Sanity-check a number before I believe it

A result jumped 40% and you want to rule out your own query first.

I got a surprising result. Before I go and tell anyone, help me rule out that it is an artefact.

Work through in this order, cheapest checks first:
1. **The query**: a join multiplying rows, a changed filter, a date boundary, a timezone shift, a NOT IN with NULLs.
2. **The data**: a backfill, a duplicate load, a late-arriving batch, a schema or enum change upstream, a source system deploy.
3. **The population**: did the denominator change rather than the numerator? Did a large customer arrive or leave?
4. **The definition**: did anyone change how this is calculated, including in a BI tool layer?
5. **The world**: only after the above, consider that the change is real.

For each candidate, give me the specific query or check that confirms or eliminates it. Rank by how often each one turns out to be the answer. Do not offer an explanation of the business cause until the artefacts are eliminated.

What I measured and what I expected:
Query:

Numbers and metrics. Pinning down what a metric means, what a report measures, and whether a surprising number is real.