Never stuck
If the committee cannot gather, Sequentia does not halt. It looks at Bitcoin, sees that time has passed, and carries on.
The price other chains pay for finality
Chains with immediate finality share a weakness. They need a quorum of signers for every block, and when too many signers are missing, there is no block. From the inside, the protocol cannot tell signers that are gone from messages that are merely slow, so it waits. Liquid and Algorand both stall when a third of their signers are absent.
Sequentia has something those chains do not: a clock outside itself. Every block already commits to a Bitcoin block, and Bitcoin’s progress is a measure of elapsed time that nobody can forge.
The escape rule
Normally a block needs signatures from a majority of the committee. The escaping-stall rule relaxes that, down to a single signer, when two things are true:
- the new block points to a Bitcoin block at least three blocks later than the one its parent points to, and
- at least one Bitcoin block interval of time, about ten minutes, has really elapsed between those two Bitcoin blocks.
Both conditions are read from data written in the blocks themselves, so every node reaches the same verdict.
In practice: if the chain stalls, it starts moving again after roughly half an hour, at a slow pace of one block per three Bitcoin blocks, until enough committee members return.
Whitepaper 8.1Why the rule also demands elapsed time
The time condition was added after an incident on the testnet on 17 July 2026.
A gap of three Bitcoin heights is not, by itself, evidence that Sequentia has stalled. Bitcoin’s test networks frequently produce bursts of very easy blocks, and three of them can arrive within seconds while Sequentia is entirely alive. On that day, exactly this let a block be produced under the escape rule thirty seconds after its parent.
A block made with fewer signatures must be hard to make, and it is only hard if the evidence behind it is Bitcoin time, which nobody can manufacture. The rule now measures the interval using Bitcoin’s own median time.
Second-class blocks
Blocks made under the escape rule are deliberately weaker than ordinary ones. The whitepaper calls them sub-quorum blocks.
- They are not final. Finality requires a majority.
- They lose to any fully certified block at the same height.
The escape rule provides liveness. It does not confer authority.
This is what keeps a network split from becoming permanent. Suppose a group of nodes is cut off and keeps the chain alive on its own with sub-quorum blocks. When it reconnects and meets a branch carrying a fully certified block, that branch wins and the isolated nodes switch to it automatically.
Whitepaper 8.2How competing blocks are ranked
A full quorum can be reached at most once per height, so a certified block normally has no rival. Sub-quorum blocks can legitimately coexist at one height. They are ordered as follows:
- a fully certified block prevails over a sub-quorum block;
- between two blocks of the same kind, the one whose proposer drew the lower lottery output prevails, and between equal outputs the one with the lower block hash.
The ranking uses the raw lottery output, not the stake-weighted score, because the raw output can be computed from the block alone without consulting the stake registry, so every node obtains the same result.
Any block that loses is discarded together with everything built on it.
Locking in finality
Each node keeps a finalized point: the most recent fully certified block that has stood on its chain for three seconds. The node rejects any block that would fork the chain at or below that point, even one that later gathers more signatures, even one pointing to a more recent Bitcoin block.
The finalized point is saved to disk and survives a restart. A final block stays final however long the chain then runs without a quorum.
Members who sign twice
Two dishonest committee members who sign two different blocks at the same height can get both certified, if the honest members happen to be split almost evenly between the two. An attacker with as little as 1% of the stake can try to provoke such a split, by releasing its own proposal at the last instant of the collection window, and retrying costs it nothing.
This is why finality waits three seconds.
Two valid certificates at one height can only exist if someone signed both, so their appearance is itself proof of cheating. A certificate is about 300 bytes and crosses the network in well under a second. If a second certificate shows up during the window, every node resolves the conflict with the same ranking rule above and keeps the same winner.
Whitepaper 8.4Why three seconds is enough
The window only has to let the certificate travel, not the whole block. In a test between a computer in Italy and a server in Germany, a certified block took a median of 43 ms to cross and never more than 95 ms.
A node that has verified a competing certificate does not rely on the clock alone. It keeps that height, and everything above it, undecided until it has received and judged the competing block, for at most 30 seconds. Normally both blocks have already reached every node as proposals.
To keep a split alive, an attacker would have to keep one certificate away from half the network for the whole window, which takes a real network partition, not good timing.
The window is a node setting, not a consensus rule. A single proposer that sends two different blocks of its own is a different, easier case: it is caught before anyone signs, and excluded for that round.
When honest nodes disagree anyway
One situation needs no dishonesty at all. Bitcoin reorganises away from a block and then, shortly after, back to it.
- Sequentia certifies branch A, anchored to Bitcoin block X.
- Bitcoin briefly replaces X. Nodes that see this undo branch A, as anchoring requires, and the committee certifies a replacement, branch B.
- Bitcoin returns to X. Branch A is valid again. But the committee is now final on B, while a node that never saw step 2 is still final on A.
Both branches are fully certified and correctly anchored. Neither anchoring nor the finality rule can choose between them.
This happened on the testnet on 17 July 2026.
The reconciliation rule resolves it. A node no longer throws away a rival certified branch that forks below its finalized point: it stores it. If that branch keeps growing with full certificates while the node’s own branch receives none for ten minutes, the node switches.
Whitepaper 8.5The exact conditions
A node moves its finalized point to the rival branch only when all of these hold:
- the rival branch carries a full quorum certificate at least three blocks above the local finalized height;
- the anchor of that certified block is on the node’s Bitcoin best chain and not above the last Bitcoin height on which every live branch agrees, so the switch never happens while Bitcoin is still forked;
- the rival branch does not fork below a Bitcoin checkpoint;
- the local branch has received no fully certified extension for a patience interval, ten minutes by default.
The node’s own blocks are not invalidated; they remain stored as valid but inactive history.
The rule creates no new class of attacker. The evidence it requires can only be produced by a majority of the committee, which already controls the chain going forward.
Incident report: incident-2026-07-17-finality-partition.md. Design: anchor-reorg-of-reorg-recovery-design.md.
What the testnet taught
These rules were not all in the first design. Three of them exist because the public testnet broke in an instructive way:
| Date | What happened | What changed |
|---|---|---|
| 17 July 2026 | A burst of easy Bitcoin testnet blocks triggered the escape rule on a healthy chain, and a double Bitcoin reorganisation left nodes final on different branches | The escape rule now requires elapsed Bitcoin time; the reconciliation rule was added |
| 1 October 2026 | A server failure took many committee nodes offline during a long Bitcoin testnet fork; the chain could neither reach a quorum nor escape | Producers now keep following a Bitcoin branch the chain is already committed to |
A sequence like the first is very unlikely on Bitcoin mainnet, where blocks cannot arrive in bursts. If it ever does occur, the rule is there.
Read further
- Final in seconds
- Bitcoin anchoring
- Whitepaper, Principle 8