The part that gives me pause is the block-level access list piece, because once you record accounts and storage slots at block granularity, the validation path gets a lot more stateful than the usual per-transaction warm/cold accounting. That may be fine for throughput, but it also means the failure mode is no longer just "bad tx reverts", it is "the block's declared access set was incomplete or malformed". I was a little surprised they are coupling that with a payload timeliness committee in the same upgrade. ePBS already shifts trust assumptions around builder delivery, and adding another on-time attestation layer feels like it makes the liveness story sharper but also more fiddly for operators.

What’s the expected behavior if a nontrivial chunk of Sepolia validators misses the activation window, do we get a clean recovery path or just a long fork-choice mess while clients catch up? And for the protocol changes, what baseline are you comparing against, the last testnet fork or current mainnet client diversity, and did you run any deliberately lagging client to see where the first consensus failure actually shows up?