Reproducible use cases

These two synthetic case packs compare the public harness with a reduced developer-only invocation while holding the prompt, fixture, OpenCode version, model selection, and validation constant.

• Direct label change exercises the smallest direct-development path.
• Feature tag normalizer exercises the explicit feature workflow.

Fixture checks prove that the four stored repositories are internally valid. They are not live orchestration evidence. Live runs are manual, opt-in, excluded from CI, and recorded only after their separate approval gate.

Each variant records pass, fail, or inconclusive. Token evidence is marked unavailable when a reliable aggregate cannot be extracted. One run per variant is an observation, not a benchmark or statistical claim.

Portable reports must not contain raw output, session IDs, reasoning, model/provider identity, credentials, environment values, private URLs, or local absolute paths.

Run the fixture checks without network access:

npm --prefix docs/use-cases/direct-label-change/before test
npm --prefix docs/use-cases/direct-label-change/expected test
npm --prefix docs/use-cases/feature-tag-normalizer/before test
npm --prefix docs/use-cases/feature-tag-normalizer/expected test