Open source · MIT
Published so you can contradict us
An evidence link hands you snapshots without a way to recompute from them — you still have to trust that the number came from the data beside it. These packages close that gap.
@crifine/sdk54 testsTypes, the measurement method, an independent verifier, and the decision policy.
walkLadder— the whole calculation, as a pure functionverify— recompute a published estimate from its own evidencedecide— proceed / resize / hold / defer / refuse, in one place- Zero runtime dependencies
@crifine/x40223 testsPay per request in USDC, with spend limits that are not optional.
maxPerCallis required and has no default- Payments serialised by default, so parallel calls cannot both slip past a budget
- Never holds a key — settlement is a callback you supply
- Works with any x402 API, not just ours
@crifine/cli38 testsPre-trade checks from the terminal and from CI, with exit codes a pipeline can branch on.
crifine decide— the branch already taken- Exit code 2 (unmeasured) is separate from 1 (expensive)
crifine watchpolls until a threshold breaks, then exits--jsonon everything; errors to stderr
Why the method is open and the record is not
The measurement is published as code you can run. The recorded depth series is not, and will not be — it is the one thing here that cannot be reconstructed after the fact, and it is what makes the numbers worth checking in the first place.
Opening the method makes our numbers checkable. Opening the record would leave us with nothing to be checked on.
The verifier is built to say no
A verifier that can only confirm proves nothing. The test suite feeds it payloads with flattering prices, padded ladders, mislabelled sizes and hidden closed markets, and asserts that each one is rejected. Payloads that are merely uncheckable — no ladder, no method version, no stated size — are refused rather than assumed correct.