Final in one block: 15-second slots and the two-phase BFT vote
Fourth article of the series. What happens inside one 15-second slot, why a block is final once it is committed, and what the network does when too many validators are silent.
The last article took a detour through the first token on mainnet. Back to the protocol. This one is about the question an exchange or a shop asks first: when is a payment really done?
On many chains the answer is a number of confirmations. You wait for six blocks, or twelve, or thirty, because a block can still be replaced by a longer competing chain. On Janzeer the answer is one block. Once a transaction is in a committed block, it is final. There are no confirmation counts and no reorganisations. Here is how that works, and what it costs.
Time is cut into slots
Every 15 seconds is one slot: a 12-second window to produce and vote, then a 3-second pause. Each slot has exactly one owner, a validator chosen round-robin from the active set. Every node works out the owner from the wall clock and the start time of the current epoch, so nobody has to announce whose turn it is.
That is also why clocks matter. A node whose clock drifts more than 250 milliseconds from NTP time stops producing blocks. It keeps following the chain, but it does not take its turn until its clock is right again.
If the owner of a slot is offline, the slot passes empty and the next owner continues. A validator that misses its slots is removed from the set at the next epoch boundary, and it can come back later.
Two rounds of votes
The slot owner builds a block from pending transactions and sends it to the other validators. Then two rounds of voting follow:
- PREPARE. Each validator checks the block and signs a PREPARE vote for it.
- COMMIT. When a validator has seen enough PREPARE votes, it re-executes every transaction in the block, checks the resulting state against the block's merkle roots, and signs a COMMIT vote.
"Enough" is a quorum: more than two thirds of the active validators. With N validators the quorum is ⌊2N/3⌋ + 1. Today there are seven validators, so the quorum is five. When five COMMIT votes exist, the block is final, and every node adds it to its chain.
Votes are signed messages, and at first every validator sends every vote to every other one. When a round reaches its quorum, the votes are bundled into one quorum certificate and that single message is passed on. This keeps traffic small on home connections and Raspberry Pis.
Why one block is enough
Two facts make a committed block permanent.
First, any two quorums overlap. Five of seven and another five of seven always share at least three validators. For two different blocks to be committed at the same height, those shared validators would have had to vote for both of them.
Second, voting twice is provable. A validator may cast one vote per slot per round. If it signs two different votes for the same slot, both signed messages together are a proof, and anyone can submit that proof to the chain. A validator proven twice is banned for good, and the reporter receives 5 JNZ from the treasury.
So a second block at the same height would need at least three validators to cheat, and each of them would leave signed evidence. With seven validators the network tolerates two faulty or malicious ones. The general rule is f faulty validators out of 3f + 1.
On top of that, every node that adds a block re-executes it. A node does not trust the producer, the votes, or the peer it syncs from. If a block's state does not match what the transactions produce, the node refuses the block.
The price: when too many are silent, the chain stops
The same rule works the other way. If fewer than five of the seven validators are online and voting, no block reaches a quorum, and no block is produced. The chain does not split into two histories; it waits. When enough validators return, block production resumes from the last final block.
This is a deliberate choice of safety over liveness. A payment can be delayed, but it can never be undone. It is also not theoretical. In late October the network stopped twice for about four hours: once because too many validators were offline at the same time, and once because of a bug in how votes were counted for a slot. Both are described, with the fixes, in the incident reports and in a longer post-mortem on dev.to. The lesson was not to weaken the quorum. It was to add more independent validators and to make silent failures visible. A fifth project validator was added the same week.
What it means if you build on it
For anyone who accepts JNZ, the rules are short:
- Credit a transaction when its status is
FINALand its receipt is successful. Never credit aPENDINGone. - There is nothing more to wait for. A committed block is not rolled back.
janzeer_waitForFinalityin the JSON-RPC API waits for exactly that moment. Each of the five SDKs wraps it in one call.
The integration guide has the full checklist for an exchange or a payment service.
In numbers, as of today: block 247,358, epoch 15, seven validators, one new block every 15 seconds. Everything in this article can be checked against the live explorer.
Next: one native binary, and what it takes to run a Layer-1 node on a 2 GB VPS and on a Raspberry Pi 3.
Janzeer is built with AI assistance (Claude Code); every number here is from the live chain and the protocol facts in the public documentation, and the text is mine. JNZ is the network's unit, it is not listed anywhere, and nothing in this series is financial advice.