The open data behind a product record: what it holds and what it is worth

A database is open when anyone can download it, read it with a machine and reuse it, with the sole duty of crediting it and keeping derived works open. Open does not mean verified: the licence describes what you may do with the data, not how correct that data is. The two properties must be kept apart, because a source that is very permissive about reuse can be very uneven in quality.

• An explicit licence allowing reuse, usually with attribution and share alike conditions.
• Access without a commercial key and without an individual contract: everyone queries the same service under the same rules.
• A machine readable format, not only a page designed for human eyes.
• Declared governance: you can tell who accepts changes, through which process and how long they take to appear.

The last point is the one that weighs most in daily use. A database run by a public agency publishes updates at regular intervals through a documented procedure. A collaborative database accepts contributions continuously, so it grows quickly but can hold a wrong line for days, entered in good faith.

The databases listed below are the ones that genuinely underpin records of foods, drugs, substances and places. Each row states the kind of data, who maintains it and the limit you hit first when you try to push it beyond its purpose.

None of these sources requires a paid key. That is a verifiable choice rather than a statement of intent: the full list, saying which call leaves from the server and which from the device, sits on the integrations page and is kept aligned with the feature catalogue by an automated check.

The most frequent mistake is not picking the wrong source but asking it a question it cannot carry. A geographic database knows where a clinic is, not whether a vet is on duty right now. A chemical database knows what a molecule is, not how much of it will harm a small cat.

Rewriting the question into a form the source can carry is almost always possible and changes the value of the answer. Instead of asking whether a food is suitable, ask what the label declares and compare that figure with the energy requirement worked out separately. Instead of asking whether a substance is dangerous, ask whether it is classified as such and take the quantitative question to whoever holds the toxicological references.

Pharmacovigilance archives are the most valuable and most misread open source. They collect spontaneous reports: someone observed an event after an administration and reported it. Neither causation nor frequency follows from that. A high number of reports may reflect a widely sold product, a press campaign or a particularly active reporting network.

• The denominator is missing: you know the reported events, not the total administrations, so no rate can be computed.
• Under reporting is the norm: mild and expected events are reported far less often than severe or unexpected ones.
• Causation is not established: the same record may describe a drug reaction, disease progression or a coincidence.
• Comparing products is not legitimate: two molecules with different sales volumes and populations are not comparable on raw counts.

Before basing a decision on a record it is worth spending the time of four checks. They are quick and they catch almost every error that matters, meaning the ones that would change the choice.

1. 1

   Compare with the physical object

   If you have the pack in hand, read two values off the label and look for them in the record. A mismatch on an analytical constituent signals an old transcription, or one taken from a different pack size of the same product.

2. 2

   Look at the date and the unit

   Every line should carry the date of the value and the unit it is expressed in. A figure without a unit or without a date is not data: it is a number, and it compares to nothing.

3. 3

   Ask who wrote it

   In collaborative databases the edit history is public and shows how many people touched that entry. A record with a single contribution and no label photograph should be treated as a draft.

4. 4

   Look for a second independent source

   If the value bears on a clinical decision or a repeat purchase, confirm it with a source that does not derive from the first. Two websites copying the same archive are not two confirmations.

Some information sits in no queryable archive and comes from guidelines and reference literature: vaccination protocols, recommended recalls, growth charts, interaction references between molecules, European regulatory documentation on veterinary medicines. In the product these sources are transcribed as references and generate no network call at all: no external provider is contacted when a recommendation is consulted.

• International and national vaccination guidelines: they define core components, intervals and criteria for antibody titre testing.
• Consensus statements from specialist societies: used for interaction checks inside treatment plans.
• European regulatory documentation on veterinary medicines: the basis of authorised product information.
• Reference growth charts: they support the puppy growth curve and comparison with the population.

The distinction between a queried source and a transcribed reference has a practical privacy consequence. A call leaving from the browser shows the provider the address of the user device; a call leaving from the server shows only the address of the infrastructure; a reference shows nothing to anyone, because no request exists.

Features that rely on external data declare the source in the catalogue, and the integrations page collects the list with the host contacted and the place the call leaves from. An automated check prevents a feature from citing a source not described in the register, so the list cannot quietly go stale while the product changes.

• Food records, label reading and product comparison in the nutrition area: they rely on Open Pet Food Facts.
• Active substance records, adverse reaction reports and co reported reactions in the medicines area: they rely on openFDA.
• Toxic plant and substance records in the toxicology area: they rely on PubChem for identifiers and chemical classification.
• Useful places, parks, water points and clinics currently open: they rely on OpenStreetMap queried through Overpass.
• Environmental indices such as parasite risk, heatstroke, air quality and the usable walking window: they rely on Open-Meteo.