Deterministic scoring
The score is a function of public nutrient data. Same inputs, same score, every time. There is no editorial adjustment, no manual override, no secret sauce.
About · Who is behind the score
The honest score for every food.
FoodScore is a data-driven US nutrition database that rates every food from 0 to 100 using a deterministic formula built on the USDA Dietary Guidelines for Americans 2020-2025 and the NIH Dietary Reference Intakes. No editorial bump. No brand influence. The algorithm is publicly documented and the source code is open.
Why this exists
Existing food-rating tools all have meaningful gaps. Yuka is French-market-first and its scoring does not map cleanly to US nutrition labels or the Dietary Guidelines for Americans. EWG Food Scores leans on the UK Ofcom 2004 methodology, which was never designed for the current American food supply and has been repeatedly criticised as outdated. Government databases like USDA FoodData Central are exhaustive but not built for day-to-day consumer decisions.
FoodScore fills the gap between label-level simplicity and authoritative data. It cites the USDA and NIH when it matters, shows the math, and never turns nutritional data into sales.
Editorial principles
Public algorithm
Every bonus, penalty and threshold is documented in the methodology page and in the foodscore.ts source file. You can recompute every score from the raw USDA data and this formula.
Independent
No brand sponsorships, affiliate kickbacks, or certification-body relationships influence food rankings. If advertising ever appears, it will be clearly disclosed and served programmatically.
Data over opinion
Foods are not 'healthy' or 'unhealthy' in absolute terms. A 20/100 soda eaten occasionally is not a crisis. A 90/100 yogurt eaten without restraint can still be a problem. We publish the number, you keep the judgment.
Transparent sources
Nutrient values come from USDA FoodData Central (Foundation, SR Legacy, Branded datasets). Daily Value baselines follow the FDA 2016 Nutrition Facts rule. Processing level uses NOVA classification. Every source is linked on every food page.
Versioned calibration
When the formula changes, we publish the diff and the reasoning. Historical scores are not rewritten silently. The methodology page carries a version number.
How FoodScore is built
- 01
Data ingest
Quarterly pulls from USDA FoodData Central for Foundation, SR Legacy and Branded entries. Raw nutrient panels are normalised to per-100g.
- 02
Deterministic scoring
Each food passes through the FoodScore formula (src/lib/foodscore.ts). Base 50, bonuses capped at +35, penalties with field-specific thresholds, clamped to 0–100.
- 03
Editorial content
Per-food takeaways, score explanations, and health-benefit summaries are generated from the nutrient panel via a constrained LLM prompt. Content is always grounded in the measured values and never invents studies or quotes named experts.
- 04
Public deployment
Everything is statically generated. No runtime database, no API calls on page load. The score you see is the score we compute from published USDA numbers.
Who builds this
FoodScore is built by the team behind NetLifeValue and TheCosmicPet, two other structured-data consumer tools. We have been building programmatic, data-driven websites for years, with an emphasis on search optimization that actually serves the reader rather than gaming rankings. FoodScore brings that approach to food nutrition: exhaustive US-specific data, transparent methodology, no marketing language.
Found a data error or calibration case we missed?
We read every report. Data issues get fixed on the next refresh.