Calorie app that shows sources: how to check where every number comes from

Download the Radish app

What "shows sources" actually means

The word "source" gets used loosely. In a nutrition database it has a precise, checkable meaning. A properly sourced value carries three pieces of information, not one:

• The named table. Ciqual (ANSES), USDA FoodData Central, a commercial database, or user contributions. These are not synonyms and they do not share a method.
• The release year of that table. Ciqual 2020 and Ciqual 2025 do not hold the same values: ANSES's own documentation states that the 2025 release adds "près de 300 aliments en plus" (close to 300 additional foods) and 7 new constituents.
• The entry code for the food. That code is what lets you reopen the original record and compare. Raw garlic, for instance, is Ciqual code 11000.

Without those three, "data from official sources" is a sentence, not evidence. Our methodology page sets out how we cross-check each figure before publishing.

Three checks that test any app in five minutes

None of these needs technical skill. They work on any calorie counter, ours included.

1. Open a simple food and look for a table name. Pick something raw — an apple, cooked rice, a chicken breast. If the screen names no table, the information is not hidden in a menu somewhere: it is absent.
2. Look for a code, a version or a date. A number with no entry code cannot be rechecked. A table with no release year cannot be dated.
3. Compare the same food across two apps. A 10% gap on a raw food is normal (variety, cooking, season). A 40% gap points to two different data origins — and the app that names its own is the one you can actually arbitrate.

The same logic explains why two nutrition ratings can disagree on one product: different data, different calculation.

The two public reference tables

Ciqual, the French table published by ANSES

Ciqual is published by the Observatoire des Aliments unit at ANSES, the French agency for food, environmental and occupational health safety. The official documentation for the 2025 release (dated 19 Nov 2025) states that the table covers 3,484 foods and 74 constituents, in open access. Two details are worth knowing, because they are rare:

• Each constituent carries an INFOODS code, the standard identifier maintained by the FAO's INFOODS network of experts. That is what makes two databases interoperable rather than merely comparable by eye.
• A separate file, sources_2025_11_03.xml, lists the "sources des données utilisées pour produire la teneur dite moyenne" — the sources behind each average value. The table does not just hand you a number; it documents where the number came from.

ANSES asks that any reuse cite the source explicitly, as "Anses. 2025. Table de composition nutritionnelle des aliments Ciqual", with DOI 10.57745/RDMHWY.

USDA FoodData Central, and why the data type matters

On the US side, FoodData Central describes itself as "USDA's comprehensive source of food composition data with multiple distinct data types". That plural is not decorative: the site separates Foundation Foods ("analytical data/metadata on commodity and minimally processed food samples"), SR Legacy Foods ("historical data derived from analyses, calculations, and published literature", final update 2018), Survey Foods, Experimental Foods and Branded Foods ("label data collected through a public-private partnership").

The practical consequence is simple: saying "USDA data" means almost nothing until you say which type. A lab-analysed value and a value copied off a label do not carry the same weight.

Open Food Facts, its strength and its limit

Open Food Facts describes itself as "a collaborative, free and open database of food products from around the world", "made by everyone, for everyone", built by thousands of volunteers. That is exactly what makes it unbeatable on packaged products and barcode scanning — and it is also the limit: a contribution stays a contribution. Use it to scan, not to settle a micronutrient level. Our article on free barcode scanning covers that division of labour.

What editors publish on their own pages

On 25 Aug 2026 we opened each editor's public page and looked for one thing: does the page name a nutrition data source? The answer is reproduced as found, with no interpretation.

How to read this: Yes = a source is named on the public page we opened · Not documented = that page names none. This does not mean the editor has no source: it means that page does not publish one.

No-affiliation notice. Radish is not affiliated with, sponsored by or partnered with any of the apps named here. Brand names belong to their respective owners and are used for identification only. Radish publishes one of the apps compared — which is precisely why the method and the check dates are published alongside.

What Radish shows, and what it does not

Line by line, to be precise:

• On the site, the "Where the numbers come from" page names the tables used and their release year, and every published food page shows the table, the year and the entry code.
• Nutrition values come from the official Ciqual (ANSES) and USDA FoodData Central tables, not from unreviewed contributions.
• Barcode scanning relies on Open Food Facts, a collaborative database — flagged as such, and limited to packaged products.
• What is not there yet. A detailed, screen-by-screen source display inside the app is being built and is not shipped as of 25 Aug 2026. We write that down rather than let it be assumed.

On micronutrients themselves, our 14 reference points and what ANSES, EFSA and Ciqual actually say set out where the displayed thresholds come from — a separate question from what a food contains.

Why the source matters more than the database size

The most common sales pitch is the food count: one million, five million, fourteen million. That figure mostly measures how many packaged products are listed, so how likely it is that your cereal box exists. It says nothing about the quality of the value attached to it.

For a raw food — three quarters of what you eat if you cook — the opposite holds: a small, lab-analysed table beats a huge, copied one. That is how an app with 3,484 documented entries can be closer to the truth day to day than an app with millions of entries of unknown origin. The useful question is not "how many foods?" but "where did this particular number come from?".

The same reasoning applies to privacy: what calorie apps actually know about you is readable in their documentation, not in their marketing. And the Nutri-Score, computed from sourced values, is only ever worth as much as those values.

Sources and check dates

• ANSES — Table de composition nutritionnelle des aliments Ciqual 2025, Documentation, document dated 19 Nov 2025: 3,484 foods, 74 constituents, INFOODS codes, file sources_2025_11_03.xml. ciqual.anses.fr, checked 25 Aug 2026. Citation requested by ANSES: "Anses. 2025. Table de composition nutritionnelle des aliments Ciqual", DOI 10.57745/RDMHWY.
• FAO — INFOODS network, food component identifiers: fao.org/infoods, checked 25 Aug 2026.
• USDA — FoodData Central, definition and data types (Foundation, SR Legacy, Survey/FNDDS, Experimental, Branded): fdc.nal.usda.gov, checked 25 Aug 2026.
• Open Food Facts — project self-description (collaborative, free and open database): world.openfoodfacts.org, checked 25 Aug 2026.
• Cronometer — free-account features and "1M+ lab-analyzed or verified foods": cronometer.com/gold, checked 25 Aug 2026.
• Yazio — official product page, no nutrition table named: yazio.com/en, checked 25 Aug 2026.
• Cal AI — official product page, no nutrition database named: calai.app, checked 25 Aug 2026.

Editors' pages change. If you spot a discrepancy with this check, tell us: the line will be corrected with a new check date.

General educational information, not medical advice. If you have a medical condition, are pregnant or follow a specific diet, talk to a healthcare professional.