How should block author equivocation be handled?

[zdave-parity]

What should a node do if it sees an author has authored two blocks in one slot?

[davxy]

In Polkadot, IIRC we announce this via dedicated "Equivocation" messages, and the block author may report it to the runtime.

Perhaps we could adopt a similar approach but use a slashing service with slashing capabilities (e.g., the staking service).
This might even be the same service responsible for "designating" new validators.

Presumably, this service holds some of the validators’ locked funds and should have the ability to withdraw a portion of those funds and transfer them to a treasury.
However, the exact behavior would ultimately be determined by the service itself.

The same (or similar) mechanism can be used if Grandpa equivocations (i.e. double votes) are detected.

I also guess that these critical service accumulation calls should be part of the "always_acc" sequence?

But if the question is more about the structure of the offchain protocol to announce equivocations, perhaps this can be some kind of simple gossip targeting the the current (and maybe next) epoch validators?

[zdave-parity]

I think there are a few questions here:

• Does equivocation need to be punished?
• If so, how should validators inform the relevant builders of equivocations? This could be done by having validators publish equivocations on chain. It could also be done via some off-chain mechanism. Placing them on-chain is probably simplest? They should be extremely rare so this isn't likely to introduce much overhead.
• When a validator sees an equivocation, what should be communicated to other validators? Should the equivocating blocks be distributed as normal? Should there be a special communication channel for this? Should the equivocating blocks simply be dropped? (I would guess this is a bad idea, but I'm not certain. If equivocations need to be communicated to builders then we certainly need to communicate them to other validators somehow)

All of these questions apply to GRANDPA equivocations too yeah. And possibly there are other equivocation cases we need to consider. eg A validator could produce both a positive and a negative judgment for a report. In most cases this is punishable by picking one of these judgments to include in a dispute. But the author including the dispute in a block could choose not to punish all equivocating validators. Is this acceptable?