Skip to content

Final in seconds

Live on testnet Simplifies the whitepaper, Principle 6

Once a block is certified, nothing inside Sequentia can replace it. Only Bitcoin can undo it, by undoing the Bitcoin block it points to.

Final, modulo Bitcoin

On Bitcoin, a payment becomes safer with every block built on top of it. On Sequentia a block does not get gradually safer. It is certified, and from then on it is final against every competitor inside Sequentia: no later block, no block with more signatures, no longer chain can displace it.

The one exception is deliberate. Every block names a Bitcoin block, and if Bitcoin reorganises that block away, Sequentia undoes its own. The whitepaper calls this finality modulo Bitcoin.

Put together: Sequentia’s history is exactly as final as Bitcoin’s, and it gets there in about ten seconds.

Who certifies a block

Each block has one proposer, chosen by the lottery, and a committee that checks and signs it.

  • The committee has up to 250 seats, shared among stakers in proportion to stake.
  • A block is certified when holders of a majority of seats have signed: 126 of 250.
  • The signatures are combined into one. The proof that a block is certified is about 300 bytes, whatever the committee’s size.

Any two majorities of the same committee must overlap, so two different blocks at the same height cannot both be certified by honest members.

Whitepaper 6.2Why a public committee of fixed size

The 2022 paper imagined a private committee: a staker belonged to it if its own secret random value fell below a threshold. The size then varied from block to block. With a target of 100 members it ranged from about 62 to 141, and on about 40% of blocks it exceeded 102, the size at which two entirely separate 51-member quorums fit. Two blocks could each be certified at the same height by the same eligible set, splitting the network into two histories, each immediately final, with no rule to choose between them.

The fixes were unattractive: a quorum of 83%, which stalls as soon as a third of members are offline, or committees of several thousand, as in Algorand.

Sequentia instead made membership public and fixed in size. The committee for a height is the first K entries of a public schedule derived from the seed, with K = min(registered stakers, 250). Every node agrees on who the members are. In force since July 2026.

If the set has K members and the quorum is Q, any two quorums share at least 2Q − K members. Requiring K ≤ (Q − 1) × 2 guarantees an overlap of at least two. So the quorum is a strict majority, plus one when K is odd: 126 of 250, 21 of 40, 1 of 1.

Only membership became public. The proposer’s identity stays secret until its block appears.

Why 250

A committee is a sample of the stakers. Too small and it is easy to stall or capture. Too large and it is slow. Three things were measured to choose the size.

How often the chain would stall

A block stalls when so many committee members are offline that the rest cannot reach a majority. The table gives the average time between such stalls.

Stakers offline Committee of 100 Committee of 250
20% 117,600 years effectively never
30% 35 days 117,700 years
35% 12 hours 3 years
40% 38 minutes 22 hours

With 35% of all stakers offline, a committee of 100 stalls twice a day. A committee of 250 stalls once in three years. And a stall is not fatal: see Never stuck.

How long a hostile group would wait to capture a block

A group captures a block when it holds a majority of that block’s committee by itself.

Hostile share of all stake Committee of 100 Committee of 250
25% 107 years effectively never
30% 85 days 286,800 years
33% 5 days 294 years
40% 1 hour 1 day

The classical safety bound for this kind of system is one third. A committee of 250 is the smallest that meets it with margin: a group holding a third of all stake would wait nearly three hundred years for its first captured block.

What a bigger committee would cost

Beyond 250, safety barely improves, while every extra member adds time to each round. A committee of 500 would take 13 seconds to sign instead of under 7, and each failed attempt would cost 18 seconds instead of 9.5.

Whitepaper 6.3The full tables, and what they assume

All figures use exact hypergeometric arithmetic, not simulation, and assume ten thousand stakers, the ceiling the minimum stake imposes.

