Incident log¶
One line per incident, newest first. Each line links to a full postmortem in
docs/incidents/<YYYY-MM-DD>-<slug>.md. Format:Severity scale matches
POSTMORTEM_TEMPLATE.md:
- S0 — silent data corruption (highest; tool answered with the wrong data without the agent or user noticing).
- S1 — tool returned a wrong answer that the agent could detect by reading our own provenance / error envelope.
- S2 — tool failed loudly; agent retried or surfaced an error; no incorrect data delivered.
- S3 — cosmetic / log-only / non-impacting.
Open¶
None.
Closed¶
None yet. This project's incident log starts at v0.1.0 and will accumulate entries as upstream UniProt schema changes are observed by the nightly integration.yml workflow. The absence of entries here does not imply the project has had no failures during development — bugs caught before a release are tracked through CI / commits / AUDIT.md. This file specifically records post-release incidents that affected, or could have affected, users.
Compliance officer view¶
A reader auditing this project in 2030 should expect to find this file populated rather than empty. An empty incident log is one of three things: (a) the project is too young to have failed yet, (b) the project does not detect its own failures, or (c) the project hides them. Topologica Bio commits to (a) being temporary and (b)/(c) being detectable: every release of uniprot-mcp carries a nightly live-API drift workflow that auto-files integration-drift issues, and the fix-PR-with-postmortem policy below is enforced through review.