If bandwidth costs are material, rational operators won’t bother syncing it at all (opting instead for a centralized third party). Large transaction sizes are a strong argument for PBS, with access to many mempools being a more important specialized activity.
And censorship issues (otherwise just use L1-sequencing). Without PBS inclusion-lists, forcing trasactions is quite expensive (L1 proof gas).
Yes. That proposal uses simple leader election because Fernet is currently broken.
Let me explain the bribery attack:
A malicious party deploys BribeContract to L1. This contract reads L2, and pays out to all provers for a block whenever a proof is missed.
I claim that as long as BribeContract holds prover-block-reward, rational provers will not prove the current block.
This is because prover-block-reward (bribe) > prover-block-reward - compute costs (honest).
I see. This allows sequencers to halt the chain for proof-time (e.g. 10 min), at the cost of a single block (see bribery attack).
See bond pricing for my suggested level of economic security.