A verdict is not authority.
This first public release turns six agent-assurance failure modes into a zero-signup blind challenge and explorable deterministic lab.
- forged derived verdict
- exact-action parameter swap
- one-use nonce replay
- correlated evaluator quorum
- evidence deletion
- silent monitoring
Verification: production build passed; 14 tests passed; hostile fixtures block or revoke; every packet fixes execution_authorized to false.
Live lab: https://vrtxomega.github.io/veritas-agent-trust-lab/
The VERITAS product gate remains INCONCLUSIVE because evidence is author-side until independent participants contribute blind labels or other review. This is a reference demonstrator, not certification, an independent audit, factual truth from signatures, or execution authority.
A bounded $750 founding pilot is offered for one agent workflow and up to five consequential operations.