sisodiabhumca When to invoke
- When monitoring weekly/daily KPI dashboards for sudden changes.
- When preparing an investigation checklist for metric movements.
Inputs needed
--input: Path to a CSV file with columnsdate(YYYY-MM-DD) andvalue.- Optional
--window: Rolling window size in days (default 14). - Optional
--z: Z-score threshold (default 3.0). - Optional
--out: Path to write a JSON report.
Workflow
- Parse dates and values; sort by date.
- Compute rolling mean and standard deviation.
- Flag points with [ z = \frac{x - \mu}{\sigma} ] above threshold.
- Summarize recent anomalies and basic context (last 7 days trend).
Output format
JSON with:
anomalies: list of{date, value, z, mean, std}.recent_summary: last value, 7d min/max, direction.
Guardrails
- If standard deviation is zero, do not divide by zero; skip anomaly scoring.
- Purely statistical heuristic; not a root cause analysis.
Reference code
kpi_anomaly_triage.py