Skip to content

Decision:

  • Status: Proposed (then Accepted / Rejected / Superseded)
  • Date: YYYY-MM-DD
  • Related issue/PR: link
  • Owners / participants: relevant maintainers and contributors

Context and question

What problem are we trying to solve? Which current behaviour or methodological assumption is at issue? Give a reproducible public/synthetic example.

Options considered

Describe the current approach and credible alternatives. Note the strengths, costs, uncertainty and who might be affected.

Evidence and limitations

Link dated sources, exact model/evaluation variants, experiments and counterexamples. Name what is unknown or disputed. Do not include private user data or raw prompts.

Proposed decision

What would change in code, data model, evidence contracts, ranking, documentation or workflow? Be explicit about what will not change.

Consequences and compatibility

How does this affect historical data, public exports, privacy, licences, generated registries, cost, performance or dependent projects?

Validation and rollout

How will we test the change (including golden evaluations and migration tests where relevant)? What is the rollback or forward-fix path?

Outcome and follow-up

Record the final decision, date and reasoning after review. A proposed ADR is not approval to make production changes. Link superseding records if revisited.