Product brief

The fictional Northstar collection editor needs a known asset despite inconsistent tags. The decision is whether transparent lexical rules and a reviewed one-record correction help satisfy explicit intent. A metadata drawer makes evidence inspectable without turning search into a dashboard.

The principal alternative is semantic ranking. It could accommodate unfamiliar terminology but cannot be honestly demonstrated with this static sample. The chosen alternative is all-token lexical matching over title and controlled tags, plus one visible bike/bicycle expansion. No real assets, employer taxonomy, external search, embeddings or proprietary results are used.

Smallest observable outcome: find the commuter artwork after reviewing its missing bicycle tag, explain each matched field, and reverse an excluding facet. Material choices are provisional until human evaluation. Mo owns product/program direction; AI assists implementation and verification. Publication status is centralized in Validation.md.