Reference
The data model.
Every object in the response, field by field — types and nullability transcribed from the backend source of truth.
Top level
Whether a catalog product confidently matched. false = limited-coverage shape.
Canonical catalog name of the matched product.
Manufacturer or brand name.
One-line editorial identity for the product.
Long-form synthesized verdict (dropped in compact mode).
The composite 0–100 score.
How many independent sources were synthesized.
ISO timestamp of the evaluation snapshot.
Scores
dimension (experts / users / value / …), score, and a synthesized summary.
All 27 criteria across 7 dimensions, each with a name and 0–100 score.
Claims
Verified strengths: title + detail, each traceable to sources.
Reported issues, same shape — the honest half of the synthesis.
Consensus & commerce
Per-source rows: source, source_type, score, stance, and a per-source summary.
Live offers: retailer_name, price, price_as_of, and a monetizable buy_url.
List price at launch, when known.
Primary purchase link.
Resolution
Match confidence plus other_matches[] (name + match_score) so your agent can sanity-check the match.
Where the synthesis came from: catalog snapshot and evaluation lineage.