Average time between stalled blocks, by share of stakers offline:

Offline 50 100 150 200 250
25% 6 d 34 yr 67,600 yr very large very large
30% 7 h 35 d 10 yr 1,081 yr 117,700 yr
35% 49 min 12 h 7 d 90 d 3 yr
40% 10 min 38 min 2 h 7 h 22 h
45% 4 min 6 min 8 min 12 min 16 min

Average time before a hostile coalition captures a block, by its share of stake:

Stake 50 100 150 200 250
25% 19 d 107 yr 212,500 yr very large very large
30% 18 h 85 d 25 yr 2,635 yr 286,800 yr
33% 4 h 5 d 139 d 11 yr 294 yr
40% 18 min 1 h 3 h 11 h 1 d
45% 5 min 7 min 11 min 15 min 21 min

Above 40% no practical committee defends by sampling at all. A larger population with the same hostile percentage makes capture easier, because the sample is drawn from a deeper pool.

Because the committee is public, a coalition does not need luck: it can see in advance a height it captures and act then. The tables are therefore expected waiting times before the first opportunity.

Capture lets an attacker replace the proposer, take the fees and censor transactions for that block, or certify two blocks at one height. The second case is handled by the observation window described in Never stuck.

Data notes: committee-regenesis-parameters.md.

What happens in the ten seconds

  1. The lottery winner assembles a block and broadcasts it, unsigned by the committee.
  2. For a short collection window (6.75 seconds with a full committee), members gather the proposals they receive. Usually there is one.
  3. Each member picks the best one: anchored to the most recent Bitcoin block, then the lowest lottery number. It checks the block against every rule and signs.
  4. Members sign independently, without talking to each other. Any node that has collected a majority of signatures combines them into a certificate.
  5. The node waits a further three seconds to be sure no competing certificate exists at that height. Then the block is final.

About seven seconds to certify, plus three to observe: roughly ten seconds from proposal to finality. The next block comes at least sixty seconds after the last.

If the chosen proposal fails to gather a majority, the committee moves to the next-ranked proposal in a new round of 9.45 seconds. An absent or dishonest proposer costs one round.

Whitepaper 6.4, 6.5The round in detail
  • Every staker publishes a verification key for the lottery and, to be eligible for the committee, an aggregation key with a proof that they hold it.
  • When a block is certified, its Bitcoin anchor and height derive the public seed for the next.
  • From the seed, every node derives the committee: K seats apportioned by stake.
  • The proposer embeds its proof of eligibility in the block.
  • The certificate lives in a part of the block excluded from the hash the committee signs, so the hash does not depend on who signs. This is what lets members sign independently and any node aggregate any majority.
  • The signature scheme (BLS) aggregates without interaction: no exchange of nonces to organise.

The durations are node settings, not consensus rules, and all nodes use the same defaults: collection window of 500 ms plus 25 ms per member, round of 700 ms plus 35 ms per member. A fixed 500 ms window stalled a 100-member committee in tests.

Cost of verifying. Checking each incoming signature share separately takes about 2.3 ms on an ordinary laptop: 0.6 seconds for 250, or 2.1 seconds on a machine limited to a quarter of one processor core. The finished certificate is verified in a single operation of about 35 ms.

Size. A certificate is the proposer’s signature, one 96-byte aggregated signature, one bit per member (32 bytes at 250) and the signers’ seat total. An empty certified block on the testnet is under 700 bytes.

Clocks. In a test where members’ clocks were shifted against each other by up to four rounds, two certified blocks at one height never occurred. Clock disagreement costs speed, not safety.

A larger committee does not certify more slowly: a block is certified at the speed of its fastest majority, not of its slowest member.

What this means for you

  • A payment in a certified block is done. A shop, an exchange or a bridge does not need to wait for more Sequentia blocks.
  • The remaining risk is Bitcoin’s. A service that would wait three confirmations for a bitcoin payment can apply the same policy to Sequentia, by looking at how deep the block’s anchor is in Bitcoin.
  • Blocks certified with less than a majority are different. They exist only when the chain is escaping a stall, and they are never final on their own. See Never stuck.

Read further