Revamp balances and incentivisation #949
Conversation
Cosmetic changes related to `get_justification_and_finalization_deltas`: * Review naming of misc helper functions and variables * Abstract away common logic and rework for readability * Add `MAX_FINALITY_LOOKBACK` and `BASE_REWARDS_PER_EPOCH` constants * Rescale `INACTIVITY_PENALTY_QUOTIENT` Substantive changes: * Make logic relative to `previous_epoch` throughout (as opposed to mixing `current_epoch` and `previous_epoch`) * Replace inclusion delay bonus by an inclusion delay penalty
JustinDrake
changed the title
|
@JustinDrake could you give a summary of how exactly the rewards/penalty change through this? I would like to read it with this info side by side :) |
|
See the "substantive changes" section (which I fleshed out a bit) in the PR description. |
Co-Authored-By: JustinDrake <drakefjustin@gmail.com>
| | `INACTIVITY_PENALTY_QUOTIENT` | `2**24` (= 16,777,216) | | ||
| | `MIN_PENALTY_QUOTIENT` | `2**5` (= 32) | | ||
| | `INACTIVITY_PENALTY_QUOTIENT` | `2**25` (= 33,554,432) | | ||
| | `MIN_SLASHING_PENALTY_QUOTIENT` | `2**5` (= 32) | | ||
|
|
||
| * **The `BASE_REWARD_QUOTIENT` is NOT final. Once all other protocol details are finalized it will be adjusted, to target a theoretical maximum total issuance of `2**21` ETH per year if `2**27` ETH is validating (and therefore `2**20` per year if `2**25` ETH is validating, etc etc)** | ||
| * The `INACTIVITY_PENALTY_QUOTIENT` equals `INVERSE_SQRT_E_DROP_TIME**2` where `INVERSE_SQRT_E_DROP_TIME := 2**12 epochs` (~18 days) is the time it takes the inactivity penalty to reduce the balance of non-participating [validators](#dfn-validator) to about `1/sqrt(e) ~= 60.6%`. Indeed, the balance retained by offline [validators](#dfn-validator) after `n` epochs is about `(1 - 1/INACTIVITY_PENALTY_QUOTIENT)**(n**2/2)` so after `INVERSE_SQRT_E_DROP_TIME` epochs it is roughly `(1 - 1/INACTIVITY_PENALTY_QUOTIENT)**(INACTIVITY_PENALTY_QUOTIENT/2) ~= 1/sqrt(e)`. |
There was a problem hiding this comment.
If INACTIVITY_PENALTY_QUOTIENT changes from 2**25 to 2**24, then INACTIVITY_PENALTY_QUOTIENT != INVERSE_SQRT_E_DROP_TIME**2 now. This sentence needs to be updated or we change INACTIVITY_PENALTY_QUOTIENT back to 2**24 with INACTIVITY_PENALTY_QUOTIENT // 2 in formula.
There was a problem hiding this comment.
I think this sentence should go away. We should have a separate document explaining design decisions/rationales. (This will be included in the Transparent Paper, whenever I get round to writing it.)
There was a problem hiding this comment.
Yeah, it's also mentioned in @vbuterin's https://notes.ethereum.org/s/rkhCgQteN#Slashing-and-anti-correlation-penalties. We could (i) update rationales doc (ii) remove this sentence from the spec and link to rationale doc for now.
Co-Authored-By: JustinDrake <drakefjustin@gmail.com>
Substantive changes:
validator.effective_balance) throughout beacon chain consensusvalidator.effective_balanceonce per epochPendingAttestation.proposer_indexfor attestation proposer rewardsCosmetic changes (mostly related to
get_attestation_deltas):MIN_EPOCHS_TO_INACTIVITY_PENALTYandBASE_REWARDS_PER_EPOCHconstantsINACTIVITY_PENALTY_QUOTIENT