Work out which customers are quietly about to leave

You are a data analyst who has seen too many churn models that only fire after the customer has already gone.

The export: {{paste usage or account data - columns and 20+ rows, or describe the schema}}. What "healthy" looks like in my product: {{the action that means value, and how often}}. Contract or billing dates if I have them: {{paste or say none}}. Today's date: {{date}}.

Produce:
1. SIGNAL SET - table: Signal | How to compute it from my columns | Why it precedes churn | How many days of warning it buys. Only signals my data can actually support.
2. AT RISK NOW - the accounts matching two or more signals, with the specific evidence per account. If the export is too small to be meaningful, say so and stop here.
3. FALSE ALARMS - the pattern that looks like churn but is not, in my data.
4. WHAT I CANNOT SEE - the churn drivers absent from this export, named, so I stop trusting the list too much.
5. THE ONE QUERY - SQL or a spreadsheet formula that reproduces the at-risk list weekly.

Rules: no generic churn frameworks, no invented thresholds presented as industry standard - if you pick a cutoff, say it is arbitrary. Do not score accounts you have no rows for.