Trust, safety & governance
methods and gates that keep AI and releases auditable
Instruments that keep a rule enforced rather than declared. ship-panel is a reusable release-review panel: six independent single-axis judges plus a synthesis pass that sorts findings into what blocks a ship versus what only advises. It runs as the review stage of the ship-prep gate over a repository, a bundle, or a single file alike; the governance-patterns case study covers its decomposition and reuse.
Around the panel sit the release gates wired into every public repo — branch protection and required status checks, with cosign keyless signing, an SBOM, and SLSA provenance riding along at the public tier — so 'is this safe to ship' is a gate, not a judgment call.
The third instrument is an enforcement root-cause method from production trust-and-safety work: trace a misclassification to its cause, correct it, and close the loop so the same class of error is caught next time. The enforcement-RCA case study describes the method — employer-agnostic, specifics kept general.
Governance patterns — instruments that keep a rule enforced
An adversarial multi-judge review panel and an access control paired with a reconciliation loop — two shapes for turning a declared rule into an enforced one.
Enforcement RCA — a root-cause program for a live ML model
A repeatable root-cause-analysis program for a live ML enforcement system. Methodology only — no platform-internal data.