I see. So stalling costs min(TOTAL_COMMITEE_REWARD * 1/3, SEQUENCER_BLOCK_REWARDS * 13) per epoch.
SEQUENCER_BLOCK_REWARDS * 13 can also grief the previous epoch for TOTAL_COMMITTEE_STAKE + TOTAL_SEQUENCER_STAKE + EPOCH_REWARDS.
Forced inclusion seems expensive. Will it support batching?
I meant high-latency-compute (rather than decentralized), which differentiates commodity from specialized hardware. This prevents a monopoly by the fastest-chip-producer, which would be an easy target for censorship. MIN_PROVER_TIME does not apply here because sequencing is separated. Again, proving is already centralized because one operator will have the lowest overhead.
I claim that token-holders will pay an efficient price for proofs under proof-racing. Racing can be modeled as an all-pay auction, which is theoretically optimal for the seller (any “wasted compute” is paid by the bidders).
To price PROVER_REWARD as efficiently would require targeting basisPointFee to 100%. This seems nontrivial.
Note Re-Based does not use proof-racing, it relies on a bonded first-price auction (on L1).