Account health audit (claims vs independently re-derived evidence)

The insight: re-running the provider that filled the field is not an audit — agreement with yourself is not evidence. Most "data audits" re-enrich from the same source that populated the record, get the same answer, and report the book as clean. The record was never tested; one function was asked to confirm itself.

But independence alone is not enough either, and this is the part that decides the design. Two 1-credit enrichment providers, queried on the same domain on the same day, returned headcounts of 17,112 and 11,303 — 5,809 people apart, 51% above the lower figure, neither flagged as uncertain. And worse: each provider contradicted itself inside its own payload. One returned an exact count of 17,112 alongside a size band of 5,001-10,000. The other returned 11,303 alongside a band of 5001 to 10000 — which excludes it — and a second range field of 10001–20000, which does not. Three headcount claims in one record, two of them mutually exclusive.

So there is no source of truth to overwrite toward, and the deliverable cannot be a corrected record: