Closes #70 (part of the property-testing epic, issue 62).

motel check previously analysed the topology in isolation: a topology that passes all checks could still violate bounds when a scenario adds calls or retries. It now reads the scenarios: block already present in the topology YAML — per the design discussion on the issue, no new input flags are needed.

How it works

• Combination enumeration: scenario windows are swept at their start/end boundaries to find every distinct combination of co-active scenarios (active sets only change at boundaries, so this covers every combination that occurs in time). This answers the issue's open questions: rather than varying activation randomly per sample or sweeping continuously, every reachable combination is checked deterministically.
• Static analysis under overrides: worst-case depth, fan-out, and span analysis apply each combination's add_calls/remove_calls via the engine's existing effectiveCalls resolution. Exported MaxDepth/MaxFanOut/MaxSpans/SampleTraces signatures are unchanged; they delegate to override-aware internals.
• Sampled exploration under overrides: each combination is sampled with its overrides active, using a shared seed so combinations sample the same trace sequence.
• Worst-case attribution: each check reports the combination that produced the worst case (by static value, sampled maximum as tie-breaker; ties go to the baseline) in a new CheckResult.Scenarios field, rendered as a scenarios: annotation in the CLI output.
• Opt-out: --skip-scenarios restores baseline-only checking, per the issue discussion.

Example

$ motel check topology.yaml --samples 100
PASS  max-depth: 1 (limit: 10)
      path: gateway.handle → backend.process
      p50: 1  p95: 1  p99: 1  max: 1  (100 samples)
PASS  max-fan-out: 5 (limit: 100)
      worst: gateway.handle
      scenarios: retry storm
      p50: 2  p95: 2  p99: 2  max: 2  (100 samples)
PASS  max-spans: 6 static worst-case, 3 observed/100 samples (limit: 10000)
      scenarios: retry storm
      p50: 3  p95: 3  p99: 3  max: 3  (100 samples)

Testing

• Unit tests for ScenarioSets enumeration (empty, overlapping windows, disjoint dedup) and for Check attribution (scenario producing the worst case fails a tight limit and is named; baseline wins ties and call-removing scenarios)
• CLI tests covering the scenarios: output annotation and --skip-scenarios
• All existing check/property tests pass unchanged; make test and make lint green

Independent of PR 140 — branched from main.

https://claude.ai/code/session_01TgWavXRfzQiTWDZccuS2pf

Generated by Claude Code