URSY.food: finding a dependable answer in conflicting food data

Two sources can describe the same food differently. The ingredient list in a database may differ from the packet. A product name may have changed. A safety notice may arrive after an older record was collected.

URSY.food is being built to keep track of those differences and explain which source supports a published answer.

By 11 September 2026, its working database held more than fifty million stored claims, each tied to evidence. One product can have many claims attached to it, including older versions and conflicting information. The size of the collection makes the individual decisions more important.

An assistant may eventually ask a short question about a product. The work behind a useful answer can involve deciding which of several records deserves to be used.

The packet and the database#

The published source policy gives different kinds of information their own order of preference. For an ingredient list, the physical pack comes ahead of a database. For a recall, a regulator can take precedence over the manufacturer.

When one value wins, the other remains in the evidence with the decision recorded. If the rules cannot settle a material disagreement, the field stays unknown with a reason. Some safety-sensitive information is held for a person to review before publication.

That gives the eventual reader more than a filled-in box. It provides a way to understand where the answer came from and why an alternative was set aside.

The current source register includes USDA FoodData Central and Open Food Facts, with their recorded permissions and review dates. Approval to use a source is kept separate from whether its material has actually been collected and published.

This is the food-information part of URSY’s wider ambition. An assistant checking a product needs a useful answer, and sometimes the useful answer is that the record cannot yet settle the question.

For us, that is the value of keeping the sources and their history. A larger collection should help explain a product more clearly, including where the evidence falls short. An empty field with a reason can be more useful than a confident answer borrowed from the wrong record.