Self-audit
Nobody outside has audited Vettai. Here is exactly what the review was, what it found, and what was done about it.
Say it plainly: Vettai is self-audited. No outside firm, no bug bounty, no third party has looked at this code. The threat model says the same thing in its last line, and it is worth repeating here rather than burying it.
What exists instead is a review we ran on ourselves before the backend gate closed, and one command anybody can run to check the parts that can be checked.
The review
The whole backend diff went through a structured code review before any frontend work started.
| Review angles | 8 |
| Independent verifiers | 5 |
| Findings confirmed and fixed | 13 |
The angles covered the money path, the world and socket layer, the quest engine, race conditions, configuration and boot refusals, the proof command itself, and code hygiene. The verifiers each re-checked findings against the code rather than against the report.
The fixes that came out of it are the parts of the design the docs now describe as normal:
advisory locks on the three caps, held claims released on a minute timer instead of lost,
giving up on a stuck sending row after three attempts, filtering the landlord stake read to
one UTC day, judging shop order expiry against block time instead of wall clock, writing
every inspected payment to received_payments, trusted-peer proxy configuration instead of a
hop count, connection ids so a stale socket cannot evict a live one, an ordered chain for tick
events, one transaction for the claim and the quest, the paid-order sweep, and clamping the
streak reward at creation.
What is actually proven
Thirteen fixes are a story. The prove-it run is the evidence: twelve checks against the live Nimiq chain, with real payouts and real refusals, and the output saved in the repository.
The test suite sits behind that, including property tests over the simulation that hold the rules under generated input rather than over chosen examples.
What is honestly not proven
- Nothing here has been reviewed by anybody outside the project.
- The advisory locks are proven by the statements they issue, not by a real race on a busy Postgres.
- A scripted client that sends exactly what a phone sends cannot be detected. The caps bound what it can take, and that is all they do.
- The treasury key is a single hot key on the host. There is no hardware signer and no multisig, which is why the wallet is funded with what the game needs and no more.
The full list is on What we did not fix, quoted from the threat model without edits.