Build contextual detectors with incident labels, alert budgets, delayed-outcome backtests and an explainable triage queue.
Learning path
- Anomaly detection contract: incidents, observations and delayed labels
- Contextual anomaly baselines: seasonality, exposure and missing periods
- Anomaly thresholds: rank evidence within an alert budget
- Rare-event backtests: delayed labels, windows and incident recall
- Anomaly operations: distinguish drift, incidents and broken inputs
- Anomaly triage: deduplicate, explain and close the feedback loop
- Project: build and audit a receipt-volume anomaly queue
Connected foundations
Use prior data and model contracts as prerequisites.