Methodology
How AnyUPC decides what a product is.
Every fact on an AnyUPC product page is traceable to a source, dated, and marked verified, unverified, or conflicting. This page explains exactly how a record is compiled and scored, so the result can be checked rather than taken on faith.
Where the data comes from
AnyUPC is a hybrid of structured data feeds, targeted scraping of brand and retailer pages, and AI extraction used only at the margins — never AI-only. Each source lands in the record with an explicit authority class:
- Brand websites — the manufacturer's own product page. Highest authority (authoritative), because the brand is the primary source for its own product.
- GTIN registries & government databases — the USDA FoodData Central nutrition database and GS1 identification standards.
- Retail listings — Amazon, Walmart, Target and category-matched retailers, used for offers, pricing, and corroborating attributes. Confirmed against the GTIN wherever the retailer exposes it.
- Structured aggregators — Open Food Facts (nutrition, ingredients, images) and UPC databases, used to fill gaps cheaply before any scraping.
- AI cross-referencing — used only to fill fields no source covered, always marked observational and unverified.
How facts are chosen (the truth engine)
Sources routinely disagree — a weight here, an ingredient there. AnyUPC does not silently pick one. Every observation is stored; a single canonical value is then selected by source authority, then verification state, then freshness. A higher-authority or independently verified fact outranks a lower one, and a verified fact is “sticky” — a weaker source cannot overwrite it. When sources genuinely conflict, the record keeps the conflict visible rather than hiding it.
The Data Authority score
Each record carries an explainable score built from four components:
- Completeness — the share of the fields its category is expected to have (a food product is measured against a nutrition panel; a tool against its specs).
- Confidence — the mean source authority across its facts, weighted by verification state and how recently each was confirmed.
- Coverage — how many independent sources corroborate the record.
- Identity — whether the core identifiers (title, brand, valid GTIN, canonical category) are present and sound.
Every score is inspectable on the record itself — “view all sources” shows the exact contribution of each source and fact. Records below the quality bar are held back rather than published.
Corrections
Facts change and sources are imperfect. Anyone can flag a correction on a record, and brands can claim their catalog to manage it directly. Corrections are reviewed against the same authority rules as any other source.
