Vendor-feed parity is not one number.

The problem

Two feeds can appear comparable while representing different coverage, timing, instrument definitions or correction behavior. A simple row count or blended match rate is not enough to decide whether a replacement is safe.

Synthetic validation artifact

This synthetic diagram shows the control states, not licensed source records. Population coverage, contract-level matching, field-level parity and unmatched taxonomy are distinct measures with distinct denominators.

The validation design

1. Define a canonical comparison schema.
2. Match in stages, retaining the reason each record did or did not match.
3. Use explicit denominators for each question.
4. Measure field-level parity separately from population coverage.
5. Classify unmatched records before attributing error.
6. Retain reproducible checks for later releases.

The boundary

Unmatched records do not automatically imply vendor error. They may describe a different active population, a timing boundary or a legitimate semantic difference. The decision record preserves that uncertainty.