I suggest slashing all sequencer+provers if a proof is missed.
This way, the cost to halt-bribe is sequencer stake, rather than sequencer-reward/prover-reward, and it improves cooperation incentives.
Remaining concerns:
- The threat of data withholding by sequencers/provers
- even if unfavored in equilibrium, it may cause centralization pressure to distribute the risk
- A maximum halt-bribe of S is considerably lower than PBS (
S * block-rate * miss-frecency
) - The security is S, but the slash is
S + S * N
- This is inefficient, and gives a multiplier of N advantage to economic adversaries (burn S to cause S*N harm)
- High operating cost centralization effects on sequencers
- This would be solved if Fernet adopts L1 sequencing