Reviewed: 2026-07-10
Scope: GTFS Scorecard website, pipeline, public API, GitHub Action, alerts, and
optional serverless infrastructure. The product performs deterministic validation and
scoring; it does not use model inference or make eligibility decisions.
| Review | Artifact | Result | Recheck |
|---|---|---|---|
| Consequence scan | audits/consequence-scan.md |
Proceed with named safeguards | Methodology or ranking change |
| Bias and disparate-impact review | audits/bias-review.md |
Proceed; comparisons remain bounded and optional RT stays neutral | Rubric/data-source change |
| DPIA-lite | audits/dpia-lite.md |
Low privacy risk; minimize subscriptions and logs | New personal-data field or processor |
| Threat model | audits/threat-model.md |
Key abuse paths mitigated and CI-gated | New write API, auth, or infrastructure |
- No sensitive rider data, trip histories, precise user locations, or protected-class attributes are collected.
- Public agency data is scored, not people. Grades are not service-quality, compliance, procurement, or staff-performance certifications.
- Missing realtime is neutral. Absolute rankings and individual percentiles are not published. Aggregate scores and named changes require the current rubric, measured schedule categories, and an unambiguous canonical feed identity.
- Feed corrections and agency claims require evidence and human review. Opening a request never verifies control or changes the registry automatically.
- Alert subscriptions use double opt-in, store only the minimum delivery information, and support removal.
- The accessibility target is WCAG 2.2 AAA. Automated evidence is merge-blocking; human assistive-technology evidence remains a maintained review artifact, not something automation can fabricate.
These are engineering assessments, not legal opinions or certifications.