About Our Data: Lab-Verified Data Methodology
What "Lab-Verified Data" means
FoodCanEat is a data editorial team. We do not run clinical trials, and we do not generate nutrition numbers — nobody outside a laboratory does. What we do is take authoritative public laboratory data, attach it to the right food with documented confidence, and publish it in a form you can act on and check. The name on our masthead — Lab-Verified Data — is a commitment: every figure on this site traces to a specific, citable laboratory measurement, and every data card prints the ID you need to verify it yourself.
This page documents the whole pipeline: where the data comes from, how foods are matched to lab entries, how we grade our own confidence, what we do when no reliable number exists, and the known limits of all of it. If you have ever wondered whether a nutrition site made its numbers up, the answer on this site is "check us" — and this page is the manual.
Where the data comes from
We use two public datasets from the United States Department of Agriculture:
USDA FoodData Central, SR Legacy (April 2018 release). The Foundation and SR Legacy branch of FDC is laboratory-derived composition data for thousands of core foods — the direct descendants of the datasets that have underpinned American nutrition science for a century. Our working table extracts 7,231 entries with computable net carbohydrates from the full 7,793-entry set, covering eleven nutrients per food: total carbohydrates, fiber, sugars, fat, saturated fat, protein, calories, potassium, sodium, magnesium and cholesterol.
USDA FoodData Central, FNDDS Survey Foods (October 2024 release). Where SR Legacy lacks a food that matters — a plain crepe, for instance — we supplement from the survey dataset, clearly labeled on the card as a different source with its own FDC ID.
Both datasets are public. The full datasets are downloadable from USDA, which means every figure we publish is independently reproducible by anyone with a spreadsheet and an afternoon.
How a food gets matched to a lab entry
A verdict article is not about an abstract food; it is about a specific laboratory entry. Matching "the white rice you cook at home" to "the white rice the lab measured" is a real engineering task, and errors there poison everything downstream. Our matching pipeline:
1. Exact and near-exact matching. Food names are normalized (stemmed, synonym-expanded — "lychee" to "litchi", "starfruit" to "carambola") and matched against the USDA table with category guardrails: a beef kidney must not match to a bean, however the strings line up.
2. Form matching. Raw must match raw, dried must match dried, canned must match canned. A dried shiitake mushroom and a fresh one differ by a factor of several in most nutrients; form errors are the most damaging class of data errors we know of, and the pipeline explicitly weights raw/fresh forms and demotes juice, flour and powder variants.
3. Manual verification of every borderline case. Every match the engine could not make with high confidence — 45 entries in our most recent audit — was re-verified by hand against the USDA source. Nineteen were corrected (a red kidney bean entry that had silently matched red bell pepper, a pork liver that had matched ham), thirteen were confirmed and promoted to full confidence, and the remainder carry explicit approximation disclosures. The full correction log is public further down this page.
Confidence grades, disclosed on every card
Not every match deserves the same trust, so every food in our verdict library carries one of three grades:
| Grade | Count | Meaning |
|---|---|---|
| HIGH | 628 | Exact species, exact form, matched to a specific FDC entry |
| MED | 16 | Closest authoritative approximation — e.g. no "beef neck" entry exists, so the chuck region stands in — always with a note saying so |
| NO_DATA | 18 | No reliable laboratory match exists; no figures are shown |
The no-data policy: an honest gap beats a confident guess
Eighteen foods in our library — mostly Japanese ingredients with no US laboratory entry, like sea urchin or dried nori — carry the NO_DATA grade. Their data cards show no numbers at all: not zero, not an estimate, nothing. This is a deliberate policy with one rule behind it: we do not publish figures we cannot trace to a laboratory measurement.
The corollary is also worth stating: a genuine zero is published as a zero. Salmon really does carry 0 g net carbohydrates, and the card says 0 because the lab said 0 — which is a different statement from "we don't know". Distinguishing "measured as zero" from "not measured" is one of the small honesty markers that separates a data operation from a content mill.
How the numbers are computed and displayed
Net carbs = total carbohydrates − fiber, per 100 grams, on every food, everywhere. The net carbs guide explains the reasoning; the display convention is that sugars and fiber are shown as child lines under total carbohydrates — a parent-child breakdown, never three parallel numbers, because sugars and fiber are subsets of the total and adding them up would be an error.
The 100-gram scale is universal. No serving-size arithmetic, no brand-specific portions. Two verdict lines are applied to the same figure: keto (YES ≤ 5 g, IN MODERATION ≤ 20 g, NO above) and low-carb (YES ≤ 20 g, IN MODERATION above). Both badges appear on every card so you can read any verdict through whichever pattern you follow — the logic is in Keto vs Low-Carb.
What our data cannot tell you
Every database has edges, and pretending otherwise is how data sites drift into fiction. Ours:
No added-sugars line. SR Legacy predates the modern added-sugars label convention for most entries; we report total sugars and say so.
No sugar alcohols for whole foods. The erythritol arithmetic on packaged "keto" products is outside our scope entirely.
No glycemic index. No GI/GL figures of sufficient quality exist across the whole table, so we publish none rather than mixing good and bad sources.
No brand entries. Supermarket brands and restaurant items are excluded by policy; SR Legacy measures foods, not marketing.
Sampling reality. A laboratory value describes the sample measured. "Year round average" entries average across seasons and regions; your local tomato will differ by a few percent. We treat small deviations as noise and refuse to fake precision — figures are shown to two decimals because the source does, not because the third decimal would mean anything about your lunch.
Correction log
A data operation that never corrects itself is not checking itself. Our corrections are public:
September 2, 2026 — full re-verification of all 45 borderline matches. Nineteen FDC mappings corrected, including five severe species/form errors (red peas matched to red bell pepper; pork liver to ham; dried wood-ear mushroom to shiitake; dried nori to agar; beef large intestine to rib eye). Two foods downgraded honestly to NO_DATA where no reliable entry exists. Thirteen verified and promoted to HIGH. Every corrected card carries a note describing the change. The recheck was triggered by a routine name-and-category audit, which is the point: the process exists to catch our own errors, not just everyone else's.
Corrections to this log are made continuously; the date of each entry reflects when the fix shipped, not when we noticed.
How to verify anything we publish
The verification loop is two steps:
1. Find the FDC ID — printed in the source line of every data card and quoted in every verdict article.
2. Look it up at the USDA's public FoodData Central search, paste the ID, and compare. If a figure on this site does not match its cited FDC entry, that is a bug and we treat it as one — our contact page accepts data corrections with priority.
That is the whole system: authoritative sources, documented matching, graded confidence, honest gaps, public corrections, and an ID on every number. "Can I eat this on keto?" is a simple question, but the answer is only as good as the data underneath it — and this page exists so you never have to take that on faith. Start with the beginner guide, browse the daily verdicts, and check our work.