How the label engine works

“Is a database-calculated label actually legal?” is the most common question in this category, so here is exactly what Nutrillius does with your recipe. The short answer to the legal question is yes — FDA accepts nutrient-database calculation, and we wrote up the regulation behind that answer.

Nutrient data: USDA FoodData Central

Ingredient nutrition comes from USDA FoodData Central — the same public database FDA points manufacturers to — with per-100g values, cup weights and piece weights per ingredient. You measure the way you cook (cups, tablespoons, scoops, pieces); we convert to grams before any math happens. AI recipe import proposes ingredients and amounts from your text or photos; the proposed values are shown for you to review, not silently accepted.

The math follows 21 CFR 101.9, rule by rule

• Serving size comes from the RACC framework in 21 CFR 101.12 — with guardrails on the metric declaration required by 101.9(b)(7) — not from a number you invent. How RACC works.
• Roundingapplies the increments in 101.9(c) per nutrient — calories to 5 or 10, fats to 0.5g or 1g, the “less than” declarations where the regulation requires them. Each rule in the engine cites the paragraph it implements. The complete tables.
• %DV uses the Daily Values, including the tiered vitamin and mineral increments in 101.9(c)(8)(iii) and whole-percent rounding under 101.9(d)(7)(ii).
• The ingredient statement lists by descending weight per 101.4; allergens are detected against the Big-9 with specific-species naming for tree nuts and fish (FALCPA / FASTER Act).
• Formats: vertical, tabular, linear and dual-column — with a guardrail that flags when dual-column is required.

How we check it

The engine runs an automated test suite of 133 tests covering the rounding increments, %DV math, ingredient statements and allergen declarations, and its output has been validated against real retail product labels. When FDA rules change, the engine and these pages are updated together.

What remains your responsibility

A database method is only as good as the inputs: your recipe amounts, your ingredient choices, and your process. For processes a database can’t model — frying, fermentation, smoking — or to substantiate a nutrient content claim, a lab is genuinely the right call, and we say so. Review your values before you print.