Scenarios

Runnable, end-to-end examples that map to real adoption questions.

Scenario	When to read
A simple end-to-end suite	First real example — the bundled quickstart, four cases.
Large suites & multi-file agents	Outgrowing one suite file; per-domain organization.
Edge cases	Empty output, exceptions, missing baselines, exotic inputs.
CI/CD integration	GitHub Actions, GitLab, CircleCI, Buildkite.
OpenAI / Anthropic SDK adapters	Skip manual instrumentation when on a supported SDK.
Performance & cost budgets	cost_lt_usd, latency_lt_ms, drift detection.
Debugging workflow	A failing case → root cause in five minutes.
Failure handling	Exception paths, judge unavailability, baseline corruption.

Every scenario follows the same five-section shape: problem → input → code → output → explanation. Copy-paste any of them as a starting point.