cb6cLLjn3fHIQ4t/xLJNvxZbmK19gye332Pt6Gvit + T1XQq04ozi6rsT4xbWB4vp1mrn7GbDoH8QZhXXjUB45m8A/+NaDiQAg131QtF2lwFXi6Rgk97xe9 + XVzq2TAvzhk277KbLiTZCugkmTMvs38gIhxQDT4pzPDNtWsiysIqwDj4PukxPu/ZJHueogDD42bm + TkW6WW/cAZs/bNhHTc62JLdsmIRxSh/X3CScq59NFJ2XiRaZkfsL9O0actxDxVFbe0zMgrmHQgU7 + mlytFozgHnbgnQ67iLN2lS+UBWcjeSB4fsD56rPgctjgrQUG/r1UlnJ7sg9kUXp70bbHALCapx2w + r8sL+2VeE6GBygWG+wRj9TF9jZmBZ4VOs6Fh57B7AGG/VR5Mp+lDTe59aTfyAgX04Z7S6Hz7+Pym + bR063aBBQ0O4lYJQFRFK6SmnlitLZBUXTUU/vFGrUvVS+MUDCsf5gvV3qxti+rJ3cHZEPItuahOx + E5ADXe4a0pvwENlinewOaHUqUeu0Gr64TFSAV3it5h36snaylKaGT9bx9N7TwWfdcDXBzd7bs9Cc + bMY2rf4gENATjbaryUThonkgzbCPXZ1o/iIv0gZm8a5QFQlH9r3Wa4zesxxQi5KxXEd5iBAdaxPn + URkxPlPVDpVR3v7yYTCxu+EANm8DRyR2tXYBnvIBINu3+KCcvz7/JVIPSf9YqQGA7G/7QPfgREBC + o2JgYJvDUgULOBLqvvArJVbY9lA5PV36CGeXbdWMdXAd6BwVEj8azPsoO+B+nd28B8/Jnyd79CA3 + nreou02hIVTPBiJo3MN5Ma8pmMqn5SBGLyLWGi1kwp06M0r1oqKOzHuGaNW7BtzpfZihrA3laE1c + DfdyZkRSO8Vgvj93AjybzwfNDV81eLldOPTVe53iP/wus7tAvrUJjvIjYeymEQg1qbSwucMIbFki + fsCiXTeaxHgE0/iAMRzPVoUfRuyX4k1rISDuk8Oln3rlYhbPHo3t4GLnbsvG1h8LBY4uH1Ljptbl + kvobRN2w6hjPQ0+23VG8oAbd53ksJI2I2zMbwGU3r3MjaCpZ2/gZoQddDKpJ7sHYUv+p//ENP07o + a2zxUT8hUgkJ9Xz2LZevtDTIu7hytEu+XbrcwMCBn778fFroc7HYJ+CM2UQP6cUtBUSuBfws0gln + xctpuWerV0Bz14Jix9bbtf66PTgkGsU+gh3btqju0StjCDssL/3Vqrke9bl1wbnDDi11waZCeVtu + NJDe95JfDqUCg4jcZiAfbulvr7iAegSaeR3Ejv34mEOb3FN8+1Rnn0GzbVBgmQifd64HOJtrc4iZ + atDqsD8QPp+0CoWZadKHLjzKtYanBrlBqEXfoVqNCbfiHXjsIER7+zG0FK5xAwVV/8xvLj3/7rNC + HdBFa8amqO7IfB6PI+IzrqJnb66NdT3QBa6k+d2adVF9Ibffd3RLwjM23bAsR14fLvAB1IRmu0ve + 8sElFGCeRRca7s8R2I6ZHIBoGwea5YHacmg35MD3sUE12OktU9DfGTJ1wtogmozX5NMOKuNwwa4C + dbCIsWqisWwA/cUb/EZ+dBjX+m6WdrHViiAbekDmWsV3sY+J2Lh1IT+4jtGT0XUpC4Wuh1OBO+zb + UsTm5X3iEH8sEA7XZDRGXzAFuO2NOz08CwDWsIIzUE2SYFvv+pSxsrkjkZCQ/un3dC9bAZ4EB2M3 + uZB07URsQv84tNEO82orjD2OYRfyFi3YnmOT+1lVyNEhpsYyFgbdno8PnKTlSB+XTgWcjCoBSufn + gXq63/i8dNMgOiQGnf/qLxfvtBjRsTGxXiSYzD98gzmGt5nb4RvYAiFagDZ5whxH5cEXk46ZMK7V + HdZ/fOG6zPb+vveP/6vaeyrAyp5hvTq1xrpwpwXZAej/4g8EWXgrsH8ZKY6qBKRMtd4OmMeA4IvC + f4xZvXomwgpi0W7HY8ap+1mFRF4wTUrzARjv7BPYhu1IXXEyCH87cRAxzn3jwguOrXjOdwv8TnKE + 8xft0nV5/maW2LGi5SRGhDschBz6scZwNiZ9KZ7yRkJfdZ/Q2H+35RZ8mxGR9rTggrQum9KXpSry + 6k70CLumFbJaa+D7dRmpjcsA8MW79mDyqh7Rii4JYI1b54AyvaEuPgTGO5UiHbjW2tAkLL/GSqdJ + gameV9jaXwSybdGnA5NBD1jnVAj+9AqGpcTwcTpCwgTcO6jzZpN6t/ptbEPyqdC1GBbqqwIg09vs + FPTd9xCHKUjJkoVqg6o8K3F0vYtk+eTpiO684tGk+Z3xmBVHR+wD8l+969l2i1oJCstJo3Fu7gwm + CyuHMsYp2JuziTCQNT18msjDpXSejEm+3Th4NIpi3rcqAVt97gfgV8qKA+Wj+9sTr3cU+HWFfSY/ + AZsmx4Z7zZEi8OPXcn7BDWp8bNLw3AKDxrflhHS2m6bXYn/88fvQe3DhvSPFVExS2t9VCV3UwwEH + xZ0nFJteDm+facPaNw59sXw3JriK3YTVWuGNrZoPvw4t/6Duzd+ReSmkDZznSxEtZRKSVfZ0CZax + mUSibr7T9RKMA2isuKXx6e4ZC4uUBq0v+/2Lb0k47VqpMJM6Eq0/f7Q58+Igzc89ersFdrnCnV3D + j558qduUT/KXf6T1NJt3VVKWsxO9Iyh8Aoa15RGkDK1dAUEn1/jwNd7+toSzBDj6iXHMRV7LroV5 + QXK+f2LnSm7kp48b/OlDBGvQkfnpXiooWt4yc+g2kYX01gUWp3qgaeSc2TLZQwDd5POcmVijdnE/ + 1IRHjQ3USmGZshjRC1h/70QqEh/4vH5dOWhHzQU7Mj8YWyQ6d6j0kUi9ijzIuvPHCxTa3KQ29afy + T0/BjKQrzbx6S5fwlvdwuTZ7HG5yC7bOiz34WZTTvK5jbiwkUXKoas2RlknzJUu8cxP4lMpuflac + brAr22KUCXtAXVdRy39+aVcdpnntK7ccL/JhBr98zKs9fsBycyL45yejLTXidLmfYwFhs8ix69an + cuHO3xze1Od73r1BX07i4qpyYNkIq8fBTmnbDhFkytvG1g9vy+HZnWStPkrz8852xrRpSq8cnDzG + 1xe20tVr7io83JWJ/vmHeUjqOyIf+xntlm36lz8YR8kRO+X4TMf3ceRgILVZyETFbYVyYg4CHajn + 9ivajLdq2P/VQ+rpKCl//t4E150eRM243ghzz7WKnHx50/R4eZEljKMZXJ29ToOIU9iakI/6V78i + mFsT6dQxNaEFsESNtsjA4uhahOKOn2jYcGG7xUfvovz4MtNyDYioviQT/v2vpXPSpIs51R+w3y8N + zu/1yxg/m/uBf3oSaL5pcCX3scF5PhX0rElfY+2y2Ib9l5nU/dX/7Q/fjN+y6CumIllHuQlgJHsu + DgdjBNR5/gcAAP//JJ3b1mswFIUfyEVVVZZL51IqflTbO/RI1TFBnn4P9guQUesw57fSZPTkya8c + fEi/dceW/EXqtKl8ftEDNH0xHwRZ/1D/aeKM5STg5Ch8pj6nlUQf3lR8rPFFo+xmR+Pdbzh0Mq8J + DZzqWcyXl/OBi5ylVO+7YzEeUBqi175HOL2ZbrfDF/Uh85L09OGwORQ9j/0r9PqN0ItIpqjeFZEi + P5SdglVdNF1y0gMi39t42eGS6QUPJf+QTRbY+PkX39n0E+0Zse14oUlFtYiVQWHDFF33RNaNwV30 + NoAdnm0cXb8G6saB8ugtu2dsaxPvsmFQDPDP8+DfmiBmS71I5W4/Y6p/tWc25jwDaDYl+GOgk2KO + 3EaD943tyYY7B2gbm28RbrtqoKqwTHR09pyhLfQL2YofprOPbqbr9yG3M2wyMvt1BbO1jfyNIYZL + /9wFkBSPgrrDjhSsn+UQdTkfUsNierGtppsEkpG75LD8PuQDUowWfebvuNHo+nDaW5DuRAs/7Ert + aDC5AWjZ+MZO/zA6nptTDwalssirGT2X19iZX/UbgfzVuGMWP20kZlxFrX42I7aoMeAuAUez5y1w + 5/EXC2hqUELVUn9nLHl+W2Sik0jYbR478qwHIr+eaEPEccP0aX8/84BTcYv17VvtZvIKDfjtlBRf + Ua4yPhaZIHs31BP+Y9udUJvHHsx3meL8sPl1XddbKfzsRX5vU78b+9s7RH7ueTjKqzgaDlkWQCSV + P3oyxQD1q55e4ov6v3eajf0GhfAjyMPqhWvZzFDzAIdW9//+gR0yJ4bvk/dotNRj6gnXHjq1NPBx + nh134jO47lUZ83jxyx2LOV5c9T1VTvjIdqQaLDCev5Ca2bcqpibUWuj+Gs2fhr+8m3YNUdC2eR2x + Lf2krF38AlxlwrChTGNE7ezuwVKvcSqpddavfukPlV9qmBzok9t4ErKii0WVftNlsxHxrUxxecC3 + cf9yJ3HMRTi0R/BblzlRZf5wBet6FaT57ugncgluoDO/YaGtT35QfmRbekfY930bseHJ+UjeeQqN + L+1RX/iCj+w2P2M94Xo2/67qKEe1tCcQNVq03X6KQD4m7h6b3pBHU+ReLMmh5Z1aQ3nuSNCKwfo8 + fI4ARWw/Pnio768dvlXqtpiS/a1FsXDEFDPzs/pnW17Xz5FxX7QrT+LLq4bXeCMLv0I/txhJvejF + hQ98oLKymmoWl7pscx1t+ZbWI7X0aXbZYdlNJ/V1gs1Pukez33ClZNfbLz3cb140V3+hKC/P8/f5 + zc94+Rqk//PVhbHuaC1q/3kXNs8FiwbFyw30RFqIXVv6RMyjMgHTqO/YTeSv3phx8IJy+nvhI+XM + jC1XNEN2kA3suV/UTYKuHNE+Po3YP37lbnTzRII4nBm1Kqpl//2qNN2tVT+606DcbCiPvUHDqfrq + swfvcY0HMn7PY8F01hAw3pFKj4ueHg/8JgTuEnLUxqeNO6287vCNTzj/HF5opEzgYNHri19D0fBJ + Hun/fAyeg6MLvHZN4SP2st+c4ZlNQpMe0TtpRf8bAcrmixWVMN2EKz4erickIMPs5X7wGqoG5w2i + LpxEmLbOF9tqWUdk4QPw7p4/fLJ7vpinw9RD9h4VemXXTJ/iDklgvSLdj+/DoE8eZSNsiMVjdyf9 + ZYy3uxQWfYFxoXRsMuPgIz9KL8eHPrSi+V7fBGhuxhOnnyFEtVP/jNUf4MM4vBHVB6uXdj4N8OHN + /9B/PVnbxKYOL+FOwKWTosfn1+Nj3AZuX02YA213IPgoqXVEcv+qrP7M35oP2Z10SxqhmQeOnniL + spa7hvHa7/yJl35suquyiK7mp8ZuVeNoNBRWg5o9bcJVtNUncrAV9NVkn4wv20D8ogfkYq522LiF + fx3vFea85ic1MzftqACRLR+5yCJbKzKK2Tr0D5i6l4Vj+X4qhL0+8Wt9xav+FtwPE9DRMhDF2dR3 + zG+Vl7zwwoWPvt1BjY6AlnpA13im32CnyS4GHkd3RYlYyDXKmm9Uo+VyVrx5JCA5gUQTV/nLpsnX + Rhhvrw3F/OB0k7Dl/FW/Y3fhF//5YV0ziy7/AnR36/uNP1+n9pcaUV8kUohi9WD6U85r7pZzy0T2 + f3lG7Z7TC7prcgP9XbFD8un6dGdlHhV58bN+07QvNCRe2YK4STSfew0JG39dOkth/XtRFdsM9cft + yMF5XzW+7JMfG65KnEMWnSJ/3CopG2URXxGXmwPFoWNmbJecNailnKeq8/I6YiE7AT/3PWxs2cxY + UiktLDyYKlNqsV0tHgnIaaBT/TYHxfBxpARuWlFgRxX2HZMvwgNdIyRSzVFOxc6Kwxyt/WxzNLmi + /95lDTJZ87Ep0wcbLyx8gZY6Ez2+kVZQtXI5EHKupO7+7URT+nvZqP0xm+JO+DBhTMURshKpdI2v + xY8fYXkf9q1N040P7TbCwoOpjpyazfG3FSB4XPDi/19dd19uWUqc7kd2hui600loW+islP3vp9tX + 45TgPy8nf68K+2LyKBpR8jdvybj4TfrEai+N7/qE77WgumNVdhJUDrGwcxB7RLnQCZApZA494IHT + CTlFmrzqbe/I3Gi4k1b8318u3Fh2C0/J4fjwbapk4VAQHsmJ9PKWM4AaCPVxYw4eLLwNnyRbdwdf + vn2gii7f//xnZm/2klMgW7r6y7nbNDFYqPijdml7haDtpRGYMpywKn2oOyl7x0Ja6KnUVks7Yt3+ + GiJ2Kl2aXHY7NKa/xANUvQJ6WuKHDU/BX3kCaZbvOyqWasjSO7Wx7qFWHxPrr5Xjw6fEuMWdzkyp + zEGO+NKvOGFX5FdGUnA+9OmjrXVirBEfJRKD/opTVCsun8Z6CH8Cz9P4sW/WemuD+Dz8sMovR0DM + MSsRMwug+DmD25jx9QMZL2+wYtwYG42hbhGmtCRj9YFoTIe9DxX5MapdHh2brkr8kE3fl6mnynE2 + J7e0RsKbxdhoLj82fstAkZd+9N+v0W+wUaAY5gM1r/PczXzdprDwUqoHsC+mg23HIIyJ6kuLXp8M + xlokglouPCRnTAn3CtJjtaPe6Uf1cdWfj0Z1qXq9VV0jHZMZWtxJ2DY3OpuMg1XDfvf16H8+9cqz + CjmS0GM3QBNb+JQG6zzH/6Fq8RNvgKOz7FDPZynryatXUBlfCmzrchDVaz3VP9XV3xw6i/3vL/ou + O5ABv0T3vx9f+DYZbZ5mVd3FPdpDImOdGmHBzBf3knI7vGMlFD4FyQZ2lIsnTqkGhl5MZQFHULeh + Qb1pdvXa/UwiWuP1JB422erfoFeCAV+Mx8Zt1Q3lpMjlgMyD6xTr/A1F4maHrb0fZGOwTLi//CBR + N5FNnT7skEdPjT9gqykqdyxkI0TFXO7oSXjuELvexhyW+omVxHwzlv1aC7DGamws/WAL7jVf9ZC/ + n0Vw5/PzwEO0uSU0N6ar2z/8LQ+beV/60trfFr0F6mALZG8+DbT9yjIgeJBsrefZlEnMAtWqb1jb + +d7if08GzK1PCZx+2GUnoa+QnaGawO2vQf39DRJyLveAat/PI5rW9XCbp0MV5VpmJNjFCbDSaKh7 + /fbFLMpHCWli/6DGloVM2LeuAJLxcKn3OShslksuRQZ+xThxbrdlOu2/YO2f3E3Uiy1M7QNAEH/4 + 2eUCG73kTSA+Ix4fnjMfER5tE1koUgObR6K4wtBzR8iEkuJTnW/daeVjpP0ifPR/g06k9tL/n+e4 + W2tg09909qVUiJLV70Wj5f9dZXFoYzLuMyEatRuVYOmfZP7CMZsK/RmDPM7v1X+wycG6Ia/6B2sG + 1nc6exOZ3DsOu19117Hz829c+QQ9Ckws1nkQ4BOI2JP8RJ9TZQ5lv/3lWNEmLyN6hXpwPKzi40UX + 9dlGsScfnRPQ6zM8R31sPgEKcC9ETE2n2y31VdbtKceWtdyKGn9fEuy/+wfWdy+563MecatfxNbf + /ZWRvvoHAAD//6RdS5eyPLf8QQzknmSIgICABAURZ4CKgohcEiC//izs58ze2Tfs1a0NyU5V7drJ + Dg4hfPsKtUO0zZdOXgqoDdKNsI9ReezSxCW4Xy63YIa13s/xaQchyUr0y8+qOX8eHCgq6hOv12T0 + U6m/2z//3zbvD0OiY87BWiIbGqRHjwnld1tDaBQHahquHM0MPEpwOD9iUknlrWJJ43RQbANGNUHQ + 8ylR7AFOYgFxyh9egGb24QRJyN1+fls1vap7AksoQWpLz3c079SuhBZyt9R5xi1b/c4CrvkzTqIC + Rl/r09/VK86e1PzV11Y/DeShdQrkpekAe1I1U9f6JMaaZOSC8OpD1ZtOwa/+mrfDhoXwt15/+mvc + nKIA7QIf0cSYT94EJVCCjvgfIpdqwab5Qya41g+pocu28aefrt1IaFhXZT5fNKcB8scXqG0cI09I + Fq77q6/5J93tGe7iDO4vj3DF51PFVr/kf+lRoPz3jgLxPCz08r541ShtQh9ePoUTKPmg5eS5uB3U + oh0KBC1CFeu+CYeOD9PH1vyxmPg5myKqw3NBD8VW96Y+/sbQqd4SNdSDbYiX/K1CfXfa4EBgViRG + 9vhSn4GYUI+7Wj170sVBQzF9sNEYQkXHYq7h/fINsFcIblXhz7FAun174ET6IsD0RXFVZXsC1E71 + XS8Zc6shNKlxIOSCni/cfVrgrPcY65s2AbP/yDh4ccxh7fJqMl7j1zNzYvAOlMZO+mVsXhBczXcf + 1E8isDnaizpShe5GHmufPdEJOw5MXFPS83dXG+QkZQ0U32kZCNR/VCP5dAnUyrdHnUU45/z0SRsQ + 8rsZH3OuzoXL2c/gliY6PZm5Gs3zUHLQOKu7QOSuNVtKeJig9bHkAJqJko+84XVQFdobzcuTXol9 + 6MRIMW45jTx4zPmz7PgAPeILtjuwMZahjAKEzkCmVmi4YL43rxJs7KmljvnIc4E3jA55tl5R/dZc + 8jHeqCI81WaEXUEymLi1RR6ameph09gHBgWXbwvJiVjY04mVC/FmEdGsFZQa1+Olny1DGaBReBHW + lu8ZiK9pfiHx89bo9mrG1TB9wgZtk4NAr9amAcKyG2WId6ZGzcO2yMWhzAN00zcZ1ji5B6O4z1P4 + FaIDPUiR2s+pUaboeS9N7F1vF4Pf2XkCWZIe8P1dh7nITfEEgy106MN79hG7d6wEanemFH+r3pvL + tUtk63YFEZdRBLOHNRPW/npmtZhHY3iKVae+BSOke6koKibnZw1txrGmGS52hiSaago337jG+Whq + /RJ0wIQULCq10abs55cfQHjVgh7v6k/Zs7HO7kg3rMeqWC2DN/aohlrBE5o4E2CLaLsyfKjui8y4 + +7JlUfsEpscHw7rHDr0YKrwIls/ZoEkpOIDH73Q9s/DF9Ord3Fz81lYtKVOyo9txv60W5nIniLaC + S6NocfLZycc7dLxpwNtNLVVka4siVN7GhUjPIvSYkhkqGob7mW4vpcbm/bGLoe/rX7wtBdugyvEo + o4PxfOFj9N5F/KW0VJRm8omm7xJHUjkbJ9i7xwgHL79g4+XGpcBnAk/Nzo4B2wijCxuIRaqZ8yYa + WmsaUGaPFcXvFvdi+74M8L4dTvg0HzQmvcIHhGcMRzJ9xwDwHKILeCvrHUtYLxlNcqmFu/hd4+QJ + UiD26RIjjTNTnJINlw+x7mUwdNMUuxslY2Q2yQtSpc5o8PIh6z/bMEaG+7kQNbpoPc/w5YQ6fcmp + Jbz0XLjLVqn+xlvXyjsbIrW7wxZNMd5LBawGnk0WDITshHdNKFckAZqJMnlzoIFKDzm/eTkdWM6F + gG1xv6/EsE15sNcmRs5p6vY81TcLhGjKafEsZjAb4cWCtivFWL/UmiHZuK+Bar0jbHjXfc4QPaew + MdOBepm2MaaPsxVRxbHHuv6aiMm1rf/mA5+pv+mJEQS1ur4PUcr20wu2Xja/+SUScM+GxJxxgYf3 + 7o7tmjeqQascHXqqm+NH9wXeIsfgBTf908R5OVmeON5kCLYJFujBP8ceL7mnCWlTqNCHsjy9+fE+ + Zuh26p1AVMAzZ1tul4JXVXr4MPJeNRl71KjTqNV4ayQ4F4uGDyGQ70YgHHegmnveHqD+NXLqF1yQ + s1mpIHoUm4YaODYB1Ub1hSLaK0Epbr/5hPJJBUk7hrRQNq9IinY3HcbLrST1YQvzUZdOLsIJDGhi + EbOS4AeWAO9vGr1sdlNPXtNcQrJ1XHwrs60hwGx+wS9uX9QJwxGMx0X3Ufnsc6KkuK2+ZIoTVM7n + PXZudhhJRa5MME/1C94/v5Enhdo9A/GLLdSC+s2bZHt6IS8ZjvgaxwdjfR4HwfBTBqpnNMbk1rtB + EQUzoMfQGfspVq4JUN+Q4tjYB96IrmWGypvKUdtjOF+6F7qrt41oUFv9jMZ01tQQjpq/4MPVF6ul + 75IYvsBGob/xZm788lGiT3HAr3g6XZt7qzwDPqHn+Ea9yU/nOwJX84Mdol4BL8Y3GaJJjoP2qfeM + Ha4tD+vScAnnX1nO4vNHhcIu5wgnkOd6z2ceqHsxv+FdCk49zweWA3cmEANgfbeAH2o+QNtHF2Or + PGlAmN6aCWqtJ9hO9Xe1xi+PJIsrMBakXc9Xy3SCv/UBsF4COhZKDb2JQmzOmcF4Jux91Oi+T/h0 + CZiAej/944vTuyD50FoygXJwi6l3vUneHJpSCr9bwcDRy8kMsQ+1GIX3m4UfrxB6A1ddJ4iEY0xD + 4f7M52A06h8+4Z9eECTHDeFhe9HpkbvbPT++rhBVh6eAY9vc93wgB5byFL9Peq/vZSTgXi1gXLwq + 7HX47PEbVnfwh8eeAgyPvwSdCkxmtFT/qkewPL9pB7P+sSFysdWioejPJ7S7awbOYsDnS9HAEG5y + KQmQtr32c/ZCuvq8v0x8IqmQT9K2nVCxP22pu9Q8m50TKCC3NyZqXubKI3nyzNDKb9j7DgePP3xY + ADWSdXjf6SdvYcrJR93+DOgFuIIxxNEnQ8/J3NL46Y2MLLMSw/snhjR3DueK/fC4fDAfO06WsEk4 + Uw2ezEHC52Xbe0TbUhMCKBmkC28TYN5ITdjx+o64jjkDmvmTjj5fqwx4zn+CJVSgCFl6yvGO+utt + U2tXRtHga6q7y8ugK37AMDyfqRcafs9aR/Fhdg5m6oUN3zPdPPvwL/6st8am7rwU6L7NNLzNOTMX + XoDX4MrfZNN9c4NM/beGK14S5MVOPm8VmKmHpRnWrnDMIEe5ItDfFCwAZJSraRiJrzy8XUpvM7UN + ZoWQoPK4GalJ3mkufHZKDKlbhcGsFZtqUQcawkRf4p8+7tnudOBhtnXWrtvLrmc//uzgjWFz1Y8L + rzEHRtesx1tVM71x6MwGKp+DhL206Pvv8Zv6f/rU++53hpBlho76S4FxutlTwOqEEijJvoWx0fNR + E3TMgnPo3AixNk7PetqQX/zT+Jzl0V88oUdyoW4qUjbN2q2EXJzc6T4S99HyDgRTdaPmSveuqueU + dp0PgVwYOItUpxJ0Op9+epRazgRAHzymBumG+aDY6ON8fvKRC8s9H9KbDr7RsovvyU+fYt0/HqPp + ML19WM6XPZnT51DN3fsA4aqnaYoX1xPgcH5BuW/DYHVFKppEuxbmg72jfqnsgZhPmxQu28wm/DkD + +XAy3x30vLwiskDf+QKui4MUbGbY9vnc68ruu8Dfegqj4rnuoCkG6A3fw6ofNCCicqtC5qMRa7qv + ebz/NRN4iA4htoVm24sfo+Qh4UOdSP6RefRrtybQd+EGZ+r5E02HD/ABG857+vt/qx7NUHHot9Sy + +bifNUOqIa/ZDXZa+cXmy01M4bQ/OTQQ+SMToq2v/X4O4hl5jA+azoKB82yDBXobbzTErQ61ENwI + C89TPz3FvoXnZ9nRC/Z0wK7B1oVRYCNq+Q6OhJUvwAWUYvB6Pk+MrPkUWPGFWtzVBPNY7u9w31BK + /fflEE1PS2xg7VqH1WHRAf92BxO6UX3Fhbjvo5XPU/DTO4l5KcA0RKEP9vomwn7N7yJewdcUvABS + 6APkirEs85zAfo+2dGuRF5vOp3uo7oLkjC2/v+Qi7A4aQJL3wnsxGfo54GgNM7QFeFtPej9FRlGC + lb9psFbgm6cKO1ncwoS6emd7LCdEBYfzdVrxbDTmPJ1VdMbcSK+FW3vsG54suOIL3Y/ZrvpidC/A + mg9QSw+33kw1GkLbKmL8yKdtLqx6GGzj7Zlqa3ws/H50oJeQI41VPogGTFwIR1of8OVZhMYs4d6F + J7280AK0asTYhhH005eXghv6pX1fCNxKbKY69p7GEqw9StZ8lgiKpHuzessn8HqT/k8P9qCMWijT + gcPG833yRF3Yd+rKdwH94dVGQh1U7otD940t9s9f/IxC5Qbr/d8RweDAAfmpPOiPP1l7rNYdycAj + CinP1fzkcxeGPYxpnmkPj7zRUYcm130CdEwe/SzvWh32uYDJkr57sOSTlEI1Np9kWr5rfitfORAS + hyerfogWubm8oBXpLvWs7xO0WIQcWL7PihqG5kb0sLwJhAE9YhOntUfI+8hD54B8okgF7Oe9IydA + tT5RgGZqe4vIhTzIpz2/8pXhCcdvGoDvIwmJ/OVOrFNPxxNCVW7R3covC+XuA8gu1yPdrvnysiWz + DF0pFrC9aeV+cPN8nY8sDRT/nhjCmfsOIC7KCh+ETu8nXQWLEuUfD+tXJ2KiH4d3tFG0UyCAUvWm + 9mkNUB5fd4w7KzJmXDl3KMmBhQ11ot7SetsBqtWC8aFlpTFdb3ULx4f8oKYxJN6SOvICW5LJhHta + A2srM47h5XN3AnalwJhjyRGh/t3mdNvFotdTVvgQFv6DhioK83GXNDy88Qdz1ct9P0ubMIBSW6b4 + ztkaEDZz7KAN8BdqkIKv2KOcMvjLnzyDF/PBfmALrL3tAuajm7f4op2oNSov2IiWNm/cZCGwSuAZ + mwMs87mWnfbPb9F9/hotHpUa+OCqF5m6UGTzoZCJGm+xSLc6++TzcJ0JMpQY//HTwt3lCT4skgbC + LZ6radbOr7/537/roB+hgEugvjkazP7h5LGXuFWRBn0XH99RVw3Fqz9Bokov8vbN0BBma6PDy2XJ + sO95dTWJOTf96TmHaz6MCZdPiZi/GUm62WPQe9HVghLHmoB/W/ucJdGhA59ruuDi9hSjP315Y+qC + Aw/QauJhZ8ELeIl06whcvzyNZkBcEwzUWlzChtfQluAaaC4+PlNszO7VC+TjuR9Xf+rjzQ0XZn/5 + Mh9eMaCxuXbps2lBeGNP/l/fbyDA7vUqASYe6hS+QP8h4soHJIAPE4J9qmBLve9zcadqJ/RIm4Y6 + TbOJxs9ujqHmLBo+XOnaUOJhQyjfPR1vsf1kAzd4C+S6z4B3q7/UviDLwBhoa9NBM+qnW+c7cPVL + gvJ6T8G06k9wnNp9wLzrN5/MjZqBVS9R3ZZP3iIlZov295MW8EC953R9XmX18+ixvBiV9CinFJnw + blNzzipGHv2iIvcrfKguDhhQ15N91bhyjO6+O9MTrPNtAGr6eP7zeyY+b6DAPxyMU/HYDz99r+y+ + AJ/KXGei+xgGwLzdAzvpd2CsVFAIA6dq6VYrHv38AlBTn2UWYT+0PcCfT0mIUtIRurOdFvT5G8pw + GxvndTxMg415WSCpIzLGpE6qeZzHApqEU9bxfeTkDK8BlFHg0yCURDCa6Kn+9Aq+YVyBqdFaDt7C + osX5j1+up86FiSZbAfT9MGKH557AV6IjvCMWrqZ3lJjI/fR2MJnzI5pMSUsB75F3oKx+jPDTc8bn + vqWOuZXAvHemBP78CYt56j/98uefFZeP8eW+LERX1gpUu135fDGCoAGY5pBI1+ZqrPg2gF0dnQh9 + jrwxH7+hD2OSzuud02VFQ6LGYPUvibLqOUnt7ou64jvdq8fWWxx7gn/5xS32PE9Y4xmdK+pSL91E + /XLSzBZJexcS5Wl5hvg9dQ384Ud+NlND+PHn+xjy+Kb2gSHJta3B8/PVkdHJ0p6d9WVCd5536bnz + eO+zGZgDDLJ90mDFwyX7Ji/EJSWHHxfhzcjHP/GI13YNPTRPWC1y83hBTjZVevcArqaQu/Hg3lNI + mPO49myhrQhd59hjr1uQMQqKdoeidCJ//CFdg62DVAB6ai9zYsyrPgfYPpWk8ejkjZF85FAmowMR + nsvZINJunv78Pt/3w1yw/H6A70OwC+pVj7HUak9w4fkj1jZ5xJYZ8AWoy61LIxzXjKX2nEEi3l9E + WGrmDeEw8394bPx+n7h5K+OEC6iWpg5Y3uiqQ9Pzb9SZ+T2bdnaUgJfUKNisq220+oUcfJ7td7B+ + nzHdFs6Fn82FBEK06SM2LZ718xcwvko3j3ikKf75afYUVWLJogWGfiMH5Pnl2fcT9AncL9oNe7li + RLTi5en3e+pYwK7GBzEymIj+nSa32zlfSrib4Fr0DUSL1BW/6dQWLue7QE0MfIMF7auBw8FUcLjq + ZbGfNwGoUrEik04dg/F23EEHEhPb1mKAGfV+9tNLOFj5izVedId5mnzxYV1vs6h9U0gMPQ3g3eSi + 6fF5qlC0DJOagmznxLnGHNDF8osL/HW8MTtedRikOAna5TRG0+f6Fn/+II0F+RP91hda/dhAKKLG + Y/DhlvDnT91K9PVGr6xNYICSUsO+j4wY8FVCMbMXbB62MBqreVLhzlTEAOqS61VuE/iqWq0nFmql + iQj+ftcKWrOhmtdjj73CC6d+PFDTFX+rX/z98u9AEZi57tB1SuCMevLLN6OF1g6BZ5bKq16+enNx + aRc1hUVAuoB7VsvKB3/xo2Py9cZa1lpUUXvGmorClc/2IkRU22Fs7teePf4VQiEcBqIu2wFML36b + oeCjmNR0tJANe8/hoRLCL842SgYm54Y4uKPqkdoGkvOOTaWPQCLb+CFmh0rYwXACPmU7uiuD7c+f + ESHL7iBQbiFv0L49iGD1pwO+cF85ua/3cP/8l+KtnvM2K6IXummLSg0f3YyXJiMNnti0Xlx67BhD + TZ9CYQlbfNjsj/0Soy6G9d0BBC4BBUzWUxdKFizoSb9WxnQCoFbR+93QIJqtfr4pl1bNtq6BfckV + 2Ug1eoLKB0vkY+aFwU4WcWE6TCW1nkAG4+HdDeDG5AWfr3e1mlZ/C/zyhd0Tj96SHrvpHx5z0Gf8 + LpUSuHwuBhE6jzem1+5dwI8KyS9fA0TkUvGnT7Ge6gaj/SPiUVbLD7J81ZlNtH0OKlP4C0HP8mhM + dnI2Ie8Nb8JmvvUmXWULEJq+Je+1/jR8H6gBJixs/JivujeOadv89ApRvSbw2DrecFHwkczHAPUD + ymUZxNg4BxUQN95kTHIHd1op4vOqvwZ403WoXy9jwNV3LWL2027gWfcCrDsKAovgxDpC0v4VCJ5h + RvO2DTP4ZeaI9VqxcnbL2haseL7qy7hf2uJN1FmOFGw5U85W/4eD2A5LbElfxFiKTyGyrXtM5JVf + p023tHD7LCsaYzB4k4VC55dvkvdzOXvs8AXDT+8Tbs3vJvwOOehNI8T6Ot8jr15rUB7RGIA1P2fZ + J9Cg9T2m1Bwfbf7nj6x6hYCzKXt/69XXrh0NVP8dEdDfHdAC2aIhs6/RjK5tCtf6UrDuq6pEOWZr + D5T5gY04HIyJ6tICPM8+EnV5hPnQPoMBFKN+wwfx0PXsJUXajx+Dqqybajk5YocOPrfF0fUoVbOJ + dxDawzVc9QCthp/f+z74O3ws1L5nl7DKYCA1C3Ztx+35kurT7/mxEzKBkXU7G4inSKM/P3nif/ts + HXVD9e6mgOHnxz4cVaZBsXtGK98HaCt8jgRuqjP7itozRb1jMhrcyDViMTjyKHelHJuicmJijK0M + tr4IsWm2KGdJvmkhdJMj3VPjVv3yC1R1ZMZG8e29uqT6glRJRNhY85UlMRbn57dgzQykaH47UY0a + 1UioAaKLN7dm6f/wlu606NbPm8fkQLJ13UBOvz6QjuM3APZ02+B9LAnGeDmbKbzOxRlnnXcGU9mE + 089PC2Bzu+U/PwVuPq5JZDAX+awcHFld+T3g9OfOaPNTmEKpG2SciYixYYQfCEctWIK39Ob6GYxN + CxszG6jf3G6RmLeH7M+PyorAiCT51Gs/f5rAr7C2aX4fXiBEZYjNlT+Wrqd3sPqb1LaWCjBT0WQk + RuUJn2yU9BMmLgcXG32wbuZZxCI2cUjYahw1nu/FmItLOaE8cfZr/rqwORi9Gkabywf7HpuN0dbL + Go7zvaD+MkuMZf6k/fgP/+qF7HPiE/Bbz6s/k4vC5sv99B22S6U1xm2bpnC4IxvfN1PKFk4JFthv + 3j2NnUoG48u3IPTS+hGMq/82obqTwbDQmoi58qymsTc49BuPU0h58A5NKYOqxCOqS/gCJLnGmtrV + yYU6TiaC9vY8OdB/bhp6uDuoWn58/z/sKFD/e0fBzXYAPXqHNlqIDVOoh9zapb1L2GJJrQrVZcdj + I+3tindM4Y7YLk7xw3GIsczd6YTE13CiW+Hk54SNR4jil64GGxRaFa/ncQ0vdzAHXPTdRcLzti+g + erQ1rH+fNuDxwyjA5daZwWzsVoXuxg5AnecSjsZ11e7Cgw6E83aPi4yFgGV321fIQdqTzrZ7MIAY + lrA4iind64VZ8feUJNBoFgPj5LTrl3k/vZDqsxBrZ+HFeLtMM8RZ8Y1GV37uKWN8gAqVUGzkolyx + 69XJkFqUF6zv1QQI+Q64sPTaGz2FqK/YrpU5GPskwVa7WExcwNeHlt7ZdJuUndFxzIYw0Lz7eoar + y4Wl515wHwsvmjsS9Zbv1GgQb/s4kDdXykZLPBRAqgMHbw/eNhKfswrBucQC9e5xmw/x1zLRXZsD + emmya768XV8FIVhaeigfO8bPrHohyRgtGu5uu1yolqiDJ1GdSb3BERC1Z+Kit7CP6eGR7QHbjdKC + hIP3xv4SVMbgXX0Tqprl451flLm0y9+hmoSDhLNNw+csh18e6GK0pYeD9s5ncUYyfNdO+Xtfj42y + JiI+OmXUQq4ZiTOsFzjo45Yeg9jL+TDkfURuTYi3fnLqhbr2M7U9btMACd69nxt7lCHaXzF98Phq + LLJZxggp8hF79P1iont3Ldi9DRnrB3E9c/J4WMpbLjDNmNtWc0CVAR7Xe4zMaX56TPHcBuohvFOr + 40dA53unqY9OfdHt4QH72UmrTDnLu5Rq73Tuh3vaxEgTm5Bex26p6G3zfUHSFC2+uq2R8zPrS6hM + lx22qKmy4dK7/lqx3FE99475stmkFtIC3JE0v1cVT6eeh+jqJvQuf8dqZF4xqcktBzgIxRtjJ06N + YfmAPk5mPQLSNfMg3Nf1l8b3h8yW9M37qDXfPr3ZG6kaIJ8RiI4nRv35yucMBPoJurukproxbbxl + PxwTKBn3A9ZDpwCzNgYpiB4lxqe+CsBUftEJBEMyEfF8iatlGziWcuWsiYDE2VfjdtB5tCNr1207 + IMZ8FT4ETWfnTbMpOjGReygygOVLDpLHc/CW9h69ID5YPd0fNyxf3FDMUGOqOfV8poEJH6UF2s8G + ETiVUk5TWymgcSpifGruzJhfT3NA549VB0q4VXuWoTGD8sF80GgbnvMx5SYX5r624Hv03eXidnmb + QPLqLz5/dKMSsIVfcJCmnhpK/qrm9GSd0C9+DZwNYL6MXx9Nz2YOVClbz7CQekFK0JypaXydXDR0 + uYSZRA0yeX3vMVGQE8RDd8a7u1iziXDMQk5lMpztz0fAOH0Z0NVPttRBFwcIk6iXQFgOId5bw1zR + 913MQOJfbXwAu70nas+7C6ARzzSjQQJodA9rqH95ATvuUYrm+HpqkGRrWxynB5MRR20byIHMppi6 + yJjkJ+5gvz14eKv7KqBgLO8o9KIGH+zNpZrOzhDDBRYavQ/PmycMk+GiRNAbusOfYyRF1SVDk7i/ + 0n0Tvap22qo8mJMuX+c38CTuvAvgW7M7bLjHF1gimCcQb78xfVAXeYz7XAZ4eiV76s7Otl/H64W6 + avLwubJmMK3tY0FYgR0O1E+Tz+S9ceH1y7bYi9SqmvSRWiC+ui9qP8+oetdzaKLuUeQUJ6d3v8BC + 9WEQ3w704HtJz2bOCeDp9NbxpXR2+bJrjhwSN86NRqjHldhhjSChUAe87z4ILOFn60KpfmyJHOfU + YM9dycEtVD7Uiq553xfa6EJYsjmoAxtES3x4JSjBg45/8cXOsVkjNe54uiePZ07ffqUifMxyuktl + 3xDZeOXgaT90+OgdnJy/0sMdFpvjhZ7DKe97y88K2NmNjA2de3qsdy4hvIr4gO3ltQdkAd8AqovN + U0cNd0D49DiGgXNUsbHzbY+6C2iAVe7sdf0+jEnZdCeVlhagrnW6GWTFa+CPyoAN93w0+Hq9t3vT + NgH1qV1684i/GfxY6ZFGU3Y2SDI5Plzq+kbPY/jtl2a/TVD9Wu9F/B62Hv9JbjwUpVDFWsL5TEhI + PSExeW0wfgtxNI/6mcDeLC3yVTd5RI8bJIPN+xjRoNxWHn+KgAx3C9AIS3u7F98d6f7i2bw/UjCe + VEkGT8l4YMs1T2Dcy9sabS+xR12YV7nIHjpB06g7eF+fv2xxQy5TnaT94OIhDxVLzHlCo6ZA7Iv3 + yWCiMMWoOuZrGPlePk+VrMNrOh2ou9uZYL4BagKOlTU+hiczkg6fWkN0mDc4wCPrCRG+IpS3vUOd + Y+x7vOlvJ7TkZIcvu+6eS4OdE7C9JB6NfbXyeOXhlOikSwAfDsorX+7IfcHqegypKz/mik2H3ITN + GWAi8junmrvRuCMrny/Yej52Hh9dNg68F7Ak1en2rNr2fDrB9urxAQd4rhr1kKp/fOK2UAf8S9Va + hPdswkETSUaZ+LQBP31VfHDRCw9PLdHR/Gr0EBxQNZunWkMgG09Uu/kLaEEMX/CIwjs+8fy3H81T + rYPrs3vj/XET5Qs5ayY6G1tCo6/x9r7GUxSRN43N2sX/WP3wBWR3sSdsja9BuG4dlNj8I+Auyikf + Nwh20IvCBkeH9M2W0ZgdtE3fcsDaIMmHjfNModF/ahz4Tcgmcj/GALbfGD/U2x0s/nO7ID5PRezN + 0sUYW3g/gVWv0cMtfTG6A0YBragDdJd0jceQwZ3A0H9SIlvbLmdhHsqIfTsbW9V89aZVP8PiULZk + 0T9nT+wjA0Le6ynhit0xYl49xNDK2QXbcFh70Bi29Tc+qSlr0XwSZgs+kprR+zof9O33Kvzp3fj5 + GvtFanYdMn2rISJV9j0fneT4b734l2FkC70aMap8waQxTO/eYhcHDojJySLSbTf1THJcGR6VTUUt + Y3s32J5tHLDiObZdqnojX1QEiJ+KYZsvX4B9tnEMmmvgkjiAnbHkWd2pxa7fYadXcc50BSdw7rHx + F88sBGoH5/5gBPMxD/K/9yuHNKUn+tAqAWV5AJZ82K14SSuWSESD6LaRqK8dDr1wWTPA+gQqasrO + 0RM72hVwnzo3qtfFYMxFuuvgFhcfGprt6idCxYIOKV1qNFqYT4fCWGAeJi/yGlqXTcrmdUIgcTvs + rPNFArMP4YGNJ3qMrrYxbzwtRdPyftOITxrAikKzkDnuMvIUTn5EY76JUZ4WHD2HwyEXH+sZkU2I + arzPNQ3wv3gaRzug2utI+mVbZjWcNFHEQbzgaqa7Wwabd5dgfDim3sy6BcJ1Pqh+3bv5kvLeHSg0 + DnF2+249yW8/DaQD22CdNFU/Tc/VARY2C+GgklXDq1dS2O9Ngu9Bve/n2PNi+Ht+58OhfhlOTgb3 + pbYNyGZ4eQQLaQqv7+BKKBG3bFaaJoGSPE3ULn1W0V0+hmCIRoH0y7Bjku/seMiIT+kPL6fvZ1iA + 9N1+qYEoYr3dTx2I93lOvcfhnbM2IuLf92nI3hjTPSWxKu8CiLc3STVo+C1daEiTRPPHYRdN6hzp + MOtDiZ6bA89mIS9diBc3ofh6dXoBDeMAf+PxEMz1DOS3qFVleq8nAzFfzbE8rXxwslb8Xtiys9gE + WGNGNHP8D5vdgrRgSofrP3367poO6im/xbe2N3tx11whWDyOw8a5NvslCt4ycrABsRVdQTUY2VUH + 1sM84XzjHHvJ2gQDyKjwoDp9lBX5IoXA4/i8UC1auy5T5W2huyDK9EHEJ5t+/FdcIU+x7HWsu3Wy + Dq3StqnX1OdoOSzpAP14KvDduzX9suov9dKwM1GaSO//9B7NZ48+XPdszDquY7S+P7Y/uDXYPMEX + zBVfobfsTIwZiDNBx6PtB/xvvq7w6kILn71g1v21reIES0g2bxn7DXhG7KcHrA/X4a2x/faLvP84 + cO9ck4BHhwFMn+r7gsbpHtM9p08GnQ65BVUUnih+vu186tIjRGaZJQTI24DNGzhbkKktCGbqxP1Q + ShoHVv6gxz3GFVOL4wut8URQL577vj93HNBv2Rev/MDoqmdhMtM+UIPs0E+1bvvQf/otPfrJUrFU + HQdwdpUj9b1QN5Y0+g5A3HIXunuaHRhe9zcBqW/L1E6krlqWA5vg+WPWeNWHBhXboQXKZWyww46b + fGpOpxoZ9lGgW3Pz6heJHWN01MyU+g956Nf46SDngh01y2tVUbB8TnDVH0TQExEsdyry0EWcTs0f + vj7nhYO+89kHX6hYXrtcDjqSat+hu+YQs/F20S044jag5iP+9F8jKCfYcHODc171mGTuswWq10NB + D9dyky8/faRtywan0akGxDDWZuWf/R0nm0LOJ5uRAiQhkbD5Ka18Vt/dBIVnUAQVWx3zVd+jz8f1 + qfM1lFV/1zEM5AliRz4dDPHkdh1EXKFQ42Xe+3Z6HjUUTaFHTydLqNg27mJo2T7FznP49vU+MVNQ + LIJPjbfNoq59PEyY7biYeiJc2IwUJQPGw0vpnrXeT4+UEGT0FICBjMb7JpYnVGmpQiQ4PKtRmpYE + jjW3EOV9Z2xp/a2GVrwPgGqc2XyGUAcnyxVxUKil8cMr8FgWa8XXoF/5Vv35Nzhu3698DhclhcUm + ulAs6VrFi+6soliRbzhC9MYWazf5aOVfrI31FK36bYHOYcLU3XJP8P3pB3eDKbV2/A6MQtlwMFWO + G3yIL+eo228+KeyqxQtkXbWj6fgyHLTyFcbfufFGg7ol/ICBD6opE7zpNfBQzaj0wFtDtSrpoIgl + PJw2F+z13+PqIHMp2G2dI81FeAJUtyITlUOWUjvVYjDZrLnDrHJzgiapiuZ9J8jwvlSILGWp5kNw + 4F+/+CCwNr2eZMk3A5PGi0QZnm0/Dfmxha+g2//wEcyzc7PAzo2eRHnEcO2pkkM4OZsT3gfw3kvx + VrUgfl0xddf8s/VT7QViQrkAvVsMmN1PLYSCXeJ9sfcM5s2TBt3NgWL3tXjeaMetA79ZOlPz4jhg + OW9yDa5+B1lOm1MvLge2qLDt40CaR+DRWr3GgPav0+pvFNVIzSuP3MZ80ONoHfpF5GMVagc404tX + t2zhtGiAWBw4MjVPUC3DO7nDNV+ga/4LyLyXX3CbfmR6kLjGW/T6xsNhA2wcSKWRz7N6WOCu8hTs + VN0rnzbg1oFs0zT4dvZrg3XVEqBl1r4B07mnQTiyd+DqtwSCdzFzoTu8Q+ANuYj9D4b98jQPJVzx + i9r7JjSm5MpCOKXkSpr7gKIZtl0BIZ4gfpySiPEg5l9QKpUD1VK4Y9Iu+5JfPBEFKpYh7h+HFkbF + IcbFd//Jp7l5iPDC8BM7zbyL+IdsvhBrrIhw3s2qpMEKXfTL53dev6mYezry6I2O04pneiU6Y6GC + sPY4cpMDls/Bi+hA31sqPegqAEsmeQ6059gkoj7qbI6vWQOHjWLjy3gjxsB/whQdW/jPn1oml7fg + 6t8RgQavaIFobb0xVzJ1NW1v8OgCfUDh08J2my45rcdD+9NX2BFj3pimrSoC8YS8n59WUWn4yOiQ + 2yyYpCXxpOtVy5BZcxzFh6PsLSQLZFg4UAuAvTBjedb0BCI5+2Cf6wLGzzvOhEztQIAUQzdWfzSF + qV2FVEu4ATDuFco/vUe302GqBnXONSBzQhBIbbpEhF6NBD1f44S91d+cD2FeQuGbyKs/Nkfzhr4z + GExdSWR9l1bLqUAubLqbhLHO3h5Ti+sLJqkQkd/zkXDtLnQKjqeAIYrA3PnPAV3EsqDR6s9OUoBc + 9ZdfifcB5UMY8gEymI8Jf82QQazu5sNDvmPY00cpZ8LO83/+dVBYnhUteTa0YObtPIBD6ObLJH9f + 8GA9Pfp4C3w09N1wB128NQkA/gPMdW2mCHF35ZdPrv4YHcC7HR16gO7b+MM/ZTrvSB67IVhyJ3Og + ZQc0EO5b0Rt/6+U91QPVSXdks/EUeUjd6YL9dN6zCbycdQcadvDhoO3yTw6/IuQ2HCWQ2pox5Mcq + g4/cy4k84zRnKt6JcPVjsXMtXTDnsZ7+4om6SqxFi8l/A7D6h3/rQfqeqhCGuTbS2/2dG1KSYh4M + slHToOCCaLk8J6gik/9g//ASc/b7+1++aP9uOS0fZQH3A90FnHH8MEaKtwrla9kHtbANPPaUb/7a + 6TLAuwqU/ZwUWQgmTofYbmPTWPV3rSZmVa96+taT9fnUH37YZ1L2SzxqL2iHzR5r9PX1yLWeOtSm + KqDbxftGI4mGP34N0HaewZQ9lhZy8s6nfjp/AXt7bQzX9UhWvcXGbjQKuOLjut623tzESg1XfxLr + qx86xOJ1gd+zuKeHmi6MndlrQt/tK6M//pnv7VVfTzw0ZMVrJjhWKKOr8nKC+XwpPT5Rzzr8+Qc7 + 9Vkb84ukGcxrxaXXXfXpmQLZCT7J3g42K38Ov3rEWh8I5IpMPRH0IQB8FGY/PcJYkF59+G1EFIjf + zcOY8k3K/+nt1Z+u/vw/8fNkZP18RZPU5pUi+SrYOl/4frbOFYFaf50pvuS4n3avRwmleFRJ61mW + sT6/Cqr1xgewceae3ucsBK2m13jbimXPNk3hQmX6cAT632u+dNghcHfa71b+GKtp+61EpAopxe7s + PKu5nC4lbMvpga0V73p42pXIuc9GoKb1O1o2x1cHeD4KqNd/Z4Pdw6kEpm829JJqPJsuN66EKz79 + 8oO+3bd3Wd0aWkHmrRGyOQHyHV41uaOZ65bVzNDuDt9G+qZux33Zn75OCjfAWnPQIn7r7loYyKVB + dxIp8uV8V07wY385jDVhiJb35Z//SgbpkxlTuc9fwPs0Z6pNPfvTrzA+ahb+q8/o9ZlHVUxaagbQ + NX78pvZf8UvNa0IYO45VBlOeXoJlzTd5/rPiZXhtf356v1z0bIGlMak4t75fb3FyPUOh4blUo6+9 + sfjP/fKr7xD+bD0Ze0nHFkqpJpDprMXecvOuIlQ+9xFvT3UaDV/b0JSf/+j04tOYgtzx4eNeJDgv + G9izkBAO6ntTDYaoKNmcWxxRV/7FRyl8e0323RQ/vUCkoX1E08UbNLD67bh4bXA17xY9gK3SioEg + ZVtD/OlHiBeIXb/gAWu1cPmrp/hLYBjL69U6sBdPAt35hRYtDRBqsP5/uh+7U8/uddBANN532Dnd + EmN5BMcXUD7FiE1bEfJZ3e01+L4MMTWu/FwxXbETZZbSZK0flGyBmLzAn/+24tlf/dA/Pp5UX/2u + JYJRDNlxqoO5Ek5sgtHEgcMWVtRYisLoToXg/tUTLEMcGKsPtIAr/wYyQbU3S8fC+fk7hK+/U7Vw + zOZgOaEtjq9JwHgYyZxaXDkeBxdl+eEjh3TOrfG2WW9v0tVlQSv/YK155v3U1G8TIuBxAfcMQPRN + 5/oOVz6i5/40VfN1amXwtXZ6wDN4MmaWlgXycmdPo5gKjNEg6aDGjucAHRQ9Eso+HeBGeesEIKvr + x4cvWj99gdeGK8ZsC3sVZpu6wcXKXxIY1+8LOJFMT6fuO65Ed8jcvYK92J3Y8hLEAB4+kYV1RXlH + K14MP3z41UcrPjpNMVr1HfWq2ylfIG5KqJa8jfftWwQT3+YQamf+jG3ps+snP3VKKC+ui53gsDe+ + 8vmSwn3dfPH+gnxDUNiiodXPo7966rRvExnaZ2fGVlNrYGoFfvWnCIe1cr2F9FdPWPEV39WPlc+/ + z2sHbiZKsfc8/qefJM07BbDT60jcbEILLbWUUJ/rCJvTuS7gKwD6n38uPLL9BPWZXPHqp0STsiSN + ygWBRuSXEPUUu1uIFDAV+HG9tv0UXaTVT1kkHORCbTAQuCeIFPVI9Yjf5tPP/67BrJIhFlLvp5fB + uTwI+HAtHzl76M8Qftx1h+uHu1WTe9Z4VG9lSjHZRgb1nZ0IzfnqEoFco375hmYIeTfTA2H1P7/y + +ZGpe+3JqBUGbrX83m/1T+lFx5rxPYT5C7BdkmLNbMWK1eoxhpvC1SlmFgEDby8Bsm/HA3VMWcul + tZ4NltY9ksl9bT1enXNd/UzHG/nKKM6n0/abwkMnqvTHH4Rklvy/9CgA/72j4P3QvjTaRmuXl9dJ + g9JZw/ioypaxCO05gDRV3UC6tvu81elBVoXmkBIgGYKxtB/tjuh9VqhX7ns23y2ygI9mhwRNScXE + tDAhlEDA463tvavFK3keHS99u7JA5wmLoIYwGKYCm/fC7ie5IxkEUD0GU09JRUIrPcGvfjjiSz9q + nqQdUAxL+Ttj3F2lfLqCYwFtTnqR5/Ei51OXKR3Uh/SMLzdF80RlE9bqsR0NrJ/klPF3PnBhXmc1 + NT/s0y/q3a7hc6PM2BaTE1jgQBsoZm5A+Nn9AHqLNQ6d9+cd9ffXFyCFGiboUpRPnEw1qKYp42I4 + T3c/mKzHpZdK1Zbhy6iW4CO+Ck88KNSFr3fXBtWrU6qBu1x52KumQzp83FaCPMgaVIM0xPkDez17 + vFoOaiYf09yTXwaj6myiw6nf0kPmb3JmUk8HJ9H7BgvLrsZUwu8LkWgfUx/dBUBeo5Op12oesVeY + 91w6B1WAlIsq0J35nPNlSN8u8o7Hd9DDYDDoUw45VJ/hl+5TkYG5kC0CR+jvcRR3YS4Fh9yFwL4f + ycZ8m9WSXvUWZvl2xHpSNnn3iHSC+q6Sgnz9PuJeaxHQO1OoldYOEKwxKOEIgz3d36II8O8lUpFr + fmrSRNcZMLOWB3jAW40+FPtSjeRCZJgNtoP1wsBMnBVVhDuDuwRCYGa9NL73JrzM8EWL990wpDyg + axfN+Yb3NHV7sR0PFvRuXEb9DnJs+iY9B+WyE3BQ40NFFKtskVV2KdVotYsE7tGaSPRzRLEk1GAm + cd2gq+aNGJddDea98JSRuz1lZHr5u7xzPJ+H3d0Z6PajYENcvHOANk9VwXs3LpiYdQMEZ/rsaBZe + emO6ctECzpxeYv/dDfnSPGkGKBe7+Lgkcj+7btCAJNlaNLb4T9Ttd1oArmFR0SIOOEaTFL3gOl7U + y5pXvuTQDuGmftyp7fgA9JJx9WF7qwQcHM+mITlhOsGQN1pqUXAylv6A73Cf6ddg3jxxPkbRokPY + EAm7Md974+54lWE1HwN8rApqzDp4iuilzgt97B9mJVpCMUFll4REZLQFkxYdQ/CS8SHQv33TEz6S + GzT2DqG5M/Jsth6Kqqx/Tw96vfFoHWx1+F7aDl93j/8j6cySVoWhILwgH2SSJI/MMgeZhDcmURRF + hgBZ/S3/u4Lk5Jzq/iqV6tgRHwznFX63fca4jLSRu7/WDNSZSojJF1G0vwJPE2utJNg4GZbDq+2o + AKExMA4ioKt89ox8+LiEKlFeBaFzCtwU8l8rXyANVYdZFE1CWCpq7CSTUnJ/+qUsUCQ5GmzA300C + YSp/7sTto2Qk4D34kDF8b+l/9e2jd25gMvktzm2TdrtY6gFiPxsibvl9AD7RMhHGsvolzmKxzvq3 + 3sspBmzjRR452w18qJh7QuKjfXEWOL17aB2bHDv73kTsxaxSeInjxj+tXxTtRxhCFLjbmxiBAqI9 + UXAAYtzeiKZczyVfakmPbm9R9tkBbeNkrTWENd9L/olpYofJD+UOJyou/twZx2inbtujcz5luK4Z + F2xsaxyg4NEc62FyKbkPUxvQ1l5PbOi52XGVVCu/FHSGnLdqB0t7zTI4JsW8CHCLujUYhwIhooRY + S8TCoURCO7jWkeHv+6sfv3CYQwhGhsMRvuXlktdZjH76TGKQFuVnoiOEJ4MtfFqvp267p6GJ3pIe + YJV1127K9ZGDysFNcMNbHKAxMjlYILsj5e+8mLsj9/CWMj3BkLzUDVqCBs1wz302PYjdf/9Z43Yl + xqkpR6pLrwINcjX5r6UvKKUPtUF77qX+LAMWbCfGlFB33FXifs9TObFqZcA9xyk5xw9JXbebK0EY + izw+b4E18g56L6jPjK/PIRJGU30bHqB1Whnn4Vd2Fp6KAqyjg4R9056jaRbBAa6GLBI97U2HQ7MI + oZ6f79gvsOasaXNpgH0Kjz+/dACv9O8v8O2hJ5IoLXQWkAEh7eIdu7GSdIxXchL8mw/pUPtgXq69 + KH7W8kLMI5TAMnq4guP1GBK51YWRNk0UogGqHJbYMHOYrgMCdNq7hP3bPe2YiesDNHDmGRfipwb7 + qEmPX1LIlThvpJTcO2srKI3SiZw7Jxy5E373kBm/HvFk4o58/vmI0AuMCuPw5QHi9L9UaH9MsYu+ + R5XmLNvCpLR0EqzDAtZ6ABxc5YODccv5Kn+4IFHMMjBhjauKssOXlkG88ME4u/RtR63rZ4VWyeVE + kj1dXRcrVv70+6+fgOafVgDGo8pw1vFSxzp+HMBv2Z+xbVzlbo+GWIDj1Z1JUuJLyV4FRUJK8Uzw + jfH4kb8R1ENbeBSk/tXD1Ld7i6oqt3xWKEqH1vVZRE99oOSSXrJyT05mAPujzPiMIwYjfy0kEXHz + 7uPzz993jswu5NN7TgolZFQq2UCCmbo/lmwovirtjE1BLPfwFxbePiP1XLxC9mL4WA8tU93xUXJR + qUQLdnnglrTdTIhQmYULDfJTt5ejECM1zgAxFUEq90Q5h0iNvuli+a9o5F+3MYWl3FnYs2u1ZJPp + GcOQswZS2IzcbSfeTpH5Sg8L0AIG7O37HUP+lDFEA+nvH2ajX5HRcNc/Pe6YY3PnRKtkcnz79Ys1 + hcmFERBu+E/vdn3DGlBPWood1R5LoqvvBbyqSsMZ92rBPmzFDp/Q8rE6NSe6tM+XArv5oBIlVPty + NkHfor/9+Udfi9bvxw+Bq7sl1n/9XQ91U8GG+j12bmUMmGMzMNDEIsLODY/dYruZD635k5MqMpOI + /HgLQH4NfUa8nwEfDHgVzeczJ9jir+ovlMeEYNQu2EnvDCURu3KoR+2XZJ7VdtNlPxWIVw8Mth/V + HfAL12XIr279Avb9EE3Pdnui/VgCYvSZSRdBrwN4B4xCsmfcgi0hywdWx+BJnLl1wCoXQwjvX0nB + Pz2lW5HYBqxDW8IOxzt0CzT/Cz6TMv5C7IxyFdrKANngPnAzJCnluys+wLgzW+LzVgq2KBIVKD6U + C3Z5ojo//jHARqWN1LHoUuZFLk/Y3mwZa90iOlROvBjaKnf0J7ltRjoYaALwBUOiccHFYdImryAr + SRQr+y9Ve/aqHp5L70H0dhcBNdMkE1+tpmOvP6rjntdBCj0ukpftwmQRM/PbFzXU7XEwGnpE1TeV + YPcDVTVaA3V/BboGqrl6kmvnhL8bZClEtay9Fv4dSM7G+NMHdk25E0Uw19+LklqA96Pl/vjtO+4D + mA34OQgG0brHCvbMX2P084//80fUN1DgrC8L9oKBc2jErsx/vvUOtgo27ll9YX+4YJyB8etsmi2L + qGple4F3uVP3ipopGFEZEge0ibo+ssYFQZt/sPx8CN0o2VQScXJ/E3/jtZJea8eHS9fYGAuh7sxO + 1vfABsYRy+/5Ral3oAa8nRYZn/OPFY2nymnBH3/4yT8AAAD//6R9Sc+DvNre/vyKT2cbHQUIYLs7 + phACxCZAJqmqgAwMIYTBBiz1v1fkeVupUnddR1GCsa/ptn2vhogjScdqLz0jsh9blE6nb2uAY3ds + GHnvKjCbBbOURR+wPcEQjLi0DXD9XAuyDyU3EuVZaNSAqneyPdfNP34gzLnFDt5j7dEFP+AhlzPm + JSuB99prkJczbGtmJd7IqWyEEP7+716pUMSgYhbAY3OCVeNxj4T4ji34W39BF+JyIxkHAXjPe0X0 + Ka9N/i6GBjSTJzHtKr6j0ei9Gsrug+DmcNqD+XHeaip2djqxR//rjXqSB0iDVszM4nAre1XyLYDm + 0x13+/PKm6/PXQB5eLzQwdbKaCxryQdrr4+YZhibcvCutIab0yywbMNKczqv7BaCkWTMHqu0ZD/8 + rdemQFmrL2cGQ1mFl7zrifXcBmC819ML1OlNIASC5RZyZxLQsYsapq9Ptjkdt4kGpznwf3xZThcj + S+Cib9hBFJ7l/NMzY30e8fya6nKgxzZQWdK7eN7nMphn7+7DYYwUcuiplI6/8b5+koJ4R9nrptvV + T+A0Zj6J7LeSzqvZaeAMYh+Pi35te5xI8KzeOios+mhe/BN8V57APCf3OrbwGyRv6U6VvGq7sSHp + CrKr7LLE+8YmN87XHvz8Ep7NoewjTG3Q2cqGOM85Mid0DSxkh16x+DM74pKxFdBN2w/snD8aPn4c + DwKKR0oCF27K8d3sK/ARDJtoHdjwpYWHAadAAexg+303ZmkqASE9XYhzOH3B1FwKDMv1aJLD5yJ5 + fPLTDFbbjuPVIVuDqdJqAe5s64pzs5S9mdG7Cx+er7F99OTldAtgAoLpbrLHurPS2TGhAOuvs2Vp + VXd88rzbChDteieL3uFD8swSZIwzYHt8ayP6XX0SKE3bGx1FG/Nx+j7O8JJ/e3K/eWW58YZgBvxY + zMTXDANI4dOt4Ky8fJKMSpdOBvhKSp4dJ/Iwm3U3yP1o/OOfoHkx+/RiO9DK6Jlsn8gxuSQo7h/e + 22/g8LkYjR5eYdkT//I+Rd1o6C4i4dogB2Uw+TS4SQB9tUnZo5wP6QyzawPn7fuMQWgJfFjwSfZ8 + JyGngzylszizFxwNOjLjFdsl7188Qe2LhczWYFO2d6ugsNwHDQVDXJTjhYEZZtPGxcoL5ClzH+OM + enII8RCrPZ9//DjNoc800cagdaMXhhYtzkwrrricJoHK4JIVOXFv2xzMX/C2UaqNFTkvfqDr75OF + VoocseAebiNR8IozcuKoJIQ+LLDo8QppUFIw3442H1zXrkF8jF9Y8OgnHb4PFYPVRZOJrb8nPjNx + cNTleTBE7dN8wgm8/vISeW9FJvUbS0BsTlWyhd2Rj+hsP0AsrF5siy2143ACBXSq+kbwbB66byXF + 2Y9vsIji1BznsT7DNKwq9pvP8/tgjOh4+TbMROouHV2ha+HzI+vklrVuOfk3ZQWqj2GxvfZuo8Xv + CvAOOkbhdjpFvC4nDckPizDcvKp0Ojpx/MNHZsc7B8y60Y8wfXt7urkrmsnfX7GBp5X2Isv76iTx + nchguFZ7FgJodzwUhgdcIXKnw1gst/LfFAg9X12TrZwAsw+RSVHXhA7RveZdjnZ0h5CWmYuV7JGn + s88yCcwn2SdH07560tnOIBRW/ozf55XazVa9dJWqbJVtq00T/fAY/fRIktio7C/99EA7W0LMe6YC + aJ7irVDytPTxCtRnPmWtuvRlD6zFr07m9DzZM3Te8YqFm43ifVHDNDhYcE+yFj74fNaCM2r38444 + w6b15mSCK/jV6I5celeLJC+RMSRv4Y+vvHEubz1Yvs/0m7IFm9sMfPV1iHWSqEGXTpumiFGDLJsF + xZWWVbPVGzTsdZlYgTDyhX996J7aNfHZQSznvJkpTNbMoZuA7qMJNR8DPqikMb1xa1P6raca9C2J + RH/seqc2GwhuwZEq246Xo2fbBtrCIcflCBQ+WjpowT6rVlg25Fe68ImMoptpYUUu9a4/zlPy0+cY + yl8z3QABSaD0Ngc8vxMGRj2fQrXsHZPsJkuLxsu8suE7e1iLX1ZN6i8V47/xulgw+uqbzP3xz5+/ + 4+b9tZyh97/sKr8g50J9dGDvyQ65znFrcpcRqvKiOrB9XrUl35g3DIddQSmsI7GbDrkO0YMKGsmY + 7qc//EM3wSrZ5aECwH3hNSJFnQO2e2y/0TiXRwoPYpFR8EFFJJoBfkFhN1yI+9l2Ji/34AyztKmY + 12yKjrn99gqt3IZsZ6t7b+zYIwAfErd4Hb4HPlvPakZlajW0t2DBp998KUKtYbjph5Lf2ChDNfCP + LCmPcdrJ7tcHEjteyY+/RjuIWoTH7Yf88p+pkuIH+s3HM3vM3izkW/+Hr8yXRb8cSXs24NaikDZn + sgdiOeMZauc2ZuYWjt4AE6LBRW+S4015/8PXiM0bon30C2fJRpOR//1oGDgnYzkj7q1+nzON6FtP + UKhgwfQNKNH9YgummmABnpz6RdVCiKPF34ywu6CQHaszS2sxdWo45Xhi5soOyl7PlVBd/Bo5VNq6 + q+hab9H30x1p+dkpJQef3P/pgz9+nS6yq8H7e9aI72Ix5a2j21D5UpEqox6b0uwkMkij0x7P8nRL + R7ret/B8njp8XfyggCLfBcvzE3elxOUvb4TvJ8dUPsvc/OEXWNYby6L4UYr2ckIocInBdCm1vXHr + HyAQpRf+0+Ozec1G+I3Cnlnrz9Ebb6t0VB9TUhPPWBVmEy9d7qS9zbGy6Mm+lvIKPki0HOn/cG8s + nrUPSIgM4gEkmmM/flv407u7YKt533bMXmCduzFxwpR7NAjaGpqKBHC5YgrvH5MnKL/8gFyeB87L + oxmD7faZU3nx+5M+6AIcv51KN00xAz40ZgK/TSphKW9njzKZZbAPQkpw87Iifq0iH86f3v/5MTDv + 1syGQ9bJlLdsZ44sqDQkuEfIvJ8+dbOPC5WLLBLrc/6Uf/WBh4c1LPTDBwihFc1QP75OzGahwzck + Ll2orVqFuapVeH/v9ze+27Sf0sVvCnDx/3/1BWm9/7Tglyesl7xbOpvhHz4xEt6qcjy/8wzuuuKK + UytbefXip8GiX9g+h8ic8fjCUMoox9cwjbxpN8U+4ML1wOLpswNTO8YveFTmB9tm+6kcj5ptQ/H9 + 4Euf7WM0vhu9hhJ9cWJ3euj9+BWM5vOEh4/n8zlTrzGsT8V28bca4Pv7O/nze940FR1T67sApHWF + SSZ/zWiyDr4PC/IwqEKBW87BPqlht9pvmLsaurTXzmkNStc8MD+Dp270bKzB85l3eFrWT/+91SFY + 8l66vhgH3tZpO8NmEDp2t/2+5OYHaHDJf5nraS3vn3IkAGGcrnhztFPOcXDP4GwJE22NMPZGLkDj + L4+zPnxXjiyiMeSiO2OFGLyczK2AoR6GKh2DHe5GdyVC0MWNQ5LunHN6+F59NH6/KtPXg9wN6BrY + 8K7bbwrfrR/NOnxZEHSSRKU3aMDkWnsI189zhNcsukQUIygBwYYzXS/vs9+c9zNcbdd3tvj96OcX + gcqlI+3DqOBLr5X+Vx9jhiWzbv6MJIEu2FQ//Cj56yuvYL3TCdvX0t2kEf/4ao4agzz33QHMYeaE + UGvDmFnT58P5dWIC0LWHSkz2/fzyM+33vllQ+O+0i937C6z9TCbRki/x/X1IwC8v2Wu33ByzNBLQ + 0CkR0+6AlvSuvxKUX32DPYZ6KGd7DjP1VoQlhbLYl2zlvmXY7scd0Tr7HfFyz2OYKiXFkmLvzZn7 + Tf3DS2Jww+42kjA5YIb2xKxemcxJ38QufCI+4u5IioiXl90K/PSwSaLRnDWQUfDsiPeXB9A9CQuI + h1Kn6m2r800Wogos9QSm6RKOpObjPGDrFj1zRbcwZ8V/B6BWV1ti2s4tbfU9HEElCxbOF//HI5tI + sBzgcsfP0SylclVQWIRGQ3YFf5nTt7clZIzGmZAIX8yhEwqMRttUiSleFTCp3kf7wy8kCutyPIqc + qheX5r/8tByr11TDtf+QiWuXLaeX3StGh+qpM6+qO7DkfSu45On08mnmX14Xw/PZtNmhXpcdHw3d + QZVSf8nhoZ1NoefdCia0G4h10BKPk7h0VGGFZ+b4vVZKxM9UeIV5T3/5+HgDtwcUrtqBGO1LMefh + VAjgUSgjbhzyKnm4+YxwqUexh8Olbtyc9RElF1oSd+cL3qALRwHdT08Xy+kp8v78F5FvBka+f/b6 + 72sVgKV+SE6prJYzGsUALvzA9GlgETsdhxdc9DjxJGx0i3/8h//wbjiYc94hA8bvYUuuz+TVUU0O + r2jBBzqKTPZGaZeP6JSClmn2bmvyXUgoPH9Em7lPDZXj6bLPoDKYa+aAzjWlIXwVv3ow8arVI5rM + wH5BfpAnhs3m2Y2Nf2zQ4seIdg+3qWjFGQZ+cnqzg9hI0TztSPGr3zLDnG2P7tbMgqv4Ef34wpus + OPOhGuAj0Zd8+Pt7/64SrHGVCIk5KhRa8kE66uxXXxs9Tb/+6gV4veRHwk9vLnkSretrw//yTqqf + YubzVuz4LtxR6E11yoznxo468Pn6YPHH5PSqp3Lct8uOgsBa6qvLVtTtgxbw52esEw2AdBtYAD7G + VyTbVFa7IUpdA4ZiamFwlLtyqQdBuEPxlQWjePWGpd75/3NHAfx/7yjYGQAzF339ctQZDtTXJCnE + WcOmm+io20Cp3oB4g7MzxwHfR6iCeGSHZK9zyZPkBzzo0ovhXS954xoGBrqs54jtjjfMhWOx6uEk + FhbR1XpjzmXRv1Bwrgj9Xo4xEEbwtGC2bzKib/w47c1ATtBDbS2m++aq67qYB1BcX30SlhlNp71t + 9aif8yNz9ushbVldaqp00C7sQUXfk07C3QebUFKpnO8vKctGM4DFONosaCVaDpPYYrC605by0+0L + xvXoOOiYXSdKjyzz5qjoMYzu9y/Z09YsZ+G2fSn6UTiy04qU5XiooQov1NsQY/vNoykzbAGZT9Fl + 2nc/pOOJTjVMZ+lCcJzlQHgXrQGfs5QxXxdGj8UNrkA2SQLR3qrPJewlNlglTU8OyneMvs9LrqF9 + HSy3XoQ7b5Iav4d0yh7s8JRMb/4K2kvtL8gitigi0MdRHiKlqX26mgkFcyC2EtwWryfJyOYEpF75 + tnBVFSPZpkcCuLGLJWSLR5NZceOVY461BwKhtGa7z6x4g8Z2yy21Q0LucHeMpvv3Y8BOaUZ2st82 + 5xlMnaWvUUN2ySsEU7XcctikGmJWCFxzip7zFdbbY8dsoFmp1EqKBcU8l9leHvbdHBUVRklBHlTp + /Ec66frk/32eDM6Wi+UugfDmNALzWP3h8zI+sC4UiWxD1wMb9s0E+EzjAwszw+PiLCcz7B9bjSxb + 0zwp4bsatsZhZkZqf/n42tEeVvLXJMfUbsppEgsfWYLvsZMJzVLKhmOG3pvvC3fA88FGu1sGOj4f + jK7Pqsn5uQgydLG7Lzl0qWNSJ7xLoFZbzLzwdCrHQUQreMJjRy5A25Ybt80lOIB0YruYTyUd7x6F + h1P1IWQmo7c8jwUZfynkdMCtN7lwiAHsbw27W3ehG9s8vsLYrmMW3xnreCKpGngguii21uw2xfmk + wmX9sCtN7+X8uptXqN77N8n6S8PFMFAgIEFZ4ra/2BE/qzKGRa0aZL/LDx1PQjdRhdT0ifN9As5v + 5h5CAg2D+MCjJgWhPqNTlkjsigExN348NzBZiw3bSWIZsav/DeAWWR1d4dMt7RszLSCUjj0z3bcV + zWHUjVDpGpeiQ3XtuKqUPiQytMg5W83ljC9fCIn9fRJ/fQ7SzRS9GxgLaoOnzGg7pllVAJPd90jS + m5p0Yhm1y3jeJpJtNI9LDy9M4HRe15TermbJdmctRP5mlTHb9LapcNuYAYr9PiDPuvU8IZ9HC3ah + W9C109xSqUKCAKd1siMmKEA05wEvoOjqX2JV1xCMjpBcYVEKFom+cRR9l8YPiI93jR3sMfLE0zsq + YGZXD3a9sbjjzlvr0bC/nDHI43s6drCngKxtjcVx03XDatQTlNfrEMsxD7nYixsMdQ2siLXWdD6e + 32EA+8vaomskiOboDl2jXjvji6W9+fWms1aq6PL2d8w6rJ2UuhcJomvu1GQnFBWY3Yu0+uEDRtTq + opkkSQHIBESqXI4C5++d84DbPimI9XgRTxK19wx5ZjCyWyVOJDqtcEYTPO9peAa7cgLJx0eBTz12 + uF5X5lQEaojECAwYcW0wp279KMBDbSyyz2OVD8MQFMhI6ES0g6l30gBDH5lg+2Kn0UmjMU+CMxTm + MqXrVNqWovjOHEhvOsRzzgZz+pC+hXKsnEjEdKHjhxLY4Pf712W8Nr/5+cPDxApxKfBeb+B7f/LJ + 1vDWfD4tdxJ8ntZjOVNfg1mYhhAWarPH611/iZb5EKIsxl+2S49hOk+970BPQJTtPzPopvfU9bAM + 3Suz4GBGvA1dCpzotSPPvsBAGgMuoRzxC8Fco9EskhqjVfUaySNkWiTstDaGJ65s2RkOW8Ck897f + dFf4IYftmpdN58AGkiAvl/nb8X7SZRXOn+rJrrQ1UmmDXg6kpRDh+bU6dPzz1mNYreY1hqW+5uN6 + 1BwEDFein4YCb3mfGbyVRGe7rdmmzUquBQT1fcQuXu55AqgOIVA/1Zc8iueu2+CZv+BTyzfMP7Fb + SWW/adAPD5PI5oCq5v78m+/MvRzPnnS6WTLMVXaj62RvpRtZcC2ITRmwH9+MAXg84M0oZGKQfLkz + QGlsaLbphXhwh/lMZjWAwQbm7Hk6GJ2wf3QYoLQlFIoPn0syZC/w4S9G0sONgwnlGoSA2BviFvEr + mr7D+wWrTpPZXU0I/37ze6Mal5YT+7BuAb99lAr4t8Yhj5du87l6n2zkZL1OQik9m3O2VQI4Lcy/ + 7em1E8moZaiKMSc26XtPPOJ9Ah1d6di506/pRk6LGD2ZHTN8OuhRczUjjFxNkNm9v7gelz+qD8+K + /qCb3dkBwvOFHz89w7wGWXyS7UMD1WCsWHzZeFzsamlEXk4+/8wXsjZcNEw5pWHdUd7363v4Gy8M + 3rRKO+t96tEbTRmVakkyZ5PyEbVTdmMLf3Opa3Mffet1hj/arulGU/gkqOy1FIMi1jspooKNov1Q + YLlBuTcW+lghdpIpuwpF7c2D07ooWm0AnZf3Jfah46t7PGc48frJnDWxfUE47kdCvo5vTjB0CiB2 + 7ZYQGJ7Nzdhvryg1PyPZHm9+OoPHrVUn8/BkbtLJgOdPhOHZZS45LHg/5ueP8+M3pp9dG0xpdVBB + qJg3CuAOA050qMF+/QypcLH9tHiUqgudc7Ulhp1oQFrwDbantYJFoJne5v1sfMi2/ptlX2XHh5Pr + +eo9SzVmPl4HPkFj9wDpLFwot+TSZLLg2jCvUciuXt5H/a0FV4BMsKFdgyrAV9iS1E/o3NmTT8wc + U1glSIwFDZd7851umo8iwzB+71l4XG9Bd0tzCpUiC/BmaDP+bS+rGNS1vGNJXmpgTGxzBOc0jfA6 + mz/lkBm2BO+t7jF3ihxT8kNHgjWiPi6O6zfnbVtkqm2qEwZWSDj3nXaGRndVSQQf90iEBnnAVTB+ + 8W3tuemEhqsE95Z6w1PGXW/cvW8QHr8xZxf7vY+kH/7vbkOE1fppdBPunwGMo9eZTgezAAN2mgDq + o/umcu/nnXDg/gy/q2dL21XSm72NFKqKtDfJCY9fjwdWVYFCYRIzU9vzGKr7Gtr3UWDuRttH0hp7 + NXiktzvZD+06mvrH1YaWgD12yMuqnGar6OEwZzq7LPpn/Nq+DYvw0DB9/753E3xcr3BI9CfZkf6Q + bkb5HsA4yK50aR1oLvh3Rnb+gsT5Uq0TZzkckW1DhjfjxejEIpgDlRnOh1hGUqZjdhFGZBzsmhjO + rkh5rNVXVbJxiNOFj7vmM6kobE4BO2yx4PVW1MAfH7Ht1CYR3ynrEKxfd0LfUe9y/kG5DYnPUzpm + xj4a948SK6ynnJDbtQITW+sv6HicEi3jDp9Sjs/gt57Wp0PR8SpdG9CN6zWWOwRL/pmOPkre6E3I + oQLeXLr+DC+95jP3o6N0tOSghgvf0/x5bMEIxrsD6PR4YOA3ccQ312lECowjepflgM+HWLlC4f25 + MUu1NJPCIbThc1fbeG27ctSOYplB5oQllpfx41M0NGCZP8Q44jrq417X0MJvVAmaIxdMgV3hfdwc + yT7ahKYA4Nv4449rVSZe99N78Se4s6s8OJF4ffQz/LzlCsuRPS1nmicNmnt9S378O2brGCKxa7a0 + HhXoMdfUW3jb1hIjUhKmPenPGtCv7ZnC+zU35yR9ZZCe1SOzbtWQspCiGITvY8jSPEZpU4H9DL4t + KxlO1KKcT75ooECQS2Kih14ufBfAZT2SeNFHEynlWo7e2CDGSAXw4xP0ktsVs2op6gY0XAUoXNaY + AuONUy5G1xYaB6sm7vu5ikboDS+I2PVMTumRcP6Zbj7kxB6Y3pfHVBzyZYdA29zJaUXMThysTQ+V + pvLJY4gO5uZRlyukPHqHpI2/TfsmBQE8Om2L58WvzH6oSbAOLIM5m8le8GZSkRJsBLY1veqnwGeY + 9NucXZLXDMZDMhhQfBd75q4S35yr29aH73p9ofm12XHJ/cgGfGmPD0W6OfHZ/ciaeobTmyz8V/ZL + W1bkkU3M9g29d2zsDwlw30eHhGe3MeeboGHYPGxAsJpM5mQESgumpjJYcqxiPnDKH3D1qSkxuX8G + PFXaCho2bIn+yQaPS+h7hWl6dsjhEVueoJ58AaLj6BN7b+496TtezuBNng5e6SY3x3xFzvCoapA9 + kXkqJ/nxvsLsEnxY1G5DMB/grYZ9N3iUtu132ZF1eKDC+tyZ/yqf5hRaxXLnSbJl+yU/mAC3V0Cx + 0UiOjvwqhXGthcrCf39+YZRJ2aP7sN2zvVB80lGgqg/Tx9UmP//XotFo4TIseFX5e2/+6R8t5RWx + EHksfKgVKIheR3a9rk4eo3UfAj/8KATrywnzLZpXqKd8v/Bx3nHBCny4LssDs2ggRVM7ni2wvC+y + +N9y0ztlos5TVhJzRXJzMAKlgbhL98y7qWr501MIGI5EHOXxARzNQqbm0cEj2xUxy0koyhq5XWpR + NbIj8LKnrwSPYUFpz6cu4s74XC1nnjiz663UDYs/B5c0NomO83U0oXiyIJJwtlQU5JT//K4gHQvy + BOfYG61e61EpVQ07pUcGBlvralgXYIkqP3onehexR80Hrpk/OE3aqw5cQeWw7X/+yOSiKkvw+VYT + 4mDAzHnRlype7wOcbddhuoznC2yeM2DL+uSUI9H58zOHRW+xKt0YEKUNYRHT446H4IDhLpuOJIHy + YA7oSGQ470OPhu7bNMXPxpLUYN8c2CPOcj7S+9mH1077kr1Q7KIpv7suXPQCsZzwY/7Nr58/cwHs + zH6jleFy52Tw4ztzlN6hjJ7r+c6M59H3JCe8C3DJB9gOg4G/2SWPUQlpRfbabtmR77oqfMbHFTMy + 7qXiq86zvzzk8dxWi/7ZqjDerGqmrzSFj+9z4qizQSDBz63FJ/Op2rA9IYXgZG+arHw/Yogfzx2F + SZdGY3qyMGoqoBN8/VSeGO3CBv3w+ZSrkScmcGihejAyQu7lzpSat2OhtUDOzJkcM/3j30VfMEeT + vyl1lJXx0y/ktpl2HU0+cgyRC212p5aX8uc3LqDb3SwWL354hrnh/83Hk7WzvXcYKCs17XarRd/i + dIyapwT1nakSvL9dSqEC+gxPqq5TGB2vHb8K+wd0hqvLtFbCS5/2Qw1ih62Jv13Pf/MBoZ3q4G92 + Vbw50OcR5epwY9u46fmST2XI9Lfm33jMyUc+Qx9vJqraI4nG4r7s8I2kO4VPKeebOrz0qIR9RVIX + hHyuU3H86VtmBvfGnK4bq0B+Gx+XfKqKpkWvwwV/iWuFuGOjfArQfRSPZDcpedfK17FAP71i7AWz + 27Rt+4BS2xZUOp9IKkkP2sKpek8/PI+GPtQwym+nHdmJ1vMfvvGHhBDzG/N0Cq3WhlUurQiRdxGf + jO8xgd/X5pefnk1eh8/+l1eSm3n2omnuXVX1wK7FgpGU0XTfjgbarfYlrUN38KaLmWWQVFFLP0se + J4Dx7sLq6IrMq3TTE+p3WKE1sSDDl87w5iY/qzDURIPor4PPN452s6B3fz6JdcAN56sk7uHiv4nD + a8tb8jDt53+Iw9IczB6+tCpWzJjWQrTiY35mLuxz0cHz5rKckJO1ALRMhMTYea90KJ9qD5G7shnh + Guu+ufhVQdSEGrGu/MCZuNNmhA/jijj+HUd9jeYz+mTGi+k584C4rLdfXoIn6wv40OOnBha8IU7y + ir3NLr8U6pI/MOfRSVGfjtkL/vDZ8MOqm17JrQLX3K0xT05dOZXA1BCKJZcdbioyJwIuFB5UY8O2 + kpWn/KrdrnD+1E86cn3k9OElV6infcycj57w8b105VlMHAuXfojzyUcG7AqpY06+v0SN9tpKkJK9 + TGcpPXvc8tEKipEysKWNUjnR/iLD3/N5U/RNO2e8rODG1wT8metVN92Hq69yNkt487qeS2HTeBl8 + WoWzdNlVUu4+jisYfMLrD1+94TVP57/566y9vTn59cpVBz+NGd7fNuXAd1EI/SDDzH0/H9GmoKgB + 4YGbVLOh4jGhk6/wWvkVM1spjIZ8OUE1RKBkmuyI3kDHvQVP22rC4t788GHItfaXtxEn5jNnXxK6 + aLM5upTj0e4G4RxRMA99Stz7/san4f7MwOPpVVSkwaXjw9rSQC0U4cKH365d8AiinewQb4+nck44 + qeB5VgpGgtWdT81jsICoeTXuX7oVbW7kKPz8DDn4zRmM2NgGv7yC7GerTrnQ7P74l7gdepr0eDct + uDHciZAhoul8cPICUuLJbAsfT875DY/w579XC15zMQpauF263jr8so7mr+AU4LoaR7Kjre4J4bRx + 1Be1JzwLysWc3gjUYEnWKXSB4PHLdJXh/BoTph/XVjcpwyCAJY8g2zW0gHQm7xrO4Fwzv9Sf/Ff/ + UK023v3VU8S5d2U4BF5GDvEqTad9mFXw2y5dkBZ9Mcb9XgO/9bDR5MabVuVUwEVfse35hM15VXcx + GGF6ozfj3XGato0Kf/WLbdS3nN5H14BUeGXkvOQ5A/GuAtxsfbbkfQ+TqpJyBalVmOwEpqKbBxhi + xK6rniy/X3aHElgqvH5z8ljwf3juGIT+rXXYNle5yS9ToKrZB47MjQ0RsN0UVAAV2YlFTRnx8fnC + GZQ0+0y8PL5H/ZLnIb14JcwqumPEV0lG4XTYFYwo35kv9YUAROa7JiRkesmFbkzg+3zEZM/rbbSJ + 3G0Pvy/xRde3qwX4Ny0MpN7pm+xCF0fzGpsVItWxxeKtouWsVGQGv/VoldkI2IIfUGdZSpLN1ICF + HzVo0wfCCvp20VRR7iLkmpwsfJGOVZRXf/72GazugK8DdIUPPU+Iht+vbsxn2VYWPczsKx+jfr8x + Vj99yfz6uTV5rWxUyLb4jfWRRuX8TEIBwd36unTNtjqhHfoQNpRbxKg7zIXw9kxgM+c+fVUHr+PD + pZQgvWsHYt/Z7I2eIxjAqk8SXfAhavLVLlaX+UA8KOOIh+dTBV+rWWUafmslz3enDKaifCJR3LiR + WMjhAy55BNHnYC75wp+wVhvMTk7opOxtPkN4clpGxWw1RXNJYfKHX1trV3t1r+Ttb34RU7GScmCX + bwyjPSvwWxJfvHskmQ1/+dw9fCUpv6Xf/q9+kd2vuinhUpHgp5BqsmN+yudWmuwfnrGFPyLmha4F + Q3fY4kknitenjnBG1nXosOKO1KThEzRg4T9meGejFNdLl8Fnej5QTnLbHG+P8gWvZpGSwyXLvB9f + gdJ6HchtvLxNagKz/dULcLF8f8aj3sMjnmsKs09ezoNTuChoz2ey/eUDPYweyKovEjnE2avjt89U + ofMMCsrP7jdidy9qf3zMdmJkRX2hjzWa71rItpfNwZy/WpAhvz0fmbnWtE7a3vwaolhwySFR82je + O9ILLfkmuy71V/7Tg7/5KAqXlynuw7iGjM0ThvVz600KrANo0wzhUa1FPvXb8Yys4bp0sQz3ZdsP + t1lttGxi+E0db1IrXAAimxMG5bwy52471apwFEN26qnc/fwYcu2VgGWg2WAQp0pDtilPdKt/t+b0 + qr8POOp7i/gVss1hycNg4PceeS7rtbPDpwW3Kb1h+caEjp02RIPvVj8Ra2/WZWN+Axc2h2SmvTAM + 0dxtlRqqBy1bNlS5gDf+t4IXut9QERT3pb6xgjAW5IYdc+bx6f1OJViZr/ZXTzPn3/htdzBkkenp + 5Sa4W2ew6CssVNeQj1uqhOoyXswZlAJMFp1fSKk+4Fc/AL/8Bq7VS0jnqN+ZIhmdB0BKmCzzqSiZ + Mz4hnOSTvNR79aiXSddDa3IOJN5c1IjWKRrBA/XGn9/pWSTJYKl/EW+pX/FOKWQINkeCxyEmnSgJ + cQXuaAqxNF70aHO/lK/fesLK4s/HM5oNKF/WEV6dO8R7MxgT6JqqhLl3Lvmon+sX7JR2ZHbyOpbj + +tIkQNsEFvHGYv/LpzP1XJAbVe97BUzL86DtY7wwEhw1LhR3TUZAIEdGqHiIsjbPrv97R8G//uu/ + /vuyQeDfdXN/vJeNAcNjGv7zf7YK/Gfzn75O3u/fvoJ/0z55Pf793/7ZgfDvb9fU3+F/DE31+PTL + 5QVo88/9Bf8emiF5/9+f/Gv5uf/5r/8FAAD//wMA/S8EZHQjAQA= + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99afd945fa8a-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:52 GMT + Server: + - cloudflare + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-allow-origin: + - "*" + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-model: + - text-embedding-3-small + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "145" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "10000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "9993623" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 38ms + x-request-id: + - req_4e89305535ce42cb47774650cea16db5 + status: + code: 200 + message: OK + - request: + body: + '{"input": ["Are counterfactuals actionable? [yes/no]"], "model": "text-embedding-3-small", + "encoding_format": "base64"}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "119" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/embeddings + response: + body: + string: !!binary | + H4sIAAAAAAAAA1RZyRKqPJve91Wc+rZ0lQxKwr9jEpmDoIi9YlAURWRIAvlvvgtPV3f1xipMFELe + PO8z/Ps//vz5pyuaWzn9868//7yf4/TPf67fVfmU//OvP//1H3/+/Pnz79/n/5t5a4tbVT0/9W/6 + b/D5qW7zP//6w//vN/836V9//iFSxZOT1RYJVfbWDfZHM0QXaTuBBXdzrBT97CO7971m8e0ugyIc + I6Kp3NOYeVdY4HjgW2JZ9wcYO94R5XuV5aQ4t1Iz5sksKx6Xmvj9Sq8NG0Zcwh03+cg/23ZBTKeO + lbd0TzF71BETloOXA6ySMpytl8TGnStDSOPNFwUULWBx86+l9PClI9t1ybAs2w0PG66tkfPQ6YB3 + cW0rprG1ibtvE0aVdn4qSn6okYFC5GGy42XFDdQKmQ9da8TJEzqYtk+D+O322TB3Y8ZKkOUzBuf2 + 0kzbW1wrRI9PGNy/ZBjX96O8quxLnMP22MzwcRAhaykmx004FJRc3j4wM0dHkQgcIB1UJ4Op/q6I + rQi7ZlazJoToOmYoJ9y7wZzPOiiw1kV3sTkZ4u6+y6F/j3ri0OHWSJd0KRVTmE8otKU+wc5HTmHd + dRzal1Dwvs8tNAHGlwn39UsrlsN8DZX7VkbkwNHBk7yZ6UqTWCMyOvmcSHuacsBdEIefXLJjy6d5 + isAusBhuSt0xpEZ9pHDf1QY5DjbwllGeOaV7BxmKDsvHG9On6iuo4AWUCK3mLW2hqUpGz3dkyD0a + +KaUMXSbq0Aivb4D0TpVrTQ/woBo/t4GQnTLeqXkbg+CWtkyaGBOLXSRKyNXM/2BpYlqKmdTf+ON + d0gatnwTE/72O5ItEwj7T/aEwdusSLx5a4Zw47oXRFecEXuXxImUS/VNSfVPhVD/6Yplb1wh/LRd + TY7yRmO0l2AI5363JXpb6Ak/iZIt01ZG5AxkPRH0tWZ2IazQpbrojNpubELLe1vIdQQleRPWvuD1 + HQ4YksNojF2cLlCnIUNulpeMpcVDhmnB8WQ/nRtvfiRpBw/ClxGXcO9hqZQ2BWJz2oWSDmAzf97c + CdaB55CQmloh7roihPfetNBFv54LkTufS3hsHjw5gWYcluAZiPAS8RqpuNhlFPLuAsUgEcLdJqqY + 8L4+bXiTLhc8V4LhSZLaxzAbzRd+WlbXMFMIc/hqqUbSZNkB+tESW9G8l4TKQzaxiR9rUfm2NCK5 + B51GyNwzr5CjVSF9Vh+AJtvtCV5m+iFJOB4YURINKi/LI8j4EjlhelFDqAvDCzmSeC3EzXK0YK/H + e4S8QzJIxVnIYSLLN+Sjj1Swd24t0DPVA4nDc2wI3vuWw6qjd3SOdyXoZ5qqyhHuPJRtvrYnaXWW + w/X9ocR11YbuszpX9L0sEI0bTW8u2LuHdaHEobyIOqPfffAE0DziUDGIMfBh7+lwXNwXsb/9K5nJ + +YihNj4qLPu32Zu7bDLhrdpsiKrXG8Ac//xS2Kk/ksNFnAHF5amDlnP7YoWihZFimXX44G4Us3eQ + eMv1dHvCuvEDkp83GhNzrsmV0PZqZD8U3VtAsB2V11a+EfN5DcD8ljYqhIQ9MSfYiLFvHUdKrX8A + Vq740Lz3+/dN2eb7I0HGhAyxtT6+cnuHMgluS+oJlZFZcD1fKHY7BKiVbDkoheIUbl+Z4jHe1Wz4 + 7BYcxsqNB3jXFT60digj6vZxB0yrs0zh+9IgmnEtwR2UHa+kXVeTIn9rhrQktQ6U4Mojtf52CS3d + RFWkqhNJNe0gw27+MBX6YAtyH2Bq2JRPCzx23REZeZc0IuQ/ORjTjqAzLB0mDktkw27qHJSqju+J + NchU6KvlE9lVfQeM9JkLxYNkkrtl7A1BL8+tEqdxg/wBzgZ747pVlJb5xL7vX2Ay3dMWpsLJJ7HY + SsOUcs5L6fVoT+wrqwY8iXoNx+vGIPZRH9m8aWqoBAXZI+Opb73Fb4UYbuOnQbxRejLW7yQM89Mm + J2qXugPfFkoH7Sx7EnXFP+pDA8MVP0NxBs9ksQ5+Civh5iHN0mnDHnGv/84z/oy37zCq3YGD+ecg + Ee9o0gLPseYqfeB8SIzci0fnWKEycDaUuMVbahaa0pdic2yHXGNSjWnKHRueAh2QwCv3QDJ2Aq9s + DxyPzNLYem0p+M9djQ+E6BYIhnmzO+kw2VKNaJXQeEt23UM4eraMTu9XBxYtd83feUMrfgAeDDWn + JDrmwt0e3QymyosMSyvRiB1fr57YHF6x8njvr+hqPfYJrj4sVmY9M9f1RoPw4s0ekqNZoWJ3PjP+ + RmgJWQc8YsNbUPBG/zWhj01GzvVXKoi9bE1FvbtlyFvmyRgD8/2C792b4R7Ix0RCapdBulNH4meJ + kUjUUDrY7vQQ3TWzGCjVFlc5qLcEc9JsJF876nIopt8A2cHZbUh32ZYK3Lky2X+7ms1WeIgBCRyR + aO5R9qZTpJ8UFlxTUrQPHyzVh0XK/u4Of/GbSgf1pQwhviNLdXxD7D5nH5qW+kbXkGIwRe1SKkVa + x0jl47SQ8sbqYVFsDHI2IgwmPQErHxJbgmjAFV0x3VN4ZvOJ3AN6KIRmeIow6JuQqLZGwTJsvhl8 + xBCiQNxayRL4MVVQmpnovDuGbJljzd6O+9HBW2vDD/NE+BDalZQgI4qV//m9sudb4pro0IznXOrB + VnhbJD3urGR+hp/wt59YvmyFZAw3nAnHjvlEfdymBAcG5ZREfT1QUjvmIGbRaMPjVbTxcg5IQj9D + rSrvfu7X+UHClk1ow8ipDn/7J9NE7wbjD3JQuPE+gHJP3YTaLnCI8Q6YwdTp0UErtc/4ewRfNmXx + pgeG5Jd4qcoM0JPDq4pXkQWZ22xm9K10I7xzFxWZ6eabzNKZjr/+FMKofRjsNjNOsRj9kBTWDzAH + yrMFWUcjchP6iyF8vUKUMxxoKIgCETxxt4ug/FoiYotKmTBeIBGEfTqTw+e+YfjOoxu8eKQhh0ml + YBG/WQS4RaPE5Yk4jOod25BftJi4epExNsUehKHaxsibwbFg149owgwjDTk4MhpJ7xYfPq/ce2VB + HaCBdIrhQRhYKFbC4uHWIqGM6gMlqN8/DGq+515pzLghXpPObHlN2qhYWbwgxxnrZB52iaWs6wnx + 1zeN5cj5GewXnSeepjjNwlS7BBGXJsS5cw82qzq14fuAP6G0ro++Mi8CJTyZJDdnzD7lMYDQcF4u + yVd84bH1eP3Wh6rez4flWd1TmFe5F3azaQMqh1P64z8hd3c8QE4yECHo2A6DSi0NwXp+LWj1V0S0 + 74zZbPJfCp9tdyRW7XXe/FCGGsJQhMTDNE2I4nY2fHmxjw5mYBoLFfblzkaai+741A9M6G4qdOSt + StLxwBi7yxWFMmuv5K4Ux6Sd8x7CgX5GFFSFAybnMapg7T8YLnxY0A34PKELZ584/qEoZrff83CU + wTfcHKvvsP7fAouuI8h6qIFHto5vwi+9Q6KPzWYYNRLVMBJ2T6Q9d8eG7jcXCx4L7k0OXNgwliyx + CgFVeXRf+Qxd8U7RPuZIvBPyEp4NGgTCTgfE4E87xlrJzSCA+I72sPyC0d0eU2XDXh1+p+YF0PX9 + K2pTXAh6WJeh7xw7hfN1ExP7RJCxmG0aw44sbbi7cxoQBGW8QZj5J+J5zaZgQpfqsJEuLbIOIvZG + d3tNZaXbfZH77hD4rP0crPWIvLiADculroTW+BgxVT9jIW69SwviSpiJW12tRJTkYgvUlAwYBBds + zHZU54pjO0dk762HN18/x1IRVKEIuWEvDd3JgSrE7OaFyzZvE5YJKAXe2ADk3b9ooA16lJCT7IAc + BS9JiGVSEcrCKCJnVt7g9/xgD6mCVn7CiKHfMVzvh0dyGL0lfY68LKkbDz98OWBUeAwmTAvIk2rz + lQw23isfpvfERtbjeUpwZuQy+CiGFPKfqizGgfkLjHfnllw278igGxZgeDWjPbrcMPKwfY+ecCJU + IFqqvBmp7yxWbo1ZEXflb0tLeVNZ9Roplmrb0C/JM7BLm5EY6heB+ac/VrwNRdUZjfEWBhkcMVJC + pgj9MHsHrVYkSQ+R30yH5sfvoOxlW3I4B7fkBYYO7n54H+FRTf7yt/V5kYXpp5iN3TEEB3mJkDGV + ZUMj1NXQD+yAnNvqDWilaCPUbS0OxcsdenP+fKrwVikbZJ69nOEsGl3gSqcCaXV085ah4k5gXQ9W + kNN5H4FuRKBV5BTK6nOfSFsWcZDTjxqyAdwDsd1VMkyP1xaVn/zL5vG2fYGHdYyw4N+eHj4TzgKo + 7zhkwv3JIH3i6LJzEhUSFlwEhNspGcF6PpBJPiqbR/e2wP0xcfCgFc9ifny2pfzyIp/E6asYGAu3 + LzC90QfPfZsA5oj9TbmHryO65qPcsNm/3ICi1zEJ8XJloqqEGRSp/kYqo5jNUbvc4PQACXLRd2D0 + +vBcOJcHQkLZ55Kxl3gfandpJvsdS71lK7xKZejDE1E3Yz4w+9qKMA9bnVRRkDJeKkEP8+1WRcU1 + uQL80SSqyLgCyAt2QkIUMYygSdOGmNKpMWjYeyr0aXBEyAlVj99vLqbc0ndFtDriPNbcRB9SVfBR + mMp6MaJyR6Fb8C3K+/GWrHhyU/LH11rrcQ+k5iaGv/NNDgN8evPn8CiVQ+5x4WLQeJjjp6eDpahd + YrYW9RbzGqjAe/RXFPr6vhDqLLR2lecGxHpNKBG/ZKPDKDtGRJ3ynTc1Qy/CVU8gV8pcg+6zLgPW + o3dD6SLO7HvKXjYoSZch/5AFbOF80AFwtwJkl9x2eHWw6ZUqiPu/49Jm/OTw68UFMWTbThZ5l4pw + K5EGrfqfCeEWuOBM6iuxBy0aGHi73V889TAVk2loal1pZlkg6nX2DPbrFwrWLeIvPE4mhqYQbHIU + EdQEDWPSTXJhR2hL9OWCPYKDsoa3edsSe3OoGRO3dg5vhNrhRvA/jN5TEsLGjBqkHbmyWM7Hh68I + 4xkR86E/GhaB7RbOQdIiR7BVwKzhUyuNdG6RMTscWPh248LLJ8hC4Ksam3m94KCRHEViyD1p8Exv + OtwlKMPJoD2GZasrC/jVT+7fjsbiH11f/vlrZe6nbJZ7jEFxFVtkak8fjNb3pUNZwCI6fBM8zPv9 + u4Rcgj4h3T7ubHaurg9sJl/Cnx+1aMcugsojLYj5fnWMkWXHydf+WxInYweDP9YvF97D9ojC6DU2 + bDN+MuV5fPAkiMU+YQd7CGH53rvEA6VuSEBWVn0TcGTVHx4e5RnClnQiUtWPX0hbx7SULuJumKcD + GsStMJZQcCIfpXyqJLPYaBBeK4WiwCBNM9OracKfn3DcprigyrWLf/iNTpVkNaJiHnkFvKs32X/4 + xevE3tzCR/V00eVzv4Pu7lQQHu5xgSx1SAoxn3MbLj7WiP46pcZyyBAHr/P2Qo7Rh3qC+uZLIC6e + /rf/fOWNWv/4HbK9RwEWo4IlFKU3I5VmUePXr+BiFVP4wyPhxzc/QvlFp3YdR2qXK5vMYiQU+tVN + Q28fFoEth9z1rBTMwWqvrM8XPlY/lLaW+4L1VRSQc+58gzh+9YK1/gbho8v1gT/v/Rf84Vnc9wMb + f/xC81oJ7fXiy1gJchG87g+LBKP6TSauizJlR3YGlsHeL9izMEQoZOZ53Y+xoNxpy4PNnktRGFzn + hux8roR1zwosAcNmfN8/oeL2HSF6b7+axWDkCbQXh5AaRXIyLmTO4fswflAFIA9mxTtROO3cAHkb + x0wWzme9Ik7bw4qXooF5EHHK48H2SNNqpxAuu9yF7nV00Gnl61SQbjY8SmVOEv50Zaz9zi7YZCYj + xlTCYa3fEOq2ERPrjcRmvh/cEp4g65Gvi1bDK+3uCXN6RyhIBlIs7AyeYC+dfJJP0jXhf/h6XM5K + yDpZSJarx2PFinmNxBaYhvHEqzJ85dYZi65vegLbyRYwnocNcsLw04wrnoOPPsnk4HYcG3Eii0Da + uQPSnhc3mcWtosplzE3EyUd5mO+72YZcf3uQcCvEhnjbijJcx0MOLztGrrGhw5Wv4+8naxMW+/wL + BFk2o1x97otZ3qhP5XMQL0Sbb0djzvgnhl/L47A8Sde/+6n8+K4qWy9AvnUeQZtsKPJfCBmLvLvx + UG8eBxRZiu6t/a0EJ2f8kLLx1ETE3RyBXB3rn143lm9ucPBabeiqb+5g5gW3h476MtB+ePCJqOWu + Bc9juEeavNEA+/lX4OgPyFr9DCpcnRiaMT5isHFeBSW7KFZ+eoCDT76gGnxQyKdNhbRb3xlzJ2cu + UOu9gexTsBL2UuPB+dFJmHCRWBDNE1vods8Yv8r3PfleCYjBtbFCpA6Pd7Fk14D764+iTAoNERe1 + qqz9i1yb4eWxCFAZTlWXIa24B95Y1papLMXTJVXdvgDNcJn98Boz0tiAbUaSQ1TvKQr0CibYO+96 + eb0/Cno/b5b+OS1g9f+x2HzVQai8WYbn08YLl8374U1FJb9g6XzO4W7Yd80SF6oJToeXi/LVXx// + 8q+u18lhC8th4ST7Kf/qrxrVb0HE/RXDRzhqSPXu34TZV8zD7/1xxqws9USkKW3heL/ySG+LZ7Kk + 5+L1O3/kcg64ZFTjciuDg5TgRSrgwD4yXMDwYA0xuiw0xMumLpUKBxHST0NnrPgfQwo8B6HjJgKt + uDFO0IBD8JcPEfd7DIHMXlfkrn4un4TNDabmJBPzlAoAr/wHsnyvY3A0aTLE360Pr7N8wbNWf5PR + Z18ZvrbbG3LLw8agD6kK4ZrvYH7lG/NOkjgoJYcOf81JBGSHv09ocKxHh3JfNMPpbVPYnvfdr794 + NLacDg6mcySn53Vi35+fsetvMVn1dDIfBleGb3zgws89Ele9WNvK2m+RQz6EzbdTgv/iI6pfNOmc + osIwtDSDaK+pTqa2nyg4HVoXGbXzGkYroVCphNIj9zXfoHMsUNBNvUNcJxG9ZS/aHfzLZ5ZCMHAj + xxCeuifFkrQN2I+PQhpijDT69b3Zd4EsRwJ4ksP8zIypXx08/YQ5rCQDKvgGoBew2PLB1DzhZpbO + Wwx+flF6OFlAVPmxhSYX7om1Xjc4KJ9w9d/++gvUG1RL8dgiYnDQMmPZ3ef85y8S8zYOA3sdR05e + 9X64eVhSs0jzYinv9EmxsIAMUOs76nC9DrcYL2D1p1W4GS8TlolTN2SZoQwi7pSQVR+yNS8p5X5R + eWIdTa2Qjma+gK86Bcj8Hj1DktJzCp/3axruwvAwSC/HcOGryr/Iux+6ZLbbpAS/+4+PNT/86bNL + zN1CZd9dErYktar8+Lij7OaGnaa+BSqkmIRTMTXzmgfBjXc9kP11lo1ZpEkIwPv+DqXk5DZ0qK7j + Xz4XiyZNxpMDdbnMrjdkbLXS+OEv2MS9Hsrpw/Fw9QExCJvhvOZfnicc2AtD1D0faPWvCv4qir7S + FMKMDPVL2PQ5fEvQ6e4ZGWyJCzpUxxFOsHyQS9HwKx4srVxmxS0UZv1djFoujvBqxvtQSNuNR+MC + YiA4sR8uuzke+IXMGVjzMBLZGmXTD+/XPAEdghp6LLjLKVz9U3LUpfOAK3q2oN3XG6LOB5CQLVio + ElkuIzqEU0NKEPPKZg/TtV5T42+e85HeFvK7WzawgEM+5KflEsqr3sGf9Nru6HhWV785KtjprVI4 + LzZDdiAd2ZJuLzJ8ZNcJy9GUN/Tx6F15w9puzU+jYXj7Qg1ZUKTksOrl2XhJOTzigx5KznEZWHYv + I7hViYzph28Mxs2pCC9vCxDLa6JC4qeghKueC+nKF6fMPYswTqMmFFc/Z2kptMDqx4UyOpgN9ZxK + l8WO9X/zk/l1u7aAqdPw4/eAKInDwbrrOTz4T8eQfnhwv9cSMvbexVhswWiVjmtJKDVLNeBbkC1w + 1Z/IHLkpmc9evoUqXDDZn9f8WKaKDs7eh4Vba3MaaBVDEX5HpOFJocYgcHTxgeG0LvK0Qk8E4yVl + sGQgCZs17+LPudQp2T3fEjRw/DAkS6wrPz61d5LPmj9NnXx6+yWq1FcH5ueWN+FAvvs13xyb0T9y + InSPSUYCAHm2mNEQAvqYFxSkjd/8zTfcev8g/w0AAP//Sg6y9Esy1nfyCjRPzChyDwD1p6vKjUKD + DCxB4Rng4hoe7GJkXFHpag4eL3QNyHE1NzYOTwm3TDIMTQrw0M8p96kq8o/KtojQTg4uC/dxKoCO + l5hWumSUQdp34PJPCbwqoJZLQSEWvMIgNz8lNQe0MKAktaJEF75UQNdYtzg3MScHsgyhtDgxPVXJ + CroCQamgKD+3oCS+JD87Na9YyUrB0BCy1kCpJL8kMQdZnAtkVS0XAAAA//8DAODA7KPgIAAA + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99b82fe6fa72-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:53 GMT + Server: + - cloudflare + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-allow-origin: + - "*" + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-model: + - text-embedding-3-small + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "32" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "10000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "9999989" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 0s + x-request-id: + - req_7de2b85bdc15a7d05b7fedcfb8159af4 + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 14-15: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew + D. White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\nlanations + are adversarial training and matched\nmolecular pairs. Adversarial perturbations + are used during training to deceive the model\nto expose the vulnerabilities + of a model109,110 whereas counterfactuals are applied post-hoc.\nTherefore, + the main difference between adversarial and counterfactual examples are in the\napplication, + although both are derived from the same optimization problem.100 Grabocka\net + al. 111 have developed a method named Adversarial Training on EXplanations (ATEX)\nwhich + improves model robustness via exposure to adversarial examples. While there + are\nconceptual disparities, we note that the counterfactual and adversarial + explanations are\nequivalent mathematical objects.\nMatched molecular pairs + (MMPs) are pairs of molecules that differ structurally at only\none site by + a known transformation.112,113 MMPs are widely used in drug discovery and\nmedicinal + chemistry as these facilitate fast and easy understanding of structure-activity + re-\nlationships.114\u2013116 Counterfactuals and MMP examples intersect if + the structural change is\nassociated with a significant change in the properties. + In the case the associated changes in\nthe properties are non-significant, the + two molecules are known as bioisosteres.117,118 The con-\nnection between MMPs + and adversarial training examples has been explored by van Tilborg\net al. 119. + MMPs which belong to the counterfactual category are commonly used in outlier\nand + activity cliff detection.113 This approach is analogous to counterfactual explanations,\nas + the common objective is to uncover learned knowledge pertaining to structure-property\nrelationships.70\nApplications\nModel + interpretation is certainly not new and a common step in ML in chemistry, but + XAI for\nDL models is becoming more important60,66\u201369,73,88,104,105 Here + we illustrate some practical\nexamples drawn from our published work on how + model-agnostic XAI can be utilized to\n14\ninterpret black-box models and connect + the explanations to structure-property relationships.\nThe methods are \u201cMolecular + Model Agnostic Counterfactual Explanations\u201d (MMACE)9\nand \u201cExplaining + molecular properties with natural language\u201d.10 Then we demonstrate how\ncounterfactuals + and descriptor explanations can propose structure-property relationships in\nthe + domain of molecular scent.31\nBlood-brain barrier permeation prediction\nThe + passive diffusion of drugs from the blood stream to the brain is a critical + aspect in drug\ndevelopment and discovery.120 Small molecule blood-brain barrier + (BBB) permeation is a\nclassification problem routinely assessed with DL models.121,122 + To explain why DL models\nwork, we trained two models a random forest (RF) model123 + and a Gated Recurrent Unit\nRecurrent Neural Network (GRU-RNN). Then we explained + the RF model with generated\ncounterfactuals explanations using the MMACE9 and + the GRU-RNN with descriptor expla-\nnations.10 Both the models were trained + on the dataset developed by Martins et al. 124. The\nRF model was implemented + in Scikit-learn125 usi\n\n----\n\nQuestion: Are counterfactuals actionable? + [yes/no]\n\nDo not directly answer the question, instead summarize to give evidence + to help answer the question. Stay detailed; report specific numbers, equations, + or direct quotes (marked with quotation marks). Reply \"Not applicable\" if + the excerpt is irrelevant. At the end of your response, provide an integer score + from 1-10 on a newline indicating relevance to question. Do not explain your + score.\n\nRelevant Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", + "stream": false, "temperature": 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4257" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAAwAAAP//hFTBbhtHDL3rK4g9aw1JtmPFtzTHBmhSuKcmELgz1C7j2eF0yJWlBv73 + YEaK5DZIctGBb9+bx0dSX2YADfvmHho3oLkxhfbN27s/3l2v4+9P9OG3f/9699B9+HP/9H7f37rP + t828MKT7TM6+sa6cjCmQscQj7DKhUVFd3q3ubm6Xd8vrCoziKRRan6y9kXa1WN20i3W7eHUiDsKO + tLmHv2cAAF/qb7EYPe2be1jMv1VGUsWemvvzRwBNllAqDaqyGkZr5hfQSTSK1fXDQEB7RzkZeFY3 + qZKCDQSTEsgWnEzRKG/R2YRBgSOMEshNATOkTJ5daRdqQzqHgfshcD8Yx77ocIbi5cjzFICLXMpk + WHkYPajlydmUqU1ZEmU7QKZQcR04AUYMB2W9grf/c4OZAFMKTB6SqLWDuCpZAPpn4h0GigYmgH5H + WTEzBqA9ljkdu0EbaERjhwGM8qhX8DDQoUpMSr6Qiy9R+pVTnUPCbFzTCYciX6L0MiLHkuYlOnUU + bX72yrFIjxQNS0RA+xSQYwnxGNslap2DTm4AVOiCiG+7XNQ7zJkpQ6I8UnVU+/jJHF+88TQcwBMl + CIT58qrCk+RH0KnvSa3uxQEcxpLHjj0BVkfY1QFrGXtRNjmZ7mjAHUuubXpyrCyxHfGxvJCyOCrr + 9n1oPk99XUfZUT6c2DsKkkpCUPd3b3r1MX6M65ebnWk7KZbDilMIp/rz+VSC9ClLpyf8XN9yZB02 + mVAllrNQk9RU9HkG8Kme5PSfK2tSljHZxuSRYhFcv14d9ZrLn8AFXS7XJ9TEMLwAFsvFj3gbT4Yc + 9MVpN0ePHPuLxOJstHba6EGNxs2WY18ujY+Xvk2b1fZm8ap7vVxeN7Pn2VcAAAD//wMAO9coUvIE + AAA= + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99b9dc83176d-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:54 GMT + Server: + - cloudflare + Set-Cookie: + - __cf_bm=YSN79BcRAveJFGvOqV0R_nweoIuBpBSKTY3paJSWtWU-1727451714-1.0.1.1-6qdk2Jk4a0ZuH2ztuLubayTecZmIRiBA3RAWNl_o1zd7RytTwvzZ0sv4XI4iMgLk7fgDsBQSfmzLY1qs.CD0QQ; + path=/; expires=Fri, 27-Sep-24 16:11:54 GMT; domain=.api.openai.com; HttpOnly; + Secure; SameSite=None + - _cfuvid=nxb3prQChp_2F.4JnUZS09dFvJKTYYCKn_T8JvX.EMM-1727451714959-0.0.1.1-604800000; + path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "1165" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9998" + x-ratelimit-remaining-tokens: + - "29998973" + x-ratelimit-reset-requests: + - 9ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_8c4faba466f21bb9b8a0617e29284dab + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 35-36: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew + D. White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\n, + S.; An, J.; G\u00b4omez-Bombarelli, R. Chemistry-informed macromolecule\ngraph + representation for similarity computation, unsupervised and supervised learn-\ning. + Machine Learning: Science and Technology 2022, 3, 015028.\n(99) Doshi-Velez, + F.; Kortz, M.; Budish, R.; Bavitz, C.; Gershman, S.; O\u2019Brien, D.;\nScott, + K.; Schieber, S.; Waldo, J.; Weinberger, D.; Weller, A.; Wood, A. Account-\nability + of AI Under the Law: The Role of Explanation. SSRN Electronic Journal\n2017,\n(100) + Wachter, S.; Mittelstadt, B.; Russell, C. Counterfactual explanations without + opening\nthe black box: Automated decisions and the GDPR. Harv. JL & Tech. 2017, + 31, 841.\n(101) Jim\u00b4enez-Luna, J.; Grisoni, F.; Schneider, G. Drug discovery + with explainable artificial\nintelligence. Nature Machine Intelligence 2020 + 2:10 2020, 2, 573\u2013584.\n(102) Fu, T.; Gao, W.; Xiao, C.; Yasonik, J.; Coley, + C. W.; Sun, J. Differentiable Scaffold-\ning Tree for Molecule Optimization. + International Conference on Learning Represen-\ntations. 2022.\n(103) Shen, + C.; Krenn, M.; Eppel, S.; Aspuru-Guzik, A. Deep molecular dreaming: inverse\nmachine + learning for de-novo molecular design and interpretability with surjective\nrepresentations. + Machine Learning: Science and Technology 2021, 2, 03LT02.\n(104) Lucic,\nA.;\nter\nHoeve,\nM.;\nTolomei,\nG.;\nRijke,\nM.;\nSilvestri,\nF.\nCF-\nGNNExplainer:\nCounterfactual + Explanations for Graph Neural Networks. arXiv\npreprint arXiv:2102.03322 2021,\n(105) + Numeroso, D.; Bacciu, D. Explaining Deep Graph Networks with Molecular Counter-\nfactuals. + arXiv preprint arXiv:2011.05134 2020,\n35\n(106) Krenn, M.; H\u00a8ase, F.; + Nigam, A.; Friederich, P.; Aspuru-Guzik, A. Self-Referencing\nEmbedded Strings + (SELFIES): A 100% robust molecular string representation. Ma-\nchine Learning: + Science and Technology 2020, 1, 045024.\n(107) Nigam, A.; Pollice, R.; Krenn, + M.; dos Passos Gomes, G.; Aspuru-Guzik, A. Beyond\ngenerative models: superfast + traversal, optimization, novelty, exploration and discov-\nery (STONED) algorithm + for molecules using SELFIES. Chemical science 2021, 12,\n7079\u20137090.\n(108) + Kim, S.; Chen, J.; Cheng, T.; Gindulyte, A.; He, J.; He, S.; Li, Q.; Shoemaker, + B. A.;\nThiessen, P. A.; Yu, B.; Zaslavsky, L.; Zhang, J.; Bolton, E. E. PubChem + in 2021:\nnew data content and improved web interfaces. Nucleic Acids Research + 2020, 49,\nD1388\u2013D1395.\n(109) Tolomei, G.; Silvestri, F.; Haines, A.; + Lalmas, M. Interpretable predictions of tree-\nbased ensembles via actionable + feature tweaking. Proceedings of the 23rd ACM\nSIGKDD international conference + on knowledge discovery and data mining. 2017;\npp 465\u2013474.\n(110) Freiesleben, + T. The intriguing relation between counterfactual explanations and ad-\nversarial + examples. Minds and Machines 2022, 32, 77\u2013109.\n(111) Grabocka, J.; Schilling, + N.; Wistuba, M.; Schmidt-Thieme, L. Learning time-series\nshapelets. Proceedings + of the 20th ACM SIGKDD international conference on Knowl-\nedge discovery and + data mining. 2014; pp 392\u2013401.\n(112) Kenny, P. W.; Sadowski, J. Structure + modification in ch\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nDo + not directly answer the question, instead summarize to give evidence to help + answer the question. Stay detailed; report specific numbers, equations, or direct + quotes (marked with quotation marks). Reply \"Not applicable\" if the excerpt + is irrelevant. At the end of your response, provide an integer score from 1-10 + on a newline indicating relevance to question. Do not explain your score.\n\nRelevant + Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": + false, "temperature": 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4294" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAAwAAAP//fFRdaxsxEHz3r1juocRgG9txa8dv+aAh0KYmLbTQFLPW7d1topNUaS+x + CfnvRTrHuVKal3vY0Y5mZrX31APIOM+WkKkKRdVOD0/P518+Vxer4mz14/eNvTu/2u5OZpvxttjt + zrJB7LCbO1Ly0jVStnaahK1pYeUJhSLrZD6dz95P5pNZAmqbk45tpZPhzA6n4+lsOF4Mxx/2jZVl + RSFbws8eAMBT+kaJJqdttoTx4KVSUwhYUrY8HALIvNWxkmEIHASNZINXUFkjZJLqbxUBbRV5J+Cp + IE9GUYBAD+RRw6P19wE86egCxIKyjRHyBSppUANtnUaD0XAYwGPFqgL0BAg1SWVzKKwH5+0D52xK + 4NjrPAluWLPsgA2cXkEKI4zgqyPFBSvUejeA76gqIQ8kgHoER9PxZN6HnINqQnhLSGSViiDZ3ArY + ArARWycPOSkO6RSaPB27JJO8XqAgrLwVUpEGbqhsdGKEo8uL1U1/AKEpSwoSrUhF7MHZGCSjhhh4 + vFhTiRpQJX17myP41ChWHSfTST/df93U5G2w8A7OUCluEjjuJz/W05s2Y7aXHl0F19REB9ckaWAD + YJOzwo5QdE6z6sTevtRtm/0/gxnBR09MQdOGTNI0jZqwZkMhpZbeRJRRsYMNySOReVNttIv5A/mA + nhOIUcHh2dS4A66d3gGm/DtahXwd4hhbsd5umiCGQkv6MtFhjfdsytGtuTWL7nP3VDQB47aZRut9 + /fmwP9qWzttN2OOHesGGQ7X2hMGauCtBrMsS+twD+JX2tPlr9TLnbe1kLfaeTCScTBYnLWH2+mvo + wMeLPSpWUHeB6fx/feucBFmHzsJnrUg25SvF+KA0Wc3CLgjV64JNGUfN7f4Xbk3vaXYyOabFcdZ7 + 7v0BAAD//wMAqvgEMwgFAAA= + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99b9dc8b252d-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:55 GMT + Server: + - cloudflare + Set-Cookie: + - __cf_bm=oRXIDnhMIwrzVuWU43TNLxwOtps1lCgWkFEvurOX1ec-1727451715-1.0.1.1-mYe5OriqH2nTJCHOfNKVGgnv_Lr814JMGAyRoBJTE1d0ErDbM6dRXFJP1DQGt4Jyj5X.gCKiPYGQka4gk.fuEw; + path=/; expires=Fri, 27-Sep-24 16:11:55 GMT; domain=.api.openai.com; HttpOnly; + Secure; SameSite=None + - _cfuvid=UZKF9DkEQdqSzH0GSGBBSxSSOZ6dijLj3FNJS3PYWj4-1727451715847-0.0.1.1-604800000; + path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "1746" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29998970" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_cca139d54c9ab4e71b0d78555087d24a + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 11-12: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew + D. White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\n) + between f and g as the \u201cfidelity\u201d.\nGraphLIME87 and LIMEtree88 are + modifications to LIME as applicable to graph neural\nnetworks and regression + trees, respectively. LIME has been used in chemistry previously,\nsuch as Whitmore + et al. 89 who used LIME to explain octane number predictions of molecules\nfrom + a random forest classifier. Mehdi and Tiwary 90 used LIME to explain thermodynamic\ncontributions + of features. Gandhi and White 10 use an approach similar to GraphLIME,\nbut + use chemistry specific fragmentation and descriptors to explain molecular property + pre-\ndiction.\nSome examples are highlighted in the Applications section.\nIn + recent work by\nMehdi and Tiwary 90, a thermodynamic-based surrogate model approach + was used to inter-\npret black-box models. The authors define an \u201cinterpretation + free energy\u201d which can be\nachieved by minimizing the surrogate model\u2019s + uncertainty and maximizing simplicity.\nCounterfactual explanations\nCounterfactual + explanations can be found in many fields such as statistics, mathematics and\nphilosophy.91\u201394 + According to Woodward and Hitchcock 92, a counterfactual is an example\nwith + minimum deviation from the initial instance but with a contrasting outcome. + They\ncan be used to answer the question, \u201cwhich smallest change could + alter the outcome of an\ninstance of interest?\u201d While the difference between + the two instances is based on the exis-\ntence of similar worlds in philosophy,95 + a distance metric based on molecular similarity is\nemployed in XAI for chemistry. + For example, in the work by Wellawatte et al. 9 distance\nbetween two molecules + is defined as the Tanimoto distance96 between ECFP4 fingerprints.97\nAdditionally, + Mohapatra et al. 98 introduced a chemistry-informed graph representation for\ncomputing + macromolecular similarity. Contrastive explanations are peripheral to counterfac-\n11\ntual + explanations. Unlike the counterfactual approach, contrastive approach employ + a dual\noptimization method, which works by generating a similar and a dissimilar + (counterfactuals)\nexample. Contrastive explanations can interpret the model + by identifying contribution of\npresence and absence of subsets of features + towards a certain prediction.36,99\nA counterfactual x\u2032 of an instance + x is one with a dissimilar prediction \u02c6f(x) in classi-\nfication tasks. + As shown in equation 5, counterfactual generation can be thought of as a\nconstrained + optimization problem which minimizes the vector distance d(x, x\u2032) between + the\nfeatures.9,100\nminimize\nd(x, x\u2032)\nsuch that\n\u02c6f(x) \u0338= + \u02c6f(x\u2032)\n(5)\nFor regression tasks, equation 6 adapted from equation + 5 can be used. Here, a counter-\nfactual is one with a defined increase or decrease + in the prediction.\nminimize\nd(x, x\u2032)\nCounterfactuals explanations have + become a useful tool for XAI in chemistry, as they\nsuch that\n\f\f\f \u02c6f(x) + \u2212\u02c6f(x\u2032)\n\f\f\f \u2265\u2206\n(6)\nprovide intuitive understanding + of predictions and are able to uncover spurious relationships\nin training data.101 + Counterfactuals create local (instance-level), actionable explan\n\n----\n\nQuestion: + Are counterfactuals actionable? [yes/no]\n\nDo not directly answer the question, + instead summarize to give evidence to help answer the question. Stay detailed; + report specific numbers, equations, or direct quotes (marked with quotation + marks). Reply \"Not applicable\" if the excerpt is irrelevant. At the end of + your response, provide an integer score from 1-10 on a newline indicating relevance + to question. Do not explain your score.\n\nRelevant Information Summary (about + 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, "temperature": + 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4358" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAAwAAAP//dFRNbxs5DL37VxC6NAHswHbSfN2CNNgWWKAosIdd7BSGLHFm2Gioqchx + nE3z3xfSOLZ7yGWA4RMfHx9FvUwADHlzC8a1Vl3Xh9nd/dXXP+t/muj/uP72dfPz20Pnrp/uPz8t + P2+8meaMuP6BTt+yzlzs+oBKkUfYJbSKmXVxtby6+Li4WpwXoIseQ05rep1dxNlyvryYza9n88td + YhvJoZhb+HcCAPBSvlkie9yaW5hP3yIditgGze3+EIBJMeSIsSIkalnN9AC6yIpcVN/HgRVTbZ0O + NgBu+2DZ5gYEbELwKC7RGj1YAesyYNcBgRi0RShMW4VYw993X+DkIefTeOQuKdXkyAb4woohUIPs + 8BTqmMC12JFoej6Dv1p8LrUGQQ8awaNi6ogRKvPUkmtBOhsCioJrLTe57BA82KCYiow4qIsdZhmW + gTh37Mov5e5QdFoZIPbkrBI3OYnScT9sdUhYxABuHaZeRzOIBbS1mmseOTW60yBjygOG9TN0xNTR + fzt62KDTmMDTTkxVnfiT7RS2H06r6hTWqE+IDDWWyjIFGcpdyg7Yt0aJoU/oqQjNFFXVWn2pX0+2 + p1BVjD/hEBmJi7shz70u3UYGtfIoU7DsM8WvY47Zb/m/oKqawvkJg9o3uoRNQpE9VbaJBGRoGhR9 + x58+xQ35PENiHUhpgzCwx5T98NmlWB81J5UpAp1lGNjFDSaQfkgUB4GEYbyTLfUyhc4+7lzujmdo + tRi/H3/ADYaziiu+Ob79CetBbF4+HkLYxV/36xRi06e4lh2+j9fEJO0qoZXIeXVEY28K+joB+F7W + dvhtE02fYtfrSuMjcia8Of848pnDQ3FAF5fnO1Sj2nAEzG+u38tbeVRLQY7W34waiZsDxXwvtHRq + 5FkUu1VN3GDqE42vQd2vlvXF/HJ9s1icm8nr5H8AAAD//wMAxVwQjRYFAAA= + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99b9ca525c1c-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:56 GMT + Server: + - cloudflare + Set-Cookie: + - __cf_bm=IILXIMFwKOCQyKIe4t968EFupyUvRYsQBF.htt_DAKY-1727451716-1.0.1.1-F1yqwTS4CuwVLYANhXnSyGZFgAr0L.AzGv5eOsnzBFQ9NRL2_1YMX04RRJxatmgDNjoIdL1XQYOm7V.JI7BKPA; + path=/; expires=Fri, 27-Sep-24 16:11:56 GMT; domain=.api.openai.com; HttpOnly; + Secure; SameSite=None + - _cfuvid=b8zaetoDNOecUNc4DfrZLUwwHg_5euumayxjJ2pW7tA-1727451716043-0.0.1.1-604800000; + path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "2349" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9998" + x-ratelimit-remaining-tokens: + - "29998965" + x-ratelimit-reset-requests: + - 9ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_db32ef9a454b07da72281b0c98abd634 + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 12-14: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew + D. White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\n + relationships\nin training data.101 Counterfactuals create local (instance-level), + actionable explanations.\nActionability of an explanation suggest which features + can be altered to change the outcome.\nFor example, changing a hydrophobic functional + group in a molecule to a hydrophilic group\nto increase solubility.\nCounterfactual + generation is a demanding task as it requires gradient optimization over\ndiscrete + features that represents a molecule. Recent work by Fu et al. 102 and Shen et + al. 103\npresent two techniques which allow continuous gradient-based optimization. + Although, these\nmethodologies are shown to circumvent the issue of discrete + molecular optimization, counter-\nfactual explanation based model interpretation + still remains unexplored compared to other\n12\npost-hoc methods.\nCF-GNNExplainer104 + is a counterfactual explanation generating method based on GN-\nNExplainer69 + for graph data. This method generate counterfactuals by perturbing the input\ndata + (removing edges in the graph), and keeping account of perturbations which lead + to\nchanges in the output.\nHowever, this method is only applicable to graph-based + models\nand can generate infeasible molecular structures. Another related work + by Numeroso and\nBacciu 105 focus on generating counterfactual explanations + for deep graph networks. Their\nmethod MEG (Molecular counterfactual Explanation + Generator) uses a reinforcement learn-\ning based generator to create molecular + counterfactuals (molecular graphs).\nWhile this\nmethod is able to generate + counterfactuals through a multi-objective reinforcement learner,\nthis is not + a universal approach and requires training the generator for each task.\nWork + by Wellawatte et al. 9 present a model agnostic counterfactual generator MMACE\n(Molecular + Model Agnostic Counterfactual Explanations) which does not require training\nor + computing gradients. This method firstly populates a local chemical space through + ran-\ndom string mutations of SELFIES106 molecular representations using the + STONED algo-\nrithm.107 Next, the labels (predictions) of the molecules in the + local space are generated\nusing the model that needs to be explained. Finally, + the counterfactuals are identified and\nsorted by their similarities \u2013 + Tanimoto distance96 between ECFP4 fingerprints.97 Unlike the\nCF-GNNExplainer104 + and MEG105 methods, the MMACE algorithm ensures that generated\nmolecules are + valid, owing to the surjective property of SELFIES. Additionally, the MMACE\nmethod + can be applied to both regression and classification models. However, like most + XAI\nmethods for molecular prediction, MMACE does not account for the chemical + stability of\npredicted counterfactuals.\nTo circumvent this drawback, Wellawatte + et al. 9 propose an-\nother approach, which identift counterfactuals through + a similarity search on the PubChem\ndatabase.108\n13\nSimilarity to adjacent + fields\nTangential examples to counterfactual explanations are adversarial training + and matched\nmolecular pairs. Adversarial perturbations are used d\n\n----\n\nQuestion: + Are counterfactuals actionable? [yes/no]\n\nDo not directly answer the question, + instead summarize to give evidence to help answer the question. Stay detailed; + report specific numbers, equations, or direct quotes (marked with quotation + marks). Reply \"Not applicable\" if the excerpt is irrelevant. At the end of + your response, provide an integer score from 1-10 on a newline indicating relevance + to question. Do not explain your score.\n\nRelevant Information Summary (about + 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, "temperature": + 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4236" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAA3xTTW/cOAy9z68gdOoCdjCZmWSS3IJsUvSQ9rBb9NAsBrRMW9qVRUGk0wyK/PdC + nq/ksL0YMB/59PhI/pwBGN+aGzDWodohhfr2bv3l8ba7vfz6bbVcLj7h8hv5h68RLxr3l6lKBTf/ + ktVD1ZnlIQVSz3EH20yoVFjP14v16uJ8fb6agIFbCqWsT1qvuF7MF6t6flXPL/eFjr0lMTfwfQYA + 8HP6FomxpRdzA/PqEBlIBHsyN8ckAJM5lIhBES+KUU11Ai1HpTipvuMxKuUOrY4YBDATtCQ2+4Za + QIGU+dm3PvbwZAJbDPDBx8JoqQ70TOGPCtCWjrEJBPSSAkYs/3L2ZOBvR0fYB69b4A4Q7Ltn31aB + F2hp4Ciai3XQbEHGvifRIuKH89ZBR6hjJgGLERoCDEqZWlAG6zD2BOoIeFTLA1Ugo3Wll4Fb320L + DYLbtpmT48Zb6Ma4kxigzzwm8BEQBg5kx0CF9Zjvg7fAcQr6WMYrBMJh3HV3NjVML5ZyUmi92FGE + BJ4xex4FBlLHrUDHGXqKlHHq6r0dUhXqME6u3z3UHz9/vi8G+Ui5gsf7jxVgbOHx8fbuvgJC6+CH + Vwet7zrKFBUwpcxoHcmUGfzgdT8T+JMkeZ0cEjoIqibDDrojKwmoQ/3NoOoGhVqYFhl8SUqZ9DjD + QCJTOpfBlLvA/YRYHWVILFo7tgcFZ0/xKV693dJM3ShYjiSOIezjr8e1D9ynzI3s8WO889GL25TJ + cCwrLsrJTOjrDOCf6bzGdxdjUuYh6Ub5P4qF8Gq93PGZ00Gf0PPFAVVWDCfg+vry/8o2LSn6IG+u + 1Owk+tifGOZHnVOjRraiNGw6H/vir98dbZc2F/OL+WJ5aZvWzF5nvwAAAP//AwCk3gJjvQQAAA== + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99b9dc2b67f9-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:56 GMT + Server: + - cloudflare + Set-Cookie: + - __cf_bm=wK.6ukfHg1BLQhOmEKuQGt_bKH23H_IyHb9et6bcrRk-1727451716-1.0.1.1-gF3JQa3K_8aTqBqVfjUjhCerwSA.mWieSmxg11iaMMZpQ6.vcJjTDvp7uDWw83y0BpYZLj3VW0EOX1H4aY8Ibg; + path=/; expires=Fri, 27-Sep-24 16:11:56 GMT; domain=.api.openai.com; HttpOnly; + Secure; SameSite=None + - _cfuvid=ssBRY_76Fejx4hGYkLB9Fim3aLOPd2XHAFZCyLFemj8-1727451716089-0.0.1.1-604800000; + path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "1940" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29998974" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_32ac99c82d71b5fd14ec43ff24a3ff1d + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 26-28: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew + D. White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\nability + and Transparency. 2018; pp 77\u201391.\n(20) Lapuschkin, S.; W\u00a8aldchen, + S.; Binder, A.; Montavon, G.; Samek, W.; M\u00a8uller, K.-R.\nUnmasking Clever + Hans predictors and assessing what machines really learn. Nature\ncommunications + 2019, 10, 1\u20138.\n(21) DeGrave, A. J.; Janizek, J. D.; Lee, S.-I. AI for + radiographic COVID-19 detection\nselects shortcuts over signal. Nature Machine + Intelligence 2021, 3, 610\u2013619.\n(22) Goodman, B.; Flaxman, S. European + Union regulations on algorithmic decision-\nmaking and a \u201cright to explanation\u201d. + AI Magazine 2017, 38, 50\u201357.\n(23) ACT, A. I. European Commission. On Artificial + Intelligence: A European Approach\nto Excellence and Trust. 2021, COM/2021/206.\n(24) + Blueprint for an AI Bill of Rights, The White House. 2022; https://www.whitehouse.\ngov/ostp/ai-bill-of-rights/.\n(25) + Miller, T. Explanation in artificial intelligence: Insights from the social + sciences. Ar-\ntificial intelligence 2019, 267, 1\u201338.\n26\n(26) Murdoch, + W. J.; Singh, C.; Kumbier, K.; Abbasi-Asl, R.; Yu, B. Definitions, meth-\nods, + and applications in interpretable machine learning. Proceedings of the National\nAcademy + of Sciences of the United States of America 2019, 116, 22071\u201322080.\n(27) + Gunning, D.; Aha, D. DARPA\u2019s Explainable Artificial Intelligence (XAI) + Program.\nAI Magazine 2019, 40, 44\u201358.\n(28) Biran, O.; Cotton, C. Explanation + and justification in machine learning: A survey.\nIJCAI-17 workshop on explainable + AI (XAI). 2017; pp 8\u201313.\n(29) Palacio, S.; Lucieri, A.; Munir, M.; Ahmed, + S.; Hees, J.; Dengel, A. Xai handbook:\nTowards a unified framework for explainable + ai. Proceedings of the IEEE/CVF Inter-\nnational Conference on Computer Vision. + 2021; pp 3766\u20133775.\n(30) Kuhn, D. R.; Kacker, R. N.; Lei, Y.; Simos, D. + E. Combinatorial Methods for Ex-\nplainable AI. 2020 IEEE International Conference + on Software Testing, Verification\nand Validation Workshops (ICSTW) 2020, 167\u2013170.\n(31) + Seshadri, A.; Gandhi, H. A.; Wellawatte, G. P.; White, A. D. Why does that molecule\nsmell? + ChemRxiv 2022,\n(32) Das, A.; Rad, P. Opportunities and challenges in explainable + artificial intelligence\n(xai): A survey. arXiv preprint arXiv:2006.11371 2020,\n(33) + Machlev, R.; Heistrene, L.; Perl, M.; Levy, K. Y.; Belikov, J.; Mannor, S.; + Levron, Y.\nExplainable Artificial Intelligence (XAI) techniques for energy + and power systems:\nReview, challenges and opportunities. Energy and AI 2022, + 9, 100169.\n(34) Koh, P. W.; Liang, P. Understanding black-box predictions via + influence functions.\nInternational Conference on Machine Learning. 2017; pp + 1885\u20131894.\n(35) Ribeiro, M. T.; Singh, S.; Guestrin, C. \u201d Why should + i trust you?\u201d Explaining the\npredictions of any classifier. Proceedings + of the 22nd ACM SIGKDD international\n27\nconference on knowledge discovery + and data mining. San Diego, CA, USA, 2016; pp\n1135\u20131144.\n(36) Dhurandhar, + A.; Chen, P.-Y.; Luss, R.; Tu, C.-C.; Ting, P.; Shanmugam, K.; Das, P.\nExplanations + based on the missing: Towards contrastive explanations with pertinent\nnegatives. + Advances i\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nDo + not directly answer the question, instead summarize to give evidence to help + answer the question. Stay detailed; report specific numbers, equations, or direct + quotes (marked with quotation marks). Reply \"Not applicable\" if the excerpt + is irrelevant. At the end of your response, provide an integer score from 1-10 + on a newline indicating relevance to question. Do not explain your score.\n\nRelevant + Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": + false, "temperature": 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4328" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAA3SQzW6DMBCE7zyF5XOoMCFAuFVVT62I1FbqoamQYxbixtiubZRWUd694ieQHHrx + Yb6d8eyePIQwL3GGMNtTxxot/PuHZLN5Waft8ZC3/Pn4neePhLC39zh4fcKLzqF2X8DcxXXHVKMF + OK7kgJkB6qBLJUmYRCuSkLgHjSpBdLZaOz9SfhiEkR+kfhCPxr3iDCzO0IeHEEKn/u0qyhJ+cIaC + xUVpwFpaA86mIYSwUaJTMLWWW0elw4sZMiUdyL51rhyiWgvO6E7AVm4luZ40ULWWdkVlK8Son6ev + haq1UTs78kmvuOR2XxigVsnuG+uUxj09ewh99iu2N62xNqrRrnDqALILJCQNh0A8X3XG0cicclTc + uOL/XEUJjnJhry6Fh4pc1nNEMPXsF8X21zpoiorLGow2fDhcpQtYQbQmS0iX2Dt7fwAAAP//AwBR + H080QQIAAA== + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99c9cbc767f9-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:56 GMT + Server: + - cloudflare + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "132" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29998967" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_d59432c0fd036ecd248c7ef42ea11f55 + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 4-5: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew D. + White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\ncation, + interpretations can be categorized as global or local\ninterpretations on the + basis of \u201cwhat is being explained?\u201d. For example, counterfactuals + are\nlocal interpretations, as these can explain only a given instance. The + second classification is\nbased on the relation between the model and the interpretation + \u2013 is interpretability post-hoc\n(extrinsic) or intrinsic to the model?.32,33 + An intrinsic XAI method is part of the model\nand is self-explanatory32 These + are also referred to as white-box models to contrast them\nwith non-interpretable + black box models.28 An extrinsic method is one that can be applied\npost-training + to any model.33 Post-hoc methods found in the literature focus on interpreting\nmodels + through 1) training data34 and feature attribution,35 2) surrogate models10 + and, 3)\ncounterfactual9 or contrastive explanations.36\nOften, what is a \u201cgood\u201d + explanation and what are the required components of an ex-\nplanation are debated.32,37,38 + Palacio et al. 29 state that the lack of a standard framework\nhas caused the + inability to evaluate the interpretability of a model. In physical sciences,\nwe + may instead consider if the explanations somehow reflect and expand our understanding\nof + physical phenomena. For example, Oviedo et al. 39 propose that a model explanation\ncan + be evaluated by considering its agreement with physical observations, which + they term\n\u201ccorrectness.\u201d For example, if an explanation suggests + that polarity affects solubility of a\nmolecule, and the experimental evidence + strengthen the hypothesis, then the explanation\nis assumed \u201ccorrect\u201d. + In instances where such mechanistic knowledge is sparse, expert bi-\nases and + subjectivity can be used to measure the correctness.40 Other similar metrics + of\ncorrectness such as \u201cexplanation satisfaction scale\u201d can be found + in the literature.41,42 In a\nrecent study, Humer et al. 43 introduced CIME + an interactive web-based tool that allows the\nusers to inspect model explanations. + The aim of this study is to bridge the gap between\nanalysis of XAI methods. + Based on the above discussion, we identify that an agreed upon\n4\nevaluation + metric is necessary in XAI. We suggest the following attributes can be used + to\nevaluate explanations. However, the relative importance of each attribute + may depend on\nthe application - actionability may not be as important as faithfulness + when evaluating the\ninterpretability of a static physics based model. Therefore, + one can select relative importance\nof each attribute based on the application.\n\u2022 + Actionable. Is it clear how we could change the input features to modify the + output?\n\u2022 Complete. Does the explanation completely account for the prediction? + Did features\nnot included in the explanation really contribute zero effect + to the prediction?44\n\u2022 Correct. Does the explanation agree with hypothesized + or known underlying physical\nmechanism?39\n\u2022 Domain Applicable. Does the + explanation use language and concepts of domain ex-\nperts?\n\u2022 Fidelity/Faithful. + Does the explanation ag\n\n----\n\nQuestion: Are counterfactuals actionable? + [yes/no]\n\nDo not directly answer the question, instead summarize to give evidence + to help answer the question. Stay detailed; report specific numbers, equations, + or direct quotes (marked with quotation marks). Reply \"Not applicable\" if + the excerpt is irrelevant. At the end of your response, provide an integer score + from 1-10 on a newline indicating relevance to question. Do not explain your + score.\n\nRelevant Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", + "stream": false, "temperature": 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4304" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAAwAAAP//dFNdbxMxEHzPr1j5OamS0HyQF1SKEAiJSgjxUYKijb13t+Dznuy9JqHq + f0e+fLVUvNzDjmduZuy97wEYdmYBxlaotm784Op6dvPx9S2/5dp/233Rdzff30yvP9TXn8a3N6af + GbL+RVaPrAsrdeNJWcIetpFQKauOZuPZ5WQ0G006oBZHPtPKRgeXMhgPx5eD4XwwnB6IlbClZBbw + owcAcN99s8XgaGsWMOwfJzWlhCWZxekQgIni88RgSpwUg5r+GbQSlELn+nNFQFtLsVGwqFRK5D+U + gINSbCIp5jAJMEHpZY0eJIIXi74PG9YKrLT5aIFWW/QJ2FFQLphc5nQn/xXrQ00YOJSgFe2Ato1H + DiDB7wAhNWS5YAscsnFLF5BNKm0VHCfbpkQpMwFVI69bpQSFRKA79C1q1u0kw/FvHKxvXZ4vzZXN + Q1x76i8NbCq2FWD6nfqwNO8TsIL1hBEq2cCGcjrvwFYYSur+yaFpFQpCbWO2IVCL42LXgdJq0+qr + pYGvFfs94ViuE0oQRDtvbFn9DpKiEmwq0orisyYxEuDZbqfmqODAeQZSPImzNMB147nrBvWZ2j7J + OkcKiR3FfEEnNnCxv4wmyh07OrRQtuzyFYCEQ87c4vMO0FZMdwQIjhJn6UNl+Vq7Ui6WYRnmjx9h + pKJNmHcgtN4f5g+nV+2lbKKs0wE/zXP+VK0iYZKQX3BSaUyHPvQAfnbb0z5ZCNNEqRtdqfymkAXn + s+lez5z39YyORqMDqqLoz8DL+ex/tJUjRfbp0RKavUUO5VlhePLZBTVpl5TqVcGhzAvC+50smtVk + OBmOX0zt2pneQ+8vAAAA//8DAMNSLCScBAAA + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99c2b896176d-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:56 GMT + Server: + - cloudflare + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "1470" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29998968" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_75fc6debc124b86df16f89e61e53f2fe + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 5-7: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew D. + White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\nlanation + use language and concepts of domain ex-\nperts?\n\u2022 Fidelity/Faithful. Does + the explanation agree with the black box model?\n\u2022 Robust. Does the explanation + change significantly with small changes to the model or\ninstance being explained?\n\u2022 + Sparse/Succinct. Is the explanation succinct?\nWe present an example evaluation + of the SHAP explanation method based on the above\nattributes.44 Shapley values + were proposed as a local explanation method based on feature\nattribution, as + they offer a complete explanation - each feature is assigned a fraction of\nthe + prediction value.44,45 Completeness is a clearly measurable and well-defined + metric, but\nyields explanations with many components. Yet Shapley values are + not actionable nor sparse.\nThey are non-sparse as every feature has a non-zero + attribution and not-actionable because\nthey do not provide a set of features + which changes the outcome.46 Ribeiro et al. 35 proposed\na surrogate model method + that aims to provide sparse/succinct explanations that have high\n5\nfidelity + to the original model. In Wellawatte et al. 9 we argue that counterfactuals + are \u201cbet-\nter\u201d explanations because they are actionable and sparse. + We highlight that, evaluation of\nexplanations is a difficult task because explanations + are fundamentally for and by humans.\nTherefore, these evaluations are subjective, + as they depend on \u201ccomplex human factors and\napplication scenarios.\u201d37\nSelf-explaining + models\nA self-explanatory model is one that is intrinsically interpretable + to an expert.47 Two com-\nmon examples found in the literature are linear regression + models and decision trees (DT).\nIntrinsic models can be found in other XAI + applications acting as surrogate models (proxy\nmodels) due to their transparent + nature.48,49 A linear model is described by the equation\n1 where, W\u2019s + are the weight parameters and x\u2019s are the input features associated with + the\nprediction \u02c6y. Therefore, we observe that the weights can be used + to derive a complete expla-\nnation of the model - trained weights quantify + the importance of each feature.47 DT models\nare another type of self-explaining + models which have been used in classification and high-\nthroughput screening + tasks. Gajewicz et al. 50 used DT models to classify nanomaterials\nthat identify + structural features responsible for surface activity. In another study by Han\net + al. 51, a DT model was developed to filter compounds by their bioactivity based + on the\nchemical fingerprints.\n\u02c6y = \u03a3iWixi\n(1)\nRegularization techniques + such as EXPO52 and RRR53 are designed to enhance the black-\nbox model interpretability.54 + Although one can argue that \u201csimplicity\u201d of models are posi-\ntively + correlated with interpretability, this is based on how the interpretability + is evaluated.\nFor example, Lipton 55 argue that, from the notion of \u201csimulatability\u201d + (the degree to which a\nhuman can predict the outcome based on inputs), self-explanatory + linear models, rule-based\n6\nsystems, and DT\u2019s can be claimed uninterpretable. + A human can pred\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nDo + not directly answer the question, instead summarize to give evidence to help + answer the question. Stay detailed; report specific numbers, equations, or direct + quotes (marked with quotation marks). Reply \"Not applicable\" if the excerpt + is irrelevant. At the end of your response, provide an integer score from 1-10 + on a newline indicating relevance to question. Do not explain your score.\n\nRelevant + Information Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": + false, "temperature": 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4313" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAAwAAAP//dFNRb9MwEH7vrzj5Oa2aUtZtbwgJBEJCgkk8MFRdnEti5viM7wwraP8d + Oc26ToKXyLnvvvN35/v+LACMa801GDug2jH65avXu48fby43TLvf77//qLf28sXbLzefPry59L9M + VRjcfCerj6yV5TF6UsfhCNtEqFSq1rvNbvuy3tUXEzByS77Q+qjLLS836812ub5cri9m4sDOkphr + +LoAAPgzfYvE0NK9uYZ19RgZSQR7MtenJACT2JeIQREnikFN9QRaDkphUn0zENC9pRQVWic2i5CA + luhP9BlLK8Ad0H30GI6/I+nArUDHCUb2ZLPHBDFR6+wxoTQnFXRss7jQAwdA1eSarCTg3R1B51ry + Tg8VJG6yaCCRCjC0IBGTOD2s4J3CSKGULJJQ4fOA0dMBijSSCn4NzhPExD9dW+5BmB+AzgVXgIkg + sAJO+rDxBA1ZzEKl1QO0PMHcdZQAQaicoSPUnGi+2w4Y+ikfOKvlkVbwLkCZZULRakIw68BJAFOf + aeZxDkqpQ6sZvUxabk1DqpRuzblOeS6qJDasw7nq03xoBTeDE5Dc9yQq/77rOBk6r+CCuH5QgeYA + ri3T7Q5ldBLJus7Zuc/HghgAvVKaujt74nkEFYx4V+g60AhZqMt+2ouWrBPHYTnjMbGlslur23Ab + rs63MVGXBYsZQvZ+jj+c1ttzHxM3MuOneOeCk2GfCIVDWWVRjmZCHxYA3yYb5WfOMDHxGHWvfEeh + FLxazzYyT8Z9Quv6EVVW9GfAuv4vb9+SovNyZkdz1OhC/1RifRI6dWrkIErjvnOhpxSTO7qzi/tN + t11fNFd1/cIsHhZ/AQAA//8DACYReYOmBAAA + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99cc7f8867f9-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:57 GMT + Server: + - cloudflare + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "988" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29998968" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_2ef8dc3cb05442e7f16a209a9e6146cc + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 21-22: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew + D. White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\nscussion\nWe + have shown two post-hoc XAI applications based on molecular counterfactual expla-\nnations9 + and descriptor explanations.10 These methods can be used to explain black-box\nmodels + whose input is a molecule. These two methods can be applied for both classification\nand + regression tasks. Note that the \u201ccorrectness\u201d of the explanations + strongly depends on\nthe accuracy of the black-box model.\nA molecular counterfactual + is one with a minimal distance from a base molecular, but\nwith contrasting + chemical properties. In the above examples, we used Tanimoto similar-\nity96 + of ECFP4 fingreprints97 as distance, although this should be explored in the + future.\nCounterfactual explanations are useful because they are represented + as chemical structures\n(familiar to domain experts), sparse, and are actionable. + A few other popular examples of\ncounterfactual on graph methods are GNNExplainer, + MEG and CF-GNNExplainer.69,104,105\nThe descriptor explanation method developed + by Gandhi and White 10 fits a self-explaining\n21\nsurrogate model to explain + the black-box model. This is similar to the GraphLIME87 method,\nalthough we + have the flexibility to use explanation features other than subgraphs. Futher-\nmore, + we show that natural language combined with chemical descriptor attributions + can\ncreate explanations useful for chemists, thus enhancing the accessibility + of DL in chemistry.\nLastly, we examined if XAI can be used beyond interpretation. + Work by Seshadri et al. 31 use\nMMACE and surrogate model explanations to analyze + the structure-property relationships\nof scent. They recovered known structure-property + relationships for molecular scent purely\nfrom explanations, demonstrating the + usefulness of a two step process: fit an accurate model\nand then explain it.\nChoosing + among the plethora of XAI methods described here is still an open question.\nIt + remains to be seen if there will ever be a consensus benchmark, since this field + sits on\nthe intersection of human-machine interaction, machine learning, and + philosophy (i.e., what\nconstitutes an explanation?). Our current advice is + to consider first the audience \u2013 domain\nexperts or ML experts or non-experts + \u2013 and what the explanations should accomplish. Are\nthey meant to inform + data selection or model building, how a prediction is used, or how the\nfeatures + can be changed to affect the outcome. The second consideration is what access + you\nhave to the underlying model. The ability to have model derivatives or + propagate gradients\nto the input to models informs the XAI method.\nConclusion + and outlook\nWe should seek to explain molecular property prediction models + because users are more\nlikely to trust explained predictions, and explanations + can help assess if the model is learning\nthe correct underlying chemical principles. + We also showed that black-box modeling first,\nfollowed by XAI, is a path to + structure-property relationships without needing to trade\nbetween accuracy + and interpretability. However, XAI in chemistry has some maj\n\n----\n\nQuestion: + Are counterfactuals actionable? [yes/no]\n\nDo not directly answer the question, + instead summarize to give evidence to help answer the question. Stay detailed; + report specific numbers, equations, or direct quotes (marked with quotation + marks). Reply \"Not applicable\" if the excerpt is irrelevant. At the end of + your response, provide an integer score from 1-10 on a newline indicating relevance + to question. Do not explain your score.\n\nRelevant Information Summary (about + 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, "temperature": + 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4246" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAAwAAAP//fJNNaxtBDIbv/hViLm3BNrbjxIlvJQk0tCEU0kOpi5Fn5N2p54uRtrUJ + +e9ldh3bKW0vC6t3JD3ilZ56AMoaNQelaxTtkxu8v549PHwy+PH+Qxp9/nU3+RK+Xk9ubmez+9sb + 1S8ZcfWDtLxkDXX0yZHYGDpZZ0KhUnU8m8ym5+PZ+KIVfDTkSlqVZDCNg8loMh2MLgeji31iHa0m + VnP41gMAeGq/BTEY2qo5jPovEU/MWJGaHx4BqBxdiShktiwYRPWPoo5BKLTUjzUBbTXlJGAs64aZ + GKQmaJggrsFHR7pxmEHHJgjlNWpp0AFtk8OAZVYGZEDwJHU0ILHTbICVQ70ZrOIW2nkZbDgpmDIZ + q0sBEOQND+FOgAWlJUD5f8dMsFAN07pxsCKNhVdq2rVKppSJKQiZwqZr8lajA5bcaGkyMbxdo7fO + Yi7AJvrCS9tEWfhdHzhhZuoDBtMWxJYTV46GCwWPtWXgpqqI5a+sDBoDrAiMzaTF7QBTcpYMxAwe + g02NK4tReqOuLf0kMMQ2lxeN6OiJ++BxY0NVpvKnAHBqGjqOUNuqdraqpbPO+hSzYNCtgyWCWjcZ + 9e7l/w9jii+GhLK3Yd8RFkrHXOADMS/UPpPplQ390svtuhyUE0qwgTsiQ4mCgeJyTV2/NwyZnMWV + dVZ2w0VYhMvTBc20bhjLfYTGuX38+bDxLlYpxxXv9UN8bYPlepkJOYay3SwxqVZ97gF8by+reXUs + KuXokywlbiiUgpfnF109dbzlozoeT/eqREF3FK5mo3+lLQ0JWscnB6o6RBuqY4XRgbMdVPGOhfxy + bUNFOWXb3es6LemcplfjM7o8U73n3m8AAAD//wMAaAurb7gEAAA= + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99cd1de4176d-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:57 GMT + Server: + - cloudflare + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "987" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29998973" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_25abef45e5a8ca421f8900cf5d52a9e5 + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 15-17: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew + D. White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\n + on the dataset developed by Martins et al. 124. The\nRF model was implemented + in Scikit-learn125 using Mordred molecular descriptors126 as the\ninput features. + The GRU-RNN model was implemented in Keras.127 See Wellawatte et al. 9\nand + Gandhi and White 10 for more details.\nAccording to the counterfactuals of the + instance molecule in figure 1, we observe that the\nmodifications to the carboxylic + acid group enable the negative example molecule to permeate\nthe BBB. Experimental + findings by Fischer et al. 120 show that the BBB permeation of\nmolecules are + governed by hydrophobic interactions and surface area. The carboxylic group + is\na hydrophilic functional group which hinders hydrophobic interactions and + addition of atoms\nenhances the surface area. This proves the advantage of using + counterfactual explanations,\nas they suggest actionable modification to the + molecule to make it cross the BBB.\nIn Figure 2 we show descriptor explanations + generated for Alprozolam, a molecule that\npermeates the BBB, using the method + described by Gandhi and White 10.\nWe see that\npredicted permeability is positively + correlated with the aromaticity of the molecule, while\n15\nnegatively correlated + with the number of hydrogen bonds donors and acceptors. A similar\nstructure-property + relationship for BBB permeability is proposed in more mechanistic stud-\nies.128\u2013130 + The substructure attributions indicates a reduction in hydrogen bond donors + and\nacceptors. These descriptor explanations are quantitative and interpretable + by chemists.\nFinally, we can use a natural language model to summarize the + findings into a written\nexplanation, as shown in the printed text in Figure + 2.\nFigure 1: Counterfactuals of a molecule which cannot permeate the blood-brain + barrier.\nSimilarity is the Tanimoto similarity of ECFP4 fingerprints.131 Red + indicates deletions and\ngreen indicates substitutions and addition of atoms. + Republished from Ref.9 with permission\nfrom the Royal Society of Chemistry.\nSolubility + prediction\nSmall molecule solubility prediction is a classic cheminformatics + regression challenge and is\nimportant for chemical process design, drug design + and crystallization.133\u2013136 In our previous\nworks,9,10 we implemented + and trained an RNN model in Keras to predict solubilities (log\nmolarity) of + small molecules.127 The AqSolDB curated database137 was used to train the\nRNN + model.\nIn this task, counterfactuals are based on equation 6. Figure 3 illustrates + the generated\nlocal chemical space and the top four counterfactuals. Based + on the counterfactuals, we ob-\nserve that the modifications to the ester group + and other heteroatoms play an important role\nin solubility. These findings + align with known experimental and basic chemical intuition.134\nFigure 4 shows + a quantitative measurement of how substructures are contributing to the pre-\n16\nFigure + 2: Descriptor explanations along with natural language explanation obtained + for BBB\npermeability of Alprozolam molecule. The green and red bars show descriptors + t\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nDo not directly + answer the question, instead summarize to give evidence to help answer the question. + Stay detailed; report specific numbers, equations, or direct quotes (marked + with quotation marks). Reply \"Not applicable\" if the excerpt is irrelevant. + At the end of your response, provide an integer score from 1-10 on a newline + indicating relevance to question. Do not explain your score.\n\nRelevant Information + Summary (about 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, + "temperature": 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4238" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAAwAAAP//fFRNb9tIDL37VxA6dQHZsBM3X7e4e2kuRbE97KJdGNSI0rAdDQczVBqj + yH9fULKTNMX2IgHDeeR7j+T8WABU3FY3UDmP6oYUlrfvLj98+Ha1u/vz/p/zvwM21I3jx+2d3NKB + qtoQ0nwlpyfUysmQAilLnMMuEypZ1s3l2eX27eZyczEFBmkpGKxPutzK8mx9tl2ur5briyPQCzsq + 1Q18XgAA/Ji+RjG29FDdwLo+nQxUCvZU3TxdAqiyBDupsBQuilGr+jnoJCrFifUnT0APjnJSSFnu + uaUCZL/oCNSjgpMxKuUOnY4YCmAmQGcasQm0gvcKRVGpzNcHabljh3ahgAqoJ3CYG3k4BHaAjlvo + s4wJOALCIIHcGOxOBJpyAqsBE+XB7JsyNEGkXTYZOUKDOTNleLPb7f6oAcsrjkAPKWA8Uihj31PR + F5zBeYw9TewoejSpc7GGA+thBX8lcpOKEA71VB/blg0P0gGqDAU4WncLWQWrTOYM1iD3lJ0MHPsJ + 6A9tluTZtEfUMZOleGXK5EcN3z07D956nMsJKQ07YJM3KygQyVnP8wE6ybDb7U5WscQV3B6Zztw5 + QpEwzsIgZWp5ylL/0leOrbWNftNGKkr52DyMLXhSyjLbYWPh8ugYQw0YuI/mwHdWb+2gzANFxQCd + 1Yl9mRI4T4O5bPpGnvl/8lRsKNFWqUBLg8Si+TQIabLBMJhSYHfsmZn6q6LTdGH+WdDqS/wSr1/u + RKZuLGgrGccQjuePT0sWpE9ZmnKMP513HLn4vQ2CRFuoopKqKfq4APh3Wubxp/2sUpYh6V7lG0VL + eHV9Meernp+P5+jm/PoYVVEMLwLr87f/h9u3pMihvHgUqpkjx/45xfqJ6KS0KoeiNOw7jj3llHl+ + I7q0P+u264vmerM5rxaPi/8AAAD//wMAbVHCmSwFAAA= + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99c969105c1c-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:58 GMT + Server: + - cloudflare + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "1889" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29998974" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_e2638adada7d4bd806097f0f99594bf7 + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Summarize + the excerpt below to help answer a question.\n\nExcerpt from wellawatteUnknownyearaperspectiveon + pages 20-21: Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew + D. White. A perspective on explanations of molecular prediction models. Unknown + journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\n----\n\ntics + Hamburg) de-\nveloped by Schomburg et al. 132.\n19\nthe \u2018fruity\u2019 scent. + There are some exceptions though, like tert-amyl acetate which has a\n\u2018camphoraceous\u2019 + rather than \u2018fruity\u2019 scent.140,141\nIn Seshadri et al. 31, we trained + a GNN model to predict the scent of molecules and utilized\ncounterfactuals9 + and descriptor explanations10 to quantify scent-structure relationships. The\nMMACE + method was modified to account for the multi-label aspect of scent prediction. + This\nmodification defines molecules that differed from the instance molecule + by only the selected\nscent as counterfactuals. For instance, counterfactuals + of the jasmone molecule would be false\nfor the \u2018jasmine\u2019 scent but + would still be positive for \u2018woody,\u2019 \u2018floral\u2019 and \u2018herbal\u2019 + scents.\nFigure 5: Counterfactual for the 2,4 decadienal molecule.\nThe counterfactual + indicates\nstructural changes to ethyl benzoate that would result in the model + predicting the molecule\nto not contain the \u2018fruity\u2019 scent. The Tanimoto96 + similarity between the counterfactual and\n2,4 decadienal is also provided. + Republished with permission from authors.31\nThe molecule 2,4-decadienal, which + is known to have a \u2018fatty\u2019 scent, is analyzed in Fig-\nure 5.142,143 + The resulting counterfactual, which has a shorter carbon chain and no carbonyl\ngroups, + highlights the influence of these structural features on the \u2018fatty\u2019 + scent of 2,4 deca-\ndienal. To generalize to other molecules, Seshadri et al. + 31 applied the descriptor attribution\nmethod to obtain global explanations + for the scents. The global explanation for the \u2018fatty\u2019\nscent was + generated by gathering chemical spaces around many \u2018fatty\u2019 scented + molecules.\nThe resulting natural language explanation is: \u201cThe molecular + property \u201cfatty scent\u201d can\nbe explained by the presence of a heptanyl + fragment, two CH2 groups separated by four\n20\nbonds, and a C=O double bond, + as well as the lack of more than one or two O atoms.\u201d31\nThe importance + of a heptanyl fragment aligns with that reported in the literature, as \u2018fatty\u2019\nmolecules + often have a long carbon chain.144 Furthermore, the importance of a C=O dou-\nble + bond is supported by the findings reported by Licon et al. 145, where in addition + to a\n\u201clarger carbon-chain skeleton\u201d, they found that \u2018fatty\u2019 + molecules also had \u201caldehyde or acid\nfunctions\u201d.145 For the \u2018pineapple\u2019 + scent, the following natural language explanation was ob-\ntained: \u201cThe + molecular property \u201cpineapple scent\u201d can be explained by the presence + of ester,\nethyl/ether O group, alkene/ether O group, and C=O double bond, as + well as the absence of\nan Aromatic atom.\u201d31 Esters, such as ethyl 2-methylbutyrate, + are present in many pineap-\nple volatile compounds.146,147 The combination + of a C=O double bond with an ether could\nalso correspond to an ester group. + Additionally, aldehydes and ketones, which contain C=O\ndouble bonds, are also + common in pineapple volatile compounds.146,148\nDiscussion\nWe have shown two + post-hoc XAI applications based on molecular counterfactual expla-\nnation\n\n----\n\nQuestion: + Are counterfactuals actionable? [yes/no]\n\nDo not directly answer the question, + instead summarize to give evidence to help answer the question. Stay detailed; + report specific numbers, equations, or direct quotes (marked with quotation + marks). Reply \"Not applicable\" if the excerpt is irrelevant. At the end of + your response, provide an integer score from 1-10 on a newline indicating relevance + to question. Do not explain your score.\n\nRelevant Information Summary (about + 100 words):"}], "model": "gpt-4o-2024-08-06", "stream": false, "temperature": + 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "4438" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAA3RTy24bORC86ysac5YESVbixy0IkCwSxLuH7F7WgdBDNmeY5bAH7KZtJTDgz3B+ + z1+yICXLioNcBpiu6mb1o75PABpvmwtoTI9qhjHM3rw9/fOy++fjt/78wx/DX98+fn7/4dPf1+31 + p607a6Ylg9uvZPQpa254GAOp57iDTSJUKlWXp6vT9avl6fJVBQa2FEpaN+pszbPVYrWeLc5mi9f7 + xJ69IWku4N8JAMD3+i0So6Xb5gIW06fIQCLYUXNxIAE0iUOJNCjiRTFqM30GDUelWFV/7gno1lAa + FawXk0VIQHuCLATswHCOSsmh0YxBwEcYE1lv1McOBg5kcsAEYiiqQJYSRnh/eQm1xzm8fVEBE4El + 5yNZQHkqQQLWO0ep5LvEQxXhYxFv6MCCdgscw7aiQoGMkt09Pod3nIBusaxg+otwdjXnK8rA8ajg + DedgoSVwGITAcaq8x/uHQvWRHu9/7B6ANiuMLF799Y74eP9ww2y308J5vH9wgROG+ofRlkhPqcVw + KCFzKBPHiGErvopaTdczSwatp4gBpOebsgBUEE3ZaE4YwPQYO5IpSDZ9mRoWYlJKYDC1HAvDx/oq + tkJlZGV5FdsG6BLnUabgowu5ol6lKkbV7UFeUecFJHcdie5lvBykwXICfO0tAZpy69iGuirf9Spl + Qz5ab7BeiIxkvPPmuJuBbQlhyRWIRJYsKAOG0tBu1k9HxlHmV/Eqnh0fcCKXBYt/Yg5hH787OCJw + NyZuZY8f4s5HL/0mEQrHcv2iPDYVvZsAfKnOyz+ZqRkTD6NulP+jWAqen5/s6jXPXn9Gl+u9Lxtl + xXAELE9+m7expOiDHDm42Wn0sXsusTgIrZ02shWlYeN87CiNye8M7cbNyq0Xr9vz5fKkmdxN/gcA + AP//AwDzfvso2QQAAA== + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99c84da3252d-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:41:58 GMT + Server: + - cloudflare + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "2157" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29998954" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 2ms + x-request-id: + - req_6afb578836c2390464966ed7361e5365 + status: + code: 200 + message: OK + - request: + body: + '{"messages": [{"role": "system", "content": "Answer in a direct and concise + tone. Your audience is an expert, so be highly specific. If there are ambiguous + terms or acronyms, first define them."}, {"role": "user", "content": "Answer + the question below with the context.\n\nContext (with relevance scores):\n\nwellawatteUnknownyearaperspectiveon + pages 11-12: Counterfactual explanations are described as actionable in the + context of XAI (Explainable Artificial Intelligence) for chemistry. They are + used to determine \"which smallest change could alter the outcome of an instance + of interest,\" indicating their actionable nature. The excerpt explains that + counterfactuals are generated by minimizing the vector distance \\(d(x, x'')\\) + between features, subject to a change in prediction \\(\\hat{f}(x) \\neq \\hat{f}(x'')\\) + for classification tasks, and \\(|\\hat{f}(x) - \\hat{f}(x'')| \\geq \\Delta\\) + for regression tasks. This suggests that counterfactuals provide \"intuitive + understanding of predictions\" and can uncover spurious relationships, making + them actionable at the instance level.\n\n9\nFrom Geemi P. Wellawatte, Heta + A. Gandhi, Aditi Seshadri, and Andrew D. White. A perspective on explanations + of molecular prediction models. Unknown journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, + doi:10.26434/chemrxiv-2022-qfv02.\n\nwellawatteUnknownyearaperspectiveon pages + 15-17: The excerpt provides evidence that counterfactuals are actionable. It + states that modifications to the carboxylic acid group in a molecule can enable + it to permeate the blood-brain barrier (BBB), as counterfactual explanations + suggest actionable changes to enhance permeability. Specifically, the addition + of atoms increases surface area, overcoming the hydrophilic nature of the carboxylic + group, which hinders hydrophobic interactions necessary for BBB permeation. + Additionally, in solubility prediction, counterfactuals indicate that modifications + to the ester group and heteroatoms are crucial, aligning with experimental findings + and chemical intuition. These examples demonstrate the practical applicability + of counterfactuals in molecular modifications.\n\n9\nFrom Geemi P. Wellawatte, + Heta A. Gandhi, Aditi Seshadri, and Andrew D. White. A perspective on explanations + of molecular prediction models. Unknown journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, + doi:10.26434/chemrxiv-2022-qfv02.\n\nwellawatteUnknownyearaperspectiveon pages + 5-7: The excerpt discusses the evaluation of explanation methods for molecular + prediction models, focusing on attributes like fidelity, robustness, and sparsity. + It mentions that Shapley values, while providing a complete explanation, are + not actionable because they do not offer a set of features that change the outcome. + In contrast, the authors argue that counterfactuals are \"better\" explanations + because they are both actionable and sparse. This suggests that counterfactuals + provide actionable insights by identifying specific changes that can alter the + prediction outcome, making them useful for decision-making processes.\n\n9\nFrom + Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and Andrew D. White. A + perspective on explanations of molecular prediction models. Unknown journal, + Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, doi:10.26434/chemrxiv-2022-qfv02.\n\nwellawatteUnknownyearaperspectiveon + pages 12-14: Counterfactuals are described as providing \"local (instance-level), + actionable explanations.\" The actionability of a counterfactual explanation + is demonstrated by suggesting which features can be altered to change the outcome, + such as modifying a hydrophobic functional group in a molecule to a hydrophilic + one to increase solubility. The excerpt discusses various methods for generating + counterfactuals, including CF-GNNExplainer, MEG, and MMACE, each with different + approaches and limitations. Despite these methods, the excerpt notes that counterfactual + explanation-based model interpretation is less explored compared to other post-hoc + methods.\n\n8\nFrom Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, and + Andrew D. White. A perspective on explanations of molecular prediction models. + Unknown journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, + doi:10.26434/chemrxiv-2022-qfv02.\n\nwellawatteUnknownyearaperspectiveon pages + 35-36: The excerpt references several works related to counterfactual explanations, + which are a method for providing interpretability in AI models. Specifically, discuss + counterfactual explanations in the context of automated decisions and the General + Data Protection Regulation (GDPR), suggesting their potential role in legal + accountability. and Numeroso & Bacciu (2020) explore counterfactual explanations + for Graph Neural Networks, indicating their applicability in complex model interpretability. + Freiesleben (2022) examines the relationship between counterfactual explanations + and adversarial examples, which may imply actionability in terms of model robustness + and decision-making.\n\n8\nFrom Geemi P. Wellawatte, Heta A. Gandhi, Aditi Seshadri, + and Andrew D. White. A perspective on explanations of molecular prediction models. + Unknown journal, Unknown year. URL: https://doi.org/10.26434/chemrxiv-2022-qfv02, + doi:10.26434/chemrxiv-2022-qfv02.\n\nValid keys: wellawatteUnknownyearaperspectiveon + pages 11-12, wellawatteUnknownyearaperspectiveon pages 15-17, wellawatteUnknownyearaperspectiveon + pages 5-7, wellawatteUnknownyearaperspectiveon pages 12-14, wellawatteUnknownyearaperspectiveon + pages 35-36\n\n----\n\nQuestion: Are counterfactuals actionable? [yes/no]\n\nWrite + an answer based on the context. If the context provides insufficient information + reply \"I cannot answer.\"For each part of your answer, indicate which sources + most support it via citation keys at the end of sentences, like (Example2012Example + pages 3-4). Only cite from the context below and only use the valid keys. Write + in the style of a Wikipedia article, with concise sentences and coherent paragraphs. + The context comes from a variety of sources and is only a summary, so there + may inaccuracies or ambiguities. If quotes are present and relevant, use them + in the answer. This answer will go directly onto Wikipedia, so do not add any + extraneous information.\n\nAnswer (about 200 words, but can be longer):"}], + "model": "gpt-4o-2024-08-06", "stream": false, "temperature": 0.0}' + headers: + accept: + - application/json + accept-encoding: + - gzip, deflate + connection: + - keep-alive + content-length: + - "6452" + content-type: + - application/json + host: + - api.openai.com + user-agent: + - AsyncOpenAI/Python 1.46.1 + x-stainless-arch: + - arm64 + x-stainless-async: + - async:asyncio + x-stainless-lang: + - python + x-stainless-os: + - MacOS + x-stainless-package-version: + - 1.46.1 + x-stainless-raw-response: + - "true" + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.12.6 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: !!binary | + H4sIAAAAAAAAA5RV22rcSBB991cUeophNIzG44zjNxM2YJa9hN3AhvViSt0lqexWt+gqzXgS/O9L + S5qLk+yCXwTquvSpU6e6vp4BZGyza8hMg2razuU379e/fdz8/PC5ePvrori8/f2j6Ntf3oWb4sF8 + ymYpIpQPZHQfNTeh7RwpBz+aTSRUSlmL9XK9uizWxdVgaIMll8LqTvNVyJeL5SpfXOWLt1NgE9iQ + ZNfw9xkAwNfhmyB6S0/ZNSxm+5OWRLCm7PrgBJDF4NJJhiIsil6z2dFoglfyA+rPJDMwofdKsUKj + PToBjARoUhVYOprDrQdtCIawJ4VQwU9PnUMezHATlSs2jA5uvZJzXJM3BG/+urk9hypEMA21LBp3 + 314FlPJ4TFeN11oSE7kkCygnGKAkg71QwrEDS0qxZT/8grToHImCadDXBJ7IkgUNgE4pDj6hVxNa + StDRA/tEiRl+OcEh0Tn82bAAp1sbpg1ZKHfQsueWv7CvhzQbMhoiWJ7iS9ItkYeKUPtIAtuGHQF5 + 6WOKwT0m9tBFsjwUNEtupoEuhg1bkhGS9qy8Iei9pZjy25QhVCeBydOCQQ+9N2FDEaTrI4deIJIb + WWy4E3izJedwi6r0yT/6sPU7wogdRenIpGuChw5rEiiKvFiez+/8nX//AxmY4IUtRbJwl5WkSvEu + e9m2pHmMI+MtaROsgONHgj8a7BztYIOuJ3nZwpS7DNqc9jgVJx1GodQM2gFb8srVDhLqpLGJTgFt + UAciji0+0rTv9gxafJxa10IvVPVu0KMlw8LB55O9i8GQCL2KuMt8fT6HDyECPWGa+lnqchscmd5h + hDbYBHkk6fsZk76uk2hP6p+Km4H0pkn6H1Ls9uIzGMvwtHNsAA1bqGPou0Q6+WZUowvB5mVE9lBi + jEwROootYcmOdQchjnyljCSJuCGHpCTcJjkSSHD96D4DdFz75LxlbVLTKXJLXtFBxYM+J0Wm+Tbo + 9ioO/lUKvMyL9ajAD33UhmIbIn3P2DQu4IJBNzsMce5oQ9+8JOVuz++Afpi2w4wm2ZQ0MjGqdhrS + k4fiRz2oej/2yp3S5tMTL6e8var2ZV6szge5C+2FJGCpDV40otKk7SSTxDB2nWOzb2gify+gqcXV + /z6x7GGD44sxveYyP90MkapeMC0m3zs3nT8fVo0LdRdDKZP9cF6xZ2nuExPBp7UiGrpssD6fAfwz + rLT+xZbKuhjaTu81PJJPCYvVxcWYMDtu0aN5ebWarBoU3Unculj/V9y9JUV2crIbsxEk+/qYYnFA + OpSayU6U2vuKfU2xizyuyqq7p0tavSsu6OoiO3s++xcAAP//AwBVC1SrMwgAAA== + headers: + CF-Cache-Status: + - DYNAMIC + CF-RAY: + - 8c9c99d72a4a22de-SJC + Connection: + - keep-alive + Content-Encoding: + - gzip + Content-Type: + - application/json + Date: + - Fri, 27 Sep 2024 15:42:02 GMT + Server: + - cloudflare + Set-Cookie: + - __cf_bm=LWPff5zX3ICbbT9QScjPn280Ah9S0inGrZ6ybIIewDI-1727451722-1.0.1.1-q5KgaN4N5L0dHLZN75tHjBW_AtJ5vsnl6zmfkPZuX3nQb6ofrTsGnI.tov8..L6pDl6EchlzSw2ix2jwjXBTjQ; + path=/; expires=Fri, 27-Sep-24 16:12:02 GMT; domain=.api.openai.com; HttpOnly; + Secure; SameSite=None + - _cfuvid=PWVdd0wya80sePSgKjRZyOELFVjKfBs7ecUlL0T1EI0-1727451722792-0.0.1.1-604800000; + path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - nosniff + access-control-expose-headers: + - X-Request-ID + alt-svc: + - h3=":443"; ma=86400 + openai-organization: + - future-house-xr4tdh + openai-processing-ms: + - "3986" + openai-version: + - "2020-10-01" + strict-transport-security: + - max-age=31536000; includeSubDomains; preload + x-ratelimit-limit-requests: + - "10000" + x-ratelimit-limit-tokens: + - "30000000" + x-ratelimit-remaining-requests: + - "9999" + x-ratelimit-remaining-tokens: + - "29998421" + x-ratelimit-reset-requests: + - 6ms + x-ratelimit-reset-tokens: + - 3ms + x-request-id: + - req_9670a7a1e5925640b8feb8354787ee0e + status: + code: 200 + message: OK version: 1 diff --git a/tests/test_paperqa.py b/tests/test_paperqa.py index c98d6bd1..7d92a86c 100644 --- a/tests/test_paperqa.py +++ b/tests/test_paperqa.py @@ -813,10 +813,11 @@ def test_pdf_reader_w_no_match_doc_details(stub_data_dir: Path) -> None: next(iter(docs.docs.values())).citation == "Wellawatte et al, XAI Review, 2023" ) + # default: ['method', 'scheme', 'host', 'port', 'path', 'query'] # this is the default list + body # this ensures vcr distinguishes requests with a different body -@pytest.mark.vcr(match_on=['method', 'scheme', 'host', 'port', 'path', 'query', 'body']) +@pytest.mark.vcr(match_on=["method", "scheme", "host", "port", "path", "query", "body"]) def test_pdf_reader_match_doc_details(stub_data_dir: Path) -> None: docs = Docs() docs.add(