AI can ship code.
But should it?
Between “the tests passed” and “production is live” lies one critical decision. EvidenceGate makes that decision visible, repeatable, and human-controlled.
Independent open-source prototype. Interactive scenarios use synthetic data. No live deployment is initiated.
More autonomy.
Less ambiguity.
One readable chain from code changes to a release decision. Each stage has an explicit control boundary; stage names indicate architectural coverage, not proof of a live execution.
Break the state.
Watch the gate hold.
Select a situation, inspect the plan fingerprint, and approve the exact planned state. No network requests or real releases occur during the exercise.
Release parameters
Release verdict
Calculating...Local illustrative approval only. This interface does not authorize a CI/CD release, authenticate a user, or attest an external pipeline.
Don't trust the badge.
Inspect the evidence.
A build-generated source manifest records reproducible file digests and optional GitLab CI identifiers. Synthetic tests are marked as synthetic; external signatures and live deployment are never implied.
Immutable plan digest, time-limited consent, drift rejection, policy checks and a machine-readable decision receipt.
GitLab Duo custom flows docs ↗This demo uses synthetic release metadata. GitLab Duo flow execution and GitLab CI delivery must be independently verified in the public project.
Competition requirements ↗