It reads your rules and shows you every case they silently fail to handle. Exhaustively. With a receipt anyone can re-check.
And once your decision space is proven solid, a cheaper model does the work just as well — prove once, then slash your token budget. See the benchmarks →
Paste a policy. Watch it find the case you never decided — the one where the bug ships.
check_completeness(refund_policy) complete: false · 3 situations your rules never decide: ▸ within window · item opened -> undetermined ▸ after window · enterprise tier -> undetermined (rule A and rule B collide) ▸ already refunded · payment disputed -> undetermined → the case you didn't decide is exactly where someone guesses in production.
No sampling, no test suite — it decides over the whole input space and names the exact gaps and contradictions. Then it seals a signed receipt an auditor re-verifies with the vendor absent.
The VEX Analyst.
Kill the 90% of scanner findings that aren't reachable in your deployment — with a proof, not a heuristic — and emit an OpenVEX receipt your customers re-check offline. Ready for the EU Cyber Resilience Act.
Start here →The requirement engineering will silently guess — and the two that contradict.
A failing case you paste straight into a regression test.
The request your policy never decided, and the rules that both grant and deny.
No fail-open hole, no contradiction, re-checkable air-gapped.
Every allow/deny a receipt; the undecided case routes to a human.
The combination your table forgot — proven, not eyeballed.
The profile where a sicker patient is denied and a healthier one qualifies.
The situation a covenant doesn't determine — where the dispute lives.
One row per untested situation — your missing-test matrix, proven.
The seam where team B assumes a value team A never guarantees.
No engine, no network — a public key and the receipt.
COLREGs collision-avoidance, patch source-contracts, pricing tiers, refund rules.
Every result seals a signed, offline-verifiable receipt. A skeptic — an auditor, a regulator, a customer — re-checks it with only a public key. No engine, no network, no trusting you. That's the difference between "we say it's complete" and "here, prove it yourself."
Get a free week → Read the guide →