How Our Data Works

Transparency is core to NutriBalance. When you make dietary decisions based on our data, you deserve to know exactly where it comes from, how confident we are, and what review process it went through.

The Data Pipeline

download

Source

Data is imported from government databases, scientific papers, or researched by AI.

visibility

Observe

Each value becomes a substance observation — tagged with its source, derivation method, and initial confidence score.

link

Evidence

Observations link to evidence items: page references, row locators, URLs, and text snippets from original sources.

rate_review

Review

AI-generated and low-confidence observations go through human review. Reviewers can approve, reject, or request revision.

check_circle

Resolve

Approved observations become resolved values — the trusted numbers you see on food pages, with a final confidence score and source summary.

Where Data Comes From

verified

Government Databases

Highest trust

Primary data from USDA FoodData Central, national food composition databases, and peer-reviewed government nutrition references. These are the gold standard for nutritional values.

Trust level: 90–100

science

Scientific Literature

High trust

Published research papers, clinical studies, and meta-analyses on substance retention, bioavailability, and food composition. Used especially for cooking-method adjustments.

Trust level: 80–95

factory

Industry Data

Moderate trust

Manufacturer-provided nutritional labels and industry food composition databases. Cross-referenced with other sources where possible.

Trust level: 60–80

auto_awesome

AI-Researched

Review required

When data is missing from primary sources, our AI agent researches scientific and government databases to fill gaps. All AI-extracted values are flagged for human review before being fully trusted.

Trust level: 40–85

Confidence Scores

Every substance value has a confidence score from 0 to 100. This tells you how reliable the data point is. On food pages, you see the average confidence across all tracked substances.

High90–100%

Directly from government databases or verified analytical data.

Good80–89%

Strong scientific evidence with minor estimation involved.

Moderate60–79%

Reasonable estimates from related foods, AI research, or calculated from components.

LowBelow 60%

Limited data available. Values are rough estimates and should not be used for critical medical decisions.

How Values Are Derived

Not all nutritional values come from the same method. Each substance observation is tagged with its derivation type so you know how the number was obtained.

biotechAnalytical

Measured in a laboratory from actual food samples.

calculateCalculated

Computed from known ingredient compositions using standardized formulas.

trending_flatEstimated

Inferred from similar foods or related entries in reference databases.

auto_fix_highImputed

Filled from a closely related food when no direct data exists.

auto_awesomeAI-Extracted

Researched by our AI agent from scientific sources. Always marked for human review.

The Review Process

Every substance value goes through a review cycle before becoming a trusted resolved value:

Pending — Newly imported or AI-researched. Awaiting review.

Approved — Verified by a reviewer. This value is trusted.

Needs Revision — Flagged for correction. The value may be inaccurate.

Rejected — Deemed unreliable. Will not appear in resolved values.

Your Feedback Matters

On every food page, you can flag inaccurate data or suggest corrections. Each submission links to the specific substance and variant, and our reviewers investigate every report. If you have a source URL or research to back up your correction, even better.

Questions about our data? Found something inaccurate?

Create an account and report it