Reduce finalization time

[kpp]

Right now we wait for finalized SequencerCommitments to generate a BatchProof and for a finalized BatchProof to generate a LightClientProof.

Because of that we use DA as a messaging system, not as a Data Availability layer. This leads to delays in how our subsystems communicate. Prover waits for a SequencerCommitment on DA before it can start proving BatchProof. But we don't need to do that!

What if Sequencer sends SequencerCommitments both on DA and directly to a Prover? Then Prover can start proving without waiting for DA to finalize.

In order to achieve that we have to split:

• BatchProof into StateDiff proof verification + inclusion + completeness
• LightClientProof into BatchProof verification + FinalizationProof

So the interaction of Sequencer-DA-Prover will be different:

1. Sequencer sends SeqComms to DA and to Prover
2. Prover receives SeqComms from Sequencer -> if SeqComms L2 range is big enough or state diff is big enough it starts proving StateDiff proof rightaway.
3. Prover receives SeqComms from DA -> wait for StateDiff proof to be ready, construct a BatchProof - which is StateDiff proof verification + inclusion + completeness of SeqComms on DA.
4. Prover sends BatchProof on DA.

But we have to deal with reorgs. If reorg happens:

1. We find which SeqComms were excluded (scan up to 3 blocks).
2. Send them again on DA.
3. Reprove BatchProofs - with new SeqComms tx hashes. But that will be cheaper because we already have StateDiff proofs.
4. Send new BatchProofs on DA.

UPD 1

What if we don't put SeqComs on DA? We may include them inside BatchProof so we don't need to create inclusion+completeness proof for SeqComms at all.

But we may even go further and remove SequencerCommitments from BatchProof, because BatchProof is the new source of trust for the new state no matter what.

[rakanalh]

I believe we are going to work on P2P communication layer between nodes, this can be how we enable communication between sequencer & prover. But i guess this solution is very plausible.