Skip to documentation
Documentation navigation

Documentation navigation

Documentation / guides

Cases and runs

Move from validated evaluation cases to bounded runs and durable results.

developer

The cases-and-runs section follows one evaluation from its durable definition to its recorded result. Start with a Scenario and its optional expectation, group scenarios into a revisioned Suite, then let eval.Run produce a report of observations and assessments.

Choose the contract

NeedContinue withWhat you will learn
Define input, identity, revision, and expectationsCases and suitesHow validation keeps fixtures stable before a target runs.
Execute trials with cancellation and bounded workersRuns and resultsHow target errors, evaluator errors, and quality outcomes remain distinct.
Persist a safe canonical reportReportingHow report/v1 and FileSink preserve evidence without raw payloads.

Keep the suite revision and target subject revision aligned. A mismatch is a target-stage error, so a run cannot quietly evaluate evidence under the wrong model or agent identity.

Source

The case contract is defined in scenario.go and suite.go, and the execution contract is in run.go.

Proof

Validation behavior is exercised by scenario_test.go, and run-stage behavior by run_test.go.

← back to documentation