Methodology v3.2 · Independently funded · No affiliate revenue Methodology · Editorial
What’s the Best Calorie Tracking App
Methodology v3.2 · Glossary

Ground Truth

By Tomás Filipovic-Reyes, PhD, MSc · Updated

Ground Truth. Ground truth is the laboratory-established true value against which a measurement instrument's output is compared. For Methodology v3.2's accuracy battery, ground truth is the calorie content of weighed reference meals computed from USDA FoodData Central per-component values.

What is ground truth?

Ground truth is the reference value against which a measurement instrument’s output is compared. For physical sciences, ground truth is established by chemical analysis or mass measurement; for clinical trials, ground truth is established by placebo-controlled outcome measurement; for dietary-assessment instruments, ground truth is established by weighed dietary records or doubly-labeled-water protocols.

For Methodology v3.2’s accuracy battery, ground truth is the calorie content of weighed reference meals computed from USDA FoodData Central per-component values. The protocol is documented in our framework article.

Why USDA FoodData Central?

USDA FDC is the largest publicly-funded, peer-curated nutrient database for foods consumed in the United States. The Foundation Foods subset has documented chemical-analysis provenance for each entry. The SR Legacy subset is curated from older but reliable analyses. The Branded Foods subset reflects manufacturer-label values.

For ground-truth purposes, