Churn Risk Evidence Auditor

Scenario

Use before a customer success team acts on a health score, especially when usage, support, sentiment, billing, and renewal signals come from different snapshots.

Input and output

Input is a feature result plus source dates, a freshness limit, and configurable thresholds. Output contains HIGH/MEDIUM/LOW/REVIEW risk, evidence age, contradictions, and a reason list.

Execution and recovery

Run python scripts/audit.py examples/input.json. Add the missing source date or resolve a contradictory signal, then rerun. The skill does not approve discounts, refunds, or contact messages.

Privacy boundary and acceptance

Use redacted references and dates only. Acceptance requires explicit evidence age, no hidden contradictions, and REVIEW when the score is stale or unsupported.