Conformance

Verification proves properties; conformance pins behavior. state/conformance replays recorded scenarios (ordered event sequences with assertions) against a machine and reports any divergence. It is the regression net: a golden test that fails the day a refactor changes what the machine does.

A scenario is serializable data (conformance.Scenario), so you can author it, commit it, and load it back with LoadScenario. Because events are typed in Go but named in the file, you bridge the two with an EventCodec.

codec := conformance.EventCodec[Event]{
    Named:   func(e Event) string { return e.String() },
    Resolve: func(name string) (Event, bool) { e, ok := byName[name]; return e, ok },
}

res := conformance.RunAgainst(m, sc, newOrder(), codec, Open)
if !res.Passed() {
    t.Fatalf("scenario %q diverged: %+v", sc.Name, res)
}

RunAgainst casts a fresh instance, fires the scenario’s events, and records a Trace (the portable, diffable record of states entered, effects emitted, and the final state), then evaluates the scenario’s assertions.

To check that two machines agree, say a reference against a refactored subject, or a Go-authored machine against its JSON round-trip, CompareMachines runs a scenario set through both and reports field-level Mismatches as an *ErrConformance:

err := conformance.CompareMachines(reference, subject, scenarios, codec, Open, newOrder)

You don’t have to hand-write the scenarios. GenerateScenarios walks the machine’s structure to produce a set that reaches every state, and verify.CoveringSuite yields event sequences that exercise every transition. Either makes a strong starting corpus you then freeze as golden.

Conformance and verification cross-check each other: the suite that gives full coverage is the suite worth freezing here.