How BioTrakk Numbers Work

Last updated: September 3, 2026

Every calorie tracker shows you numbers. Most don't tell you where they come from or how much to trust them. This page explains exactly how BioTrakk produces its estimates, correlations, and experiment verdicts — including their limits.

1. Where nutrition numbers come from

When you log a food, we resolve it through a chain of sources, most reliable first:

USDA FoodData Central — lab-verified reference data for generic foods (chicken breast, rice, banana).
Open Food Facts — community product database, strongest for European branded products.
AI estimation — only when neither database has a match. These estimates are marked with a ~ in your log so you always know which numbers are approximate.

Database and AI results are cross-checked against known calorie-density ranges per food category. When they disagree wildly, we use the value closer to the reference range — this catches errors like a beer being logged at 785 calories.

2. Photo portions are the weakest link — we treat them that way

No app can reliably judge portion weight from a photo; published research puts typical photo-AI calorie error at 15-30%, and hidden ingredients (oil, butter, sugar) are invisible to any camera. So BioTrakk:

asks a clarifying question when an unstated detail could change calories by 2x or more (fried vs grilled, dressing, milk in coffee) instead of guessing silently;
learns your personal portion bias from your corrections and applies it to future photo estimates;
uses anonymized, aggregated correction data across users to calibrate common foods.

Your corrections are the ground truth. Every "it was 200g, not 350g" reply makes your future estimates better.

3. Correlations are computed, not vibes

When BioTrakk says "your sleep drops after alcohol", that statement comes from statistics run on your daily data — Pearson correlation with p-values, tested across 0-3 day lag windows — not from an AI's general impression. We only surface relationships that are statistically significant (p < 0.05) with at least 7 days of paired data. The AI's job is to translate the math into plain language, and it is explicitly forbidden from inventing patterns the statistics don't support.

Correlation still isn't causation. That's why significant correlations come with a suggestion to test them properly — see experiments.

4. Experiments give honest verdicts

An experiment compares a baseline period against an intervention period. For each metric we run Welch's t-test on your daily values, and the report labels every change as a clear effect, a possible effect, or not distinguishable from chance. If nothing reaches statistical reliability, the verdict says so — we'd rather tell you "inconclusive, run it longer" than sell you a false win. If you don't have enough logged history when you start, BioTrakk first collects a proper baseline before the intervention begins.

5. What we don't claim

We don't claim photo-perfect calorie accuracy. Nobody can.
We don't provide medical advice — see our Health Disclaimer.
Correlations found in your data are hypotheses about you, not proven facts.

6. Your data is yours

Export everything as CSV at any time, delete your account and all data with one command, GDPR-first by design. See our Privacy Policy.