Blog

Where calorie databases get their numbers, and where they do not

September 23, 2026 · Viral Tandel

Calorie database accuracy is not one property a database either has or does not have. It depends on where each individual figure came from, and a single app can contain lab-tested numbers, manufacturer figures and unverified guesses sitting side by side, looking identical in the search results.

Key takeaways

• A calorie figure can come from a government lab survey, a manufacturer's own tested panel, or an entry someone typed in by hand. Those are three different levels of evidence.
• USDA FoodData Central entries carry a traceable ID number and a documented method behind every figure, which is what makes them checkable.
• A manufacturer's barcode figure is measured for that exact product, but only that product; it tells you nothing about a home-cooked or restaurant version of a similar dish.
• Crowdsourced entries can be accurate, but the database itself usually cannot tell a careful entry from a guess, and neither can you without checking.
• A number worth trusting is a number you can trace back to where it came from. That is the whole discipline behind this site's own figures.

Three kinds of number, one search box

Search almost any food in almost any tracking app and you get a list of results. What that list rarely shows you is where each entry actually came from, and that matters more than the food name does.

A lab-measured, standardized entry. Built from a documented testing or survey process, published with an ID you can look up, and describing a defined food as prepared a specific way. USDA FoodData Central is the clearest example available to the public.

A manufacturer's own panel. Measured by or for the brand for that specific product, printed on the package, and generally accurate for that exact item. It says nothing about any other brand's version of a similar product.

A crowdsourced or user-submitted entry. Typed in by someone using the app, sometimes copied carefully from a real label, sometimes estimated, sometimes simply wrong. The database has no reliable way to tell these apart once they are all sitting in the same results list.

USDA FoodData Central: how a lab number gets built

USDA FoodData Central entries come from a documented process, either direct laboratory analysis of food samples or large-scale national dietary surveys, and every entry carries a specific FDC ID number tying it back to that record. Some entries note the dataset behind them directly, for instance "SR Legacy" for older standard-reference lab analyses or "FNDDS" for the food and nutrient database built from national dietary survey data. That documentation is exactly what lets a figure be checked rather than simply trusted.

A dosa entry, for example, carries FDC ID 2708347, is documented against the FNDDS 2024-10-31 release, and lists 210 kcal, 5.7 g protein, 37.04 g carbs and 4.05 g fat per 100 g. Every part of that sentence is checkable against the public record. That traceability, not the specific number, is what separates this tier from the other two.

Manufacturer panels: measured, but narrow

A barcode scan works from the same principle in a narrower lane: the brand tested or calculated the figure for its specific product and is legally responsible for what it prints. That makes it a genuinely measured number, not an estimate, for the exact item you scanned. It does not extend any further than that. A generic brand's granola bar and a store brand's version can differ meaningfully even with a similar name and package, which is exactly why the barcode path scans the specific product rather than assuming a category-level number will do.

Crowdsourced entries: useful breadth, unverified depth

Letting users add entries is how many databases cover the enormous long tail of restaurant dishes, regional foods and homemade recipes that no lab will ever test. That breadth is genuinely useful. The tradeoff is that the database usually cannot distinguish a careful entry, copied faithfully from a real label, from a rough guess submitted with confidence. Duplicate entries for the same product, wrong portion sizes, and figures that were never checked against anything all end up looking exactly as credible as a properly sourced one, and a search result rarely tells you which you are looking at.

Why this matters once the number reaches your log

An unsourced figure that happens to be right and an unsourced figure that is off by 40 percent look identical in an app. The only way to tell them apart is to trace the number back to something, which most database entries do not make possible. Calorie counting margin of error covers how much that kind of untraceable error can compound across a full day of logging, and it is more than most people assume. It compounds differently again once a photo estimate is layered on top of an already shaky database match, which is the specific failure mode covered in why photo calorie estimates run low.

How KalTrack sources its own numbers

This site and the app behind it work from exactly two kinds of figure: USDA FoodData Central entries with a real ID, and a brand's own published data reached through a barcode scan. Where neither exists, the honest answer is to say so rather than to invent a number that looks as confident as the sourced ones next to it. A photo-based estimate is always presented as an estimate, with an easy way to correct it, never dressed up as a database figure it is not. How accurate are calorie counting apps and how to photograph a meal for a calorie app cover that estimate side of the picture in more depth.

Next time a search result hands you a number, ask where it came from before you trust it, and try logging with KalTrack to see the difference between a sourced figure and an estimate labeled honestly as one.

This is general information about tracking food, not medical advice. If you are managing a health condition or working to a clinical target, talk to your doctor or a registered dietitian before changing how you eat.