StarkGate (Starknet Bridge)

Rejected venues wait the longest for re-review; a rejection has to earn another look before the scheduled date.

REJECTED ON A RECENT, NOT YET RESOLVED, RELIABILITY GAP — THE NARROWEST REJECTION IN THIS BATCH. Starknet is a Stage 1 ZK rollup, confirmed directly from L2Beat: the same rating Base holds, with a comparably structured governance split between a 2-of-6 StarkWare multisig routing normal actions through an eight-day on-chain timelock and a 9-of-12 Security Council that can upgrade instantly for emergencies only, the same zero-delay-once-invoked shape already priced into Base, Arbitrum, and Optimism’s caps in this registry. What keeps this at zero is the incident record: two mainnet halts in a four-month span, September 2025 and 2026-01-05, neither losing user funds but both stopping block production outright. Roughly seven months have passed clean since the more recent one at this review, short of the twelve-month bar this registry applies elsewhere. This is the closest case to a near-term reopen in the whole bridges batch, not a structural disqualification.

Kill criteria: the events that revoke this verdict
• Twelve consecutive months with no mainnet halt or block-production outage, counted from the 2026-01-05 incident
• Any future major protocol upgrade is accompanied by a public pre-mortem or rollback plan, reducing recurrence risk
• StarkWare Multisig or Security Council composition or threshold changes are publicly disclosed within 48 hours
• Any confirmed exploit of the StarkGate bridge contracts themselves

The research file

Mechanism

Starknet is a Cairo-based ZK validity rollup. State transitions are proven with STARK proofs rather than assumed and optimistically disputed: the L1 bridge contract unlocks withdrawn funds only after verifying the STARK proof covering the corresponding L2 burn. This removes the fraud-proof challenge period entirely, and withdrawals typically clear in hours once the relevant block is proven and accepted on L1, rather than the seven days OP Stack chains require.

Control and governance

Confirmed directly from L2Beat: Starknet is rated Stage 1. Governance of the main rollup contract and its ETHBridge and STRKBridge escrows splits between a 2-of-6 StarkWare Multisig, which routes all normal actions through a DelayedExecutor timelock with an eight-day on-chain delay, and a 9-of-12 Security Council that can upgrade instantly for emergencies only. A separate 2-of-4 StarkWare Multisig 2 governs most other bridge escrows with instant upgradeability. The Operator role, which supplies state-transition proofs, is permissioned, so Starknet implements a StarknetSCMinorityMultisig letting a 3-of-12 Security Council minority enforce censorship resistance by including transactions regular Operators have not. All bridge escrows can enable a withdrawal throttle capping flow to 5% of locked funds per 24-hour period; enabling it is permissioned to a multisig, while disabling it in the core bridges requires only a 3-of-12 Security Council minority. L2Beat notes Starknet does not pass its ”walkaway test”: users cannot exit in the presence of a malicious operator if the Security Council itself disappears, a real gap Base does not carry, though a narrower one than the CRITICAL, no-delay upgrade paths that sank several other entries in this batch.

Incident record

Two confirmed mainnet halts within roughly four months: a nine-hour block-production halt in September 2025, triggered by the Grinta (v0.14.0) upgrade combined with Ethereum RPC provider failures and sequencer bugs, and a four-hour halt on 2026-01-05, caused by a proving error on a pending transaction that led engineers to pause sequencers as a precaution. No user funds were lost in either event, and no exploit of the StarkGate bridge contracts themselves was found. But two significant halts in a four-month span is a demonstrated reliability pattern none of Base, Arbitrum, or Optimism’s canonical bridges carry on their own records, and only about seven months have passed clean since the more recent incident as of this review.

Exit under stress

Normal-path withdrawals are fast, a genuine structural advantage of the validity-proof design over every optimistic bridge in this registry. But a sequencer halt — which has happened twice in the reviewed history — stops new blocks and proofs from being produced at all, so withdrawal finality is only as reliable as sequencer uptime, which has a documented, recent gap.

Comparison

Governance-wise, Starknet’s split multisig and Security Council structure with an eight-day non-emergency timelock is comparable to, and arguably somewhat more conservative than, Arbitrum’s already-approved structure, and meaningfully better disclosed than World Chain’s or Robinhood Chain’s verified weaknesses elsewhere in this batch. This is not a Linea- or Mantle-style ”None” exit-window case. The decisive difference from the three approved peers is purely the incident record: a real, recent reliability gap none of Base, Arbitrum, or Optimism have shown, not a governance-structure failure.

Sources

The claims above trace to these. Where a number could not be independently verified, the thesis says so.

• L2Beat — Starknet · primary · accessed 2026-08-17
  Supports: Stage 1 rating, 2-of-6 multisig plus 8-day timelock, 9-of-12 Security Council emergency path, fails walkaway test, withdrawal throttle mechanics
• Starknet — official incident report, January 5, 2026 · primary · accessed 2026-08-17
  Supports: January 2026 halt, proving error root cause
• The Block — Starknet outage marks second major disruption since Grinta rollout · secondary · accessed 2026-08-17
  Supports: links January halt to September 2025 Grinta incident
• Starknet Documentation — StarkGate · primary · accessed 2026-08-17
  Supports: bridge mechanism, validity-proof-gated withdrawal
• CoinDesk — Starknet goes down for over 4 hours · secondary · accessed 2026-08-17
  Supports: January 2026 outage timeline

Inherited controls

The verdict above grades the protocol layer. Every position also inherits the asset it holds and the chain it settles on. The least safe layer sets the position’s grade, and the position table names which one that is.