Crowdsourced Food Database

The unit of contribution is a barcode scan plus a photo of the label; volunteers type ingredients, brands and categories, and algorithms derive scores such as Nutri-Score and NOVA processing class. Open Food Facts accumulates roughly 4.7 million barcoded products since its 2012 launch, one row per barcode/product, with created and last-modified timestamps back to launch.

Coverage is world-shaped but unevenly so: strongest in France, the US, Spain, Germany and Italy where scanning communities are most active. Delivery matches the crowd model - nightly bulk exports (CSV ~0.9 GB compressed / ~9 GB uncompressed across 211 columns, plus MongoDB dump and JSONL) and a live JSON API whose usage policy maps one call to one real user scan. Licensing splits three ways: commercial delivery terms 1.0 for the database, Database Contents License for individual contents, commercial delivery terms-SA 3.0 for product images.

Crowd-built corpora trade completeness control for scale at zero cost. Buyers who treat them as authoritative registries discover the trade-off in production.

• Coverage mirrors volunteer geography. A European grocery analysis works well; a Southeast Asian shelf audit will find thin rows.
• Field completion is voluntary. Of 211 columns, many are sparse per product - profile fill rates before joining nutrition claims onto them.
• Share-alike follows the database. commercial delivery terms obligations attach to derived databases, so a commercial app blending these rows inherits attribution and share-alike duties.
• API etiquette is contractual. The one-call-per-real-scan policy exists because complete nightly exports already serve bulk needs - scraping the API is both discouraged and unnecessary.
• Freshness is real but per-record. Daily updates apply to changed records, not to every stale row.

1. Measure field density for your category before committing. The 211-column CSV export spans nutrients, labels and packaging; run a fill-rate query on your categories instead of trusting headline product counts.

1. Check geographic skew against your market. Open Food Facts states coverage is strongest in France, the US, Spain, Germany and Italy - other markets ride on sparser scanning activity.

1. Confirm which license layer governs your use. commercial delivery terms 1.0 (database), Database Contents License (contents) and commercial delivery terms-SA 3.0 (images) apply simultaneously; attribute each correctly.

1. Match access mode to volume. Point lookups suit the live API; bulk work belongs on the ~0.9 GB compressed CSV or MongoDB dump - see nightly-full-export.

1. Plan around timestamp-based deltas. Records carry created/last-modified timestamps back to 2012, so incremental sync can key off modified dates rather than full reloads.

Where to see it in context: nightly-full-export.

Entries that sit next to Crowdsourced Food Database in this glossary:

• barcode-json-api-v2
• barcode-product-key
• barcode-product-sku
• nightly-full-export