Evaluator Rewards

[JuaniRios]

Objective

Implement automatic rewards for evaluations based on the knowledge hub logic.

Changes Overview

• We now have a state machine that handles the "cleaning" of a finished project. We call it a "finalizer"

  • The finalizer is triggered in on_idle on a per-operation basis
  • each operation has to return the used weight
  • The finalizer has a funding success and failure version
  • For now, only the evaluation rewards are implemented.
  • Other cleaning operations are mentioned but not implemented

• Bids over the final weighted average price are counted down to the average price

  • The difference is returned to the bidder

• Unbonding an evaluation is done in a single extrinsic, irrespective of if the funding was a success or not.

• We can now represent a failed or successful funding for the project, and handle it with the new state machine.

Issues

• PLMC unbonding tests are broken
• Failed projects are panicking the runtime once on_initialize tries to initialize the finalizer.
  • This is because a new logic was added to check if the project needs to be manually accepted for funding or not. and only the automatic acceptance when the funding is >90% is implemented.
• Current cleaning operations of the finalizer return weight 0. This means all of the finalizations will occur inside one block, which might lead to longer than expected block times.
  • We should benchmark each cleaning operation and return those weights instead

[linear]

PLMC-214 Finalize Evaluations

Implement slashing and rewards

[lrazovic]

From an offline review:

• Calculating the total bonded in evaluations every time you evaluate is inefficient. We should store the total bonded somewhere and add to this on every evaluation.
• Use StorageNMap instead of having a BoundedVec for all the evaluations/bids/contributions for a single project/user pair.
• Use PerQuintill instead of Perbill. Perbill only has 9 zeroes, and our Balance type uses 10, so we have precision problems right now, which we circumvent in tests with the new macro assert_close_enough
• We are not explicitly testing failed bid currency refunds
• Test bids with original price higher than weighed average are returned the difference on the community funding new_with method.
• Failed bid return is implemented on BadMath rejection, and it needs to be implemented as well in after candle end
• Write down test for CT sold out on remainder round. Right now it will fail because of the ensure in do_funding_end
• Write test for PLMC unbonding on success and fail evaluation rounds
• Extract success or fail funding logic in do_end_funding into a function to call by the issuer in case of needing a manual acceptance (certain funding thresholds)

[JuaniRios]

• Total bonded calculation now stored in ProjectDetails
• StorageNmap change is reserved for another PR
• Perbill is now PerQuintill. assert_close_enough macro is now used with 0.01 parts per billion instead of 1 part. Calculation of reward still leads to an error though
• Failed bids due to after block end, and ct sold out are now tested for funding return
• The possibility of checking that a final price for a bid is lower is actually there, inside BidInfoFilter.
• CT sold out on remainder round now tested
• Tests for PLMC evaluation unbond in success and fail funding are written, but only success passes since the failure path is unimplemented yet.
• Extraction feedback is not clear enough, and would make sense to extract once we actually need it, so leaving it off for now.

[joepetrowski]

Suggested change

	consumed_weight += weight
	consumed_weight.saturating_accrue(weight);

[linear]

PLMC-214 Finalize Evaluations

Implement slashing and rewards