Links

Reputation & Slashing

Validators are generally expected to be honest actors, but penalties must exist to ensure the reliable performance of Validator duties and for countering subversive behaviour. Validators hold shares in keys which can not be used to move protocol funds unless at least 100 of the 150 maximum Validators are online and can sign transactions. The protocol therefore needs to make sure sufficient penalties exist to incentivise consistent, secure, and high-performance Validators.
That being said, whilst it's necessary to discourage Validators from going offline at any point, there are legitimate reasons to do so which we should respect. Updating Validator software is one such legitimate reason, and penalising Validators for minor periods of being offline could also lead to negative results for the network.
Taking this all into account, Chainflip operates a unified slashing and reputation system that rolls all penalties into a time-based system with both positive and negative reputation scores, designed to balance the above considerations.
Validators start with their reputation at 0. Until they join the Authority Set, their reputation will stay at 0. Validators can be in one of three onchain states which affect their Reputation score:
  1. 1.
    Online - The Validator is submitting heartbeats and is not suspended - the Validator will earn rewards and will not be slashed while in this state regardless of their reputation score, and will gain reputation over time.
  2. 2.
    Offline - The Validator has failed to submit a heartbeat, which is due to some or all of the Validators' critical functions failing. Reputation will decrease over time while in the offline state.
  3. 3.
    Suspended - The Validator has been given a time-penalty for failing to complete a process in time or for doing something suspicious or incorrectly. The Validator will lose as much reputation as if they had been offline for the duration of the penalty.
If a Validator has a negative reputation and is offline or suspended, they will be periodically slashed until they achieve the Online state again. The more negative the reputation, the greater the rate of slashing.

Heartbeats & Positive Reputation

Reputation will be accrued by a Validator as long as they are considered “online” on the State Chain. It makes sense to cap the amount of uptime credit available to any single Validator in order to prevent some strange scenario where a long-running Validator can be offline for many days at a time without suffering punishment. Chainflip has an initial cap of 48 hours on uptime credit, which should be more than enough time for any honest Validator to debug issues with their setup. This cap may be reduced in the future.
Being online is the only way to accrue reputation. If a Validator has a negative balance but is back online, they will start being slashed as soon as they are offline or suspended again. This is a strong incentive to fix issues and be reliable long before a reputation balance goes negative.
The way Chainflip ascertains whether a Validator is online or not is through Heartbeat Extrinsics. At regular intervals, Validators must submit a special transaction to the state chain, simply proving that their hardware is connected and active, though it doesn’t prove their overall performance.
If a Validator doesn’t submit a valid heartbeat before the interval deadline occurs on-chain, they are automatically moved into the Offline state and will start losing reputation. The Validator will be moved back into the Online state at the next interval deadline that they submit a valid heartbeat for.

Suspension Conditions

For now, Validators can be suspended for two reasons, both relating to the Egress process:
  1. 1.
    Failing to Participate in a Signing Ceremony: During a signing ceremony, it is possible that a Validator fails to sign or communicate messages before the ceremony times out. This has the impact of requiring the entire signing ceremony to be conducted again, costing users valuable time. After a failed ceremony, the online Validators vote through consensus on Validators which misbehaved or failed in the signing ceremony, knocking them into the suspended state. The ceremony is then attempted again with a new set.
  2. 2.
    Failing to Broadcast: Account-based chains require our Validators to have funded accounts in order to broadcast transactions. The process of generating an output for these chains also includes a process by which the broadcaster is selected. The signed transaction to be broadcasted should be saved on the State Chain. Failure by a Validator to submit this signed transaction indicates a failure to broadcast the transaction to the external chain. This will require a new proposer to be selected by the State Chain, incurring a slight delay. The initially-proposed broadcast Validator will then be suspended.
As Chainflip is developed, more advanced suspension conditions can be introduced to better incentivise Validators to be performant. An example would be, "consistent failure to witness' ', where a Validator does not submit a witness extrinsic for an otherwise confirmed transaction after a timeout threshold. This is considered future work, as it would be trivial to dodge penalties by simply copying witness transactions without implementing a commit-reveal scheme. We don’t consider this to be immediately necessary, but as the number of required chain clients grows, Validators may be more inclined to try and avoid maintaining so many connections, thus possibly necessitating implementing the scheme to ensure secure witnessing.
In the Suspended state:
  • The Validator will not be selected as a signer or a broadcaster,
  • The Validator is liable to be slashed, if this state is coupled with a negative reputation value,
  • The Validator is still expected to participate in witnessing and submitting heartbeats.
Validators, once the required time has passed, can have their suspended state lifted by submitting a valid heartbeat. If this does not happen by the end of the heartbeat interval, the Validator will be downgraded to offline, meaning they’ll have to wait for the next heartbeat interval before they get the chance to return to an online state.
Tuning the time-penalties and exact numbers in regards to how reputation affects slashing is important. It is impossible to know exactly why a Validator triggered a suspension condition. In otherwise performant Validators, a situation could arise where they may have just gone offline before they were called upon, and therefore a suspension occured. In this situation however, the Validator should not experience any major difference in reputation. Similarly, Validators which just submit heartbeats in an attempt to evade higher server costs will get repeatedly suspended and then slashed anyway. Finding the exact parameters to strike a balance between the different cases is an ongoing process.

Reputation Rollover

Reputation which has been accrued by a Validator should roll over into the next epoch if they make it to the next Authority Set. This allows them to drop out of the Authority Set and then re-enter in a future epoch with the same amount of reputation. It should be noted that the reputation is not taken into account during the Validator rotation process. Instead, any Validators that are offline (as defined by the State Chain), at the time of an auction being completed, will not be considered for selection in the new Authority Set.

Backup Validators

The reputation and slashing system only applies to active Validators - that is, Authority Set members and outgoing members. Backup Validators don’t have a reputation and instead are simply not rewarded when offline, and can never be slashed. Their online state is tracked using heartbeat submissions.