Evaluate sequencer decentralisation tradeoffs
"Decentralised sequencer" covers four very different arrangements, and a due-diligence answer is only useful if it names which one applies and what the operator can still do to a user's transaction.
Procedure
Classify the ordering model. The families are: (a) single centralised sequencer with a known operator key; (b) leader-elected / rotating set (e.g. multiple sequencers with a consensus); (c) shared sequencer networks where a third-party set orders for many rollups; (d) based sequencing where the L1 proposer orders, inheriting L1 censorship resistance; (e) forced-inclusion only, where the sequencer orders but anyone can bypass it after a timeout.
For each model, answer the four trust questions explicitly: who can reorder, who can censor, who can halt liveness, and who can upgrade the bridge/verifier contracts.
Read the upgrade authority on L1. A decentralised sequencer behind a single-key proxy admin is still a single point of control. Inspect the proxy and its admin:
cast call $PROXY "admin()(address)" --rpc-url $L1RPC # EIP-1967 cast storage $PROXY 0x360894a13ba1a3210667c828492db98dca3e2076cc3735a920a3ca505d382bbcfOr read the admin slot directly and check whether it is a timelocked Safe or an EOA.
Check the escape hatch and its bound. Decentralisation is only real if a censored user can force inclusion or exit through L1 within a bounded, published window. Record that window (OP withdrawal 7 days; Arbitrum delayed inbox 24 h to forceInclusion; zk priority queue).
Verify the prover/validator set independently of the sequencer. A zk rollup with a single prover is not trustless even with a decentralised sequencer; an optimistic rollup with one bonded proposer relies on the fraud-proof game being live and incentivised.
Summarise as a matrix with one row per property and the evidence (contract address, slot value, doc URL, on-chain parameter) behind each cell. Mark any property you could not verify as unknown, not as satisfied.
Pitfalls
- Reading the marketing label ("shared sequencer") and not the contracts; the actual ordering contract is what binds a user.
- Treating a sequencer-set multisig as decentralisation when the set is co-located or upgradeable by one party.
- Ignoring upgrade keys: an admin that can swap the verifier can retroactively change finality.
- Confusing DA layer decentralisation with sequencer decentralisation; they are separate risks.
- Assuming the escape hatch works at scale — a forced-inclusion path with a long window and high L1 gas is not an escape for a retail user mid-attack.
Verification
cast call $PROXY "admin()(address)" --rpc-url $L1RPC
cast call $L2_OUTPUT_ORACLE "CHALLENGE_PERIOD()(uint256)" --rpc-url $L1RPC 2>/dev/null || \
cast call $ROLLUP "confirmPeriodBlocks()(uint64)" --rpc-url $L1RPC
Report the ordering-model classification, the upgrade-authority address and whether it is timelocked, and the forced-inclusion/withdrawal window — each cell backed by the call that produced it.