Open Food Facts Bulk Export Data: Dataset Structure and Field Coverage

Scale is the headline. The project's homepage counter showed 4,697,513 products when Datadory researched the source on 2026-08-21, and records carry created and last-modified timestamps back to the 2012 launch. Coverage is worldwide but weighted toward France, the United States, Spain, Germany and Italy, where scanning communities are most active.

Commercial identity comes first: product_name, brands, quantity as printed on the pack, packaging, categories, the stores that carry the product and the countries where it is sold. Origin fields follow - origins of ingredients, manufacturing_places, labels and certifications - which is the block sourcing teams use for country-of-origin and certification screens.

Composition is the deep part of the schema. Rows carry the full ingredients_text string, allergen and trace flags, an additive count, roughly 150 per-100g nutrient columns from energy-kcal_100g through vitamins and minerals, Nutri-Score as both numeric score and letter grade from a to e, the NOVA processing group from 1 to 4, PNNS food-group classifications and an Environmental Score. Front-of-pack, ingredients and nutrition photos ship as image URLs.

Provenance metadata lets you filter for quality: creator, created_t and last_modified_t timestamps, unique_scans_n from the scanning app and a per-record completeness score. One documentation quirk is worth noting - the legacy data-fields dictionary does not yet describe nutriscore_score or nova_group even though both appear in the current CSV, so treat the export itself as the field reference.

Catalog enrichment is the dominant job. Distributors load the nightly CSV into a warehouse, deduplicate on the code column, then join on barcode to USDA FoodData Central (Branded Foods + Foundation + SR Legacy + FNDDS), whose Branded Foods table keys on gtinUpc and refreshes monthly under commercial delivery terms - the April 2026 release ships 195 MB of zipped JSON and 428 MB of zipped CSV expanding to roughly 3 GB across 1.7 million-plus branded US products. Open Food Facts contributes global barcode reach, images and Nutri-Score; FoodData Central contributes authoritative US composition with documented derivations. The Open Food Facts vs USDA FoodData Central comparison sets out where each wins.

The attribute payload feeds merchandising directly: allergen and trace flags, additive counts and NOVA groups become structured product-page facets, ingredient text powers dietary filters without anyone transcribing labels, and unique_scans_n doubles as a crude popularity signal when you rank which enriched items deserve homepage placement first.

Assortment and demand context comes from the government neighbors in the same pool. U.S. Economic Census - Wholesale Trade (NAICS 424) counts 34,604 grocery wholesale establishments (NAICS 4244) with $1.30 trillion in sales and 621,911 employees in its 2022 file, and Monthly Wholesale Trade Survey (MWTS) - Sales & Inventories tracks monthly US merchant-wholesaler sales back to January 1992. Together they give an enriched catalog a market frame to be measured against.

Deltas never capture deletions. The daily files listed in delta/index.txt cover the previous 14 days of changes only, so a warehouse loaded once drifts until periodic full re-dumps rebuild it.

Completeness varies by contributor. Every record carries a completeness score, and the export includes rows holding little more than a barcode and a brand name, so filter on completeness before publishing attributes to customers or feeding a pricing engine.

Watch the mirror trap. Kaggle - World Food Facts (Open Food Facts mirror) freezes roughly 356,000 products at version 5 dated 2017-09-18 and scores 6 in Datadory's catalog against the live export's 10 - see what the Kaggle Open Food Facts dataset actually contains. Product counts also fluctuate daily as contributions land, so quote a count with its observation date rather than treating 4.7 million as fixed.