Structural
Required artifacts, schemas, bindings, and coverage are valid.
Proof and governance
Claims require typed evidence and separated authority
A run can finish only when required gates resolve, evidence matches the frozen route and source identity, independent review is present, the run packet validates, and human-only decisions have valid approval.
Proof levels
Semantic
Commissioned adapters and thresholds prove the intended behavior.
Authoritative
An independent trusted runner, provider, signer, or authority attests the result with rollback-resistant state.
Required completion controls
Typed evidence
Proof records commands, artifacts, metrics, digests, timestamps, status, and source identity. A file that only says “passed” is not sufficient.
Independent review
The builder cannot be every reviewer. Completion requires an attested review artifact from a separated reviewer identity.
Run packet
The run packet binds intake, goal, route, controls, evidence, review, approvals, unresolved items, and the final trust decision.
Human approval
Approval permits a consequential decision. It does not replace failed or missing technical proof.
Gate policy
Every new completion claim must produce valdris.run-packet.v3 binding intake, route, goal, assurance level, packet-commit catalog bytes, accepted pre-closure gate set, proof, valdris.review.v2 provenance, and applicable RCA and semantic/authoritative artifacts. Non-HEAD and v2 packets receive integrity-only structural inspection; full revalidation requires an isolated exact-commit checkout.
Every completion requires an Ed25519-attested independent review artifact using valdris.review.v2 with exactly scout, implementer, verifier, and independentReviewer. The actorId, sessionId, and executionId values are each pairwise distinct across all four roles; an authorized independent reviewer from the committed project trust store signs the frozen evidence bundle and complete role roster.