diff --git a/.github/workflows/diff.py b/.github/workflows/diff.py
index c77d055a..f021d577 100644
--- a/.github/workflows/diff.py
+++ b/.github/workflows/diff.py
@@ -1,8 +1,15 @@
+import stats
 import sys
 import pandas as pd
 import markdown
 import re
 
+if len(sys.argv) == 2 and sys.argv[1] == "final":
+    stats.load_stats()
+    print("# Overall Statistics\n")
+    stats.output_stats()
+    sys.exit(0)
+
 if len(sys.argv) < 3:
     print("Usage: python diff.py [current.md] [main.md]")
     sys.exit(1)
@@ -46,6 +53,8 @@ def read_tables(file):
         print(f"> **Warning**\n> Skip {file}. File not found.\n")
         sys.exit(0)
 
+
+
 current = read_tables(sys.argv[1])
 main = read_tables(sys.argv[2])
 
@@ -53,6 +62,8 @@ def read_tables(file):
     print(f"> **Warning**\n> Skip {sys.argv[1]}, due to the number of tables mismatches from main branch.\n")
     sys.exit(0)
 
+flaky_benchmarks = ["## Heartbeat"]
+
 for i, ((header, current), (header2, main)) in enumerate(zip(current, main)):
     if header == header2 and current.shape == main.shape and all(current.columns == main.columns) and all(current.index == main.index):
         result = pd.DataFrame(index=current.index, columns=current.columns)
@@ -72,8 +83,20 @@ def read_tables(file):
                     result.loc[idx, col] = f"{x:_} ($\\textcolor{{red}}{{{d:.2f}\\\\%}}$)"
                 else:
                     result.loc[idx, col] = f"{x:_}"
+                if header in flaky_benchmarks:
+                    continue
+                if col.endswith("binary_size"):
+                    stats.data["binary_size"].append(d)
+                elif col.endswith("max mem"):
+                    stats.data["max_mem"].append(d)
+                else:
+                    stats.data["cycles"].append(d)
         print(result.to_markdown())
         print(f"\n")
     else:
         print(f"> **Warning**\n> Skip table {i} {header} from {sys.argv[1]}, due to table shape mismatches from main branch.\n")
 
+print(f"## Statistics\n\n")
+stats.output_stats()
+
+stats.save_stats()
diff --git a/.github/workflows/perf.yml b/.github/workflows/perf.yml
index f2ee2a57..b2ddf3c4 100644
--- a/.github/workflows/perf.yml
+++ b/.github/workflows/perf.yml
@@ -14,14 +14,21 @@ jobs:
       DFX_VERSION: 0.14.3
       IC_REPL_VERSION: 0.4.1
       MOC_VERSION: 0.9.7
+      IC_WASM_VERSION: 0.4.0
     steps:
       - uses: actions/checkout@v3
       - name: Checkout out gh-pages report
-        if: github.event_name == 'pull_request'
+        if: github.event_name == 'pull_request' && !contains(github.event.pull_request.labels.*.name, 'build_base')
         uses: actions/checkout@v3
         with:
           ref: gh-pages
           path: main/_out
+      - name: Checkout out base branch
+        if: github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'build_base')
+        uses: actions/checkout@v3
+        with:
+          ref: ${{ github.base_ref }}
+          path: main/
       - uses: actions-rs/toolchain@v1
         with:
           profile: minimal
@@ -44,7 +51,7 @@ jobs:
         if: github.event_name == 'pull_request'
         run: |
           python -m pip install --upgrade pip
-          pip install pandas markdown lxml html5lib bs4 tabulate
+          pip install pandas markdown lxml html5lib bs4 tabulate scipy
       - uses: actions/setup-node@v3
         with:
           node-version: 18
@@ -59,10 +66,19 @@ jobs:
           cd $(dfx cache show)
           wget https://github.com/dfinity/motoko/releases/download/$MOC_VERSION/motoko-linux64-$MOC_VERSION.tar.gz
           tar zxvf motoko-linux64-$MOC_VERSION.tar.gz
-          cargo install --git https://github.com/dfinity/ic-wasm.git
+          wget https://github.com/dfinity/ic-wasm/releases/download/$IC_WASM_VERSION/ic-wasm-linux64
+          cp ./ic-wasm-linux64 /usr/local/bin/ic-wasm
+          chmod a+x /usr/local/bin/ic-wasm
       - name: Start dfx
         run: |
           dfx start --background
+      - name: Run perf for base branch
+        if: github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'build_base')
+        run: |
+          cd main
+          make
+          dfx stop
+          dfx start --clean --background
       - name: Run perf
         run: make
       - name: Generate table
@@ -103,6 +119,7 @@ jobs:
               fi
             fi
           done
+          python .github/workflows/diff.py final >> DIFF.md
       - name: Read table
         if: github.event_name == 'pull_request'
         id: perf
diff --git a/.github/workflows/stats.py b/.github/workflows/stats.py
new file mode 100644
index 00000000..3674da46
--- /dev/null
+++ b/.github/workflows/stats.py
@@ -0,0 +1,41 @@
+import statistics
+import pickle
+from scipy.stats import t
+
+data = {'binary_size': [], 'max_mem': [], 'cycles': []}
+file = "stats.pk"
+
+def stats(array, ignoreZeros=True):
+    if ignoreZeros:
+        array = [x for x in array if x != 0.]
+    if len(array) == 0:
+        return f"no change"
+    elif len(array) == 1:
+        return f"{array[0]:.2f}%"
+    mean = statistics.mean(array)
+    std = statistics.stdev(array)
+    conf_level = 0.9
+    t_value = t.ppf(1 - (1 - conf_level) / 2, len(array) - 1)
+    interval = t_value * std / len(array)**0.5
+    l, r = mean - interval, mean + interval
+    res = f"{mean:.2f}% [{l:.2f}%, {r:.2f}%]"
+    return res
+
+def output_stats():
+    for name, array in data.items():
+        print(f"* {name}: {stats(array)}")
+
+def save_stats():
+    with open(file, 'ab') as fp:
+        pickle.dump(data, fp)
+
+def load_stats():
+    with open(file, 'rb') as fr:
+        try:
+            while True:
+                d = pickle.load(fr)
+                for name, array in d.items():
+                    data[name].extend(array)
+        except EOFError:
+            pass
+
diff --git a/Cargo.toml b/Cargo.toml
index 926d64b6..00169588 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -17,7 +17,7 @@ members = [
 [profile.release]
 panic = "abort"
 lto = true
-opt-level = 3
+opt-level = 2
 
 [workspace.dependencies]
 ic-cdk = "0.10.0"
diff --git a/collections/perf.sh b/collections/perf.sh
index f6fe69f6..8690d717 100644
--- a/collections/perf.sh
+++ b/collections/perf.sh
@@ -56,7 +56,7 @@ perf(zhenya, "zhenya_hashmap", init_size);
 perf(btreemap_rs, "btreemap_rs", init_size);
 perf(hashmap_rs, "hashmap_rs", init_size);
 
-output(file, "\n## Priority queue\n\n| |binary_size|heapify 50k|mem|pop_min 50|put 50|\n|--:|--:|--:|--:|--:|--:|\n");
+output(file, "\n## Priority queue\n\n| |binary_size|heapify 50k|max mem|pop_min 50|put 50|\n|--:|--:|--:|--:|--:|--:|\n");
 perf(heap, "heap", init_size);
 perf(heap_rs, "heap_rs", init_size);