Key→Spend Cap
Method key ≠ spend cap. Policy code guards the money.
–
1 Demo→89% Leak
Happy path waves 8/9 bugs through; 5 layers catch 18/18.
–
Boundary→100% Hit
Boundary schedules + verdict cache: 0% false accepts.
–
Failure→Rule PG-xx
Shrinks to minimal counterexample & blames broken rule.
–
Rust Vault→TEE Proof
458 schedules certified + Ed25519 enclave diff review.
–
Oracle pipeline & 19 candidate guards
Press “Run the loop” — 19 candidate guards (1 correct, 9 seeded bugs, 9 novel boundary drifts) vs. 5 oracle layers + content-addressed cache.
Loop results (L1 → L14)
LoopOracle layersTestsExposedFalse acceptCorpus
Layer × Defect Matrix
Private Code & Policy Review · [Diff] → [NEAR AI Cloud TEE (e2ee)] → [verifyResponse() Ed25519]
engine: checking /api/engine…
CLI & Raw JSON Evidence · SHA-256 ledger ready (0 entries)
npm test · npm run demo · node bin/proofgate.mjs incremental · scripts/near-conformance.sh → CERTIFIED (458 schedules)
No loops yet.