Review follow-up (HIGH): the three detectors fed the same evaluator/backtest, but
SuspensionFreeze is non-directional (favourite unchanged) — grading it as "favourite
won" polluted the hit-rate with the base favourite-win rate, and its high frozen-ness
score always cleared the backtest threshold.
- Add AnomalyKind.IsDirectional() (flip + steam = true, freeze = false).
- AnomalyOutcomeEvaluator returns Unresolved for non-directional kinds (favourites
still surfaced for display) so they don't distort calibration.
- RunBacktestUseCase skips non-directional anomalies when building candidates.
- Tests for the classification, the evaluator path, and the backtest skip.
Adds a calibration dashboard that joins persisted SuspensionFlip anomalies
with EventResult rows and reports whether the post-flip favourite actually
won — the single metric that says whether the detector is doing its job.
Domain:
- AnomalyEvidenceData + AnomalyEvidenceParser to read the JSON written by
AnomalyDetector without re-implementing the schema.
- AnomalyOutcomeEvaluator: pure function returning Hit / Miss / Unresolved.
Tennis-style two-way markets with a Draw winner are downgraded to
Unresolved rather than silently counted as Miss.
- AnomalySeverityThresholds: shared Low/Medium/High constants so the UI
badge and the report buckets cannot drift.
Application:
- EvaluateAnomalyOutcomesUseCase orchestrates the join + aggregation.
- AnomalyOutcomeReport carries totals, hit rate, three breakdowns
(severity / sport / score bins) and a per-event title lookup so the UI
needs no second pass over IEventRepository.
- Score bins extend below 0.30 automatically when the operator lowers the
detector threshold so the histogram total always equals ResolvedCount.
UI:
- Insights page at /anomalies/insights — hero header, 4-card KPI strip
(hit rate tinted by tone), three breakdown grids with bar visualisation,
drill-down tables for resolved and unresolved anomalies. Honors
prefers-reduced-motion. RU + EN localisation.
- Nav entry under Analysis section + chip button on the Anomaly Feed.
Tests: +42 across Domain + Application (evaluator boundary cases including
tennis two-way and Draw guard, score-bin edges, dynamic floor when
threshold is lowered, event-title pass-through). All 324 tests pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
alexei.dolgolyov