Product Data Index — methods

How every figure in the Product Data Index is produced, what is excluded and why.

Sources

Only open and government datasets are used. Commercial catalogue data is excluded from every published figure, including the totals, even where it is held for other purposes.

Family	Datasets	Records (latest edition)
Food and drink	Open Food Facts contributors (ODbL); USDA FoodData Central (U.S. Department of Agriculture)	3,609,227
Cosmetics and personal care	Open Beauty Facts contributors (ODbL)	61,633
Pet food	Open Pet Food Facts contributors (ODbL)	13,942
Other consumer products	Open Products Facts contributors (ODbL)	39,903
Books	Open Library (Internet Archive); Deutsche Nationalbibliothek; Bibliotheque nationale de France	7,119,463
Music releases	MusicBrainz (CC0); Discogs monthly data dump (CC0)	3,121,220
Medical devices (registrations)	U.S. FDA GUDID; European Commission EUDAMED	6,505,371

A “family” is the originating dataset, not a product classification: a record from a music database is counted under music even when its own data is incomplete. Records are database entries; a product can appear in more than one dataset.

Definitions

Valid GTIN-13: 13 digits (a 12-digit UPC-A is read as 0 + UPC) with a correct GS1 check digit.
Barcode issuing organisation: the GS1 member organisation that owns the prefix range (first three digits) of a valid GTIN-13. It tells you where the brand owner obtained its barcodes, not the country of manufacture or origin. Restricted (in-store), coupon, ISBN and ISSN ranges are labelled as such.
Nutri-Score, NOVA, allergens: the values recorded in Open Food Facts (grades A–E; NOVA groups 1–4; the 14 allergen groups regulated in the EU). Records without a value are excluded from that base, and each chart states its base.
E-numbers: codes such as “E330” found in the ingredient text, counted once per record. Additives written by name are not detected, so the counts are lower bounds.
Brands: only brand values recorded by the source dataset itself. Values written by any later enrichment are excluded, as are placeholders such as “unbranded” or “generic”. Spellings that differ only in case or spacing are merged; nothing else is merged. A record count is not a market share, a sales figure or a quality judgement.

Small numbers

Any cell under 10 records is shown as “<10” on the page and left blank (with a flag) in the CSV.

Verification

Each edition is computed twice. The builder streams the records and counts them in a program; a separate checker recomputes every published figure inside the database with independent code (its own check-digit test, grouping and text extraction). The edition is published only if the two agree — 1,617 figures compared for the latest edition. Product figures may differ by at most 0.02% between the two passes, because the registers keep loading while the build runs; recall figures must match exactly. The checker is tested by planting errors in a copy of the edition and confirming that each one fails.

Update cadence and licence

Rebuilt monthly; each edition stays online unchanged under its own address.
Licence: CC BY 4.0 — cite “ProductGuru, Product Data Index” with a link. Figures derived from Open Food Facts (and Open Beauty, Pet Food and Products Facts) also require the attribution “Open Food Facts contributors, ODbL”.
Related dataset: recalls by barcode, DOI 10.5281/zenodo.23943408 (concept DOI 10.5281/zenodo.22685727).
Recall statistics follow their own methods: Safety Data Index methods on product-guru.org.
Corrections: contact us; a corrected edition is republished with a note.

The text of these pages was written with AI assistance; the figures are computed by program. See how we use AI.