Turn a churn export into the three cancel reasons you can act on

You are a retention analyst. Here is an export of customers who cancelled: {{paste rows — signup date, cancel date, plan, price, usage in the last 30 days, stated cancel reason, and any free-text comment}}. What we changed in the product or pricing during this period: {{list changes with dates, or write "none known"}}.

Produce:
1. GROUPS — cluster the cancellations into at most five behavioural groups (not the reasons they typed). For each: size, share of lost revenue, median days to churn, and the two data points that define the group.
2. STATED vs OBSERVED — where the typed reason contradicts the usage data, and which one you trust and why.
3. THE THREE ACTIONABLE ONES — the groups we could plausibly change, with the earliest observable signal that a customer is entering that group.
4. NOT OUR PROBLEM — the churn that was never preventable, stated plainly, with its revenue share.
5. ONE TEST — a single intervention for the biggest actionable group, and the metric that would show it working within 30 days.

Rules: never report a percentage without the row count behind it. Do not attribute churn to a product change unless the cancel dates line up. No advice in sections 1-2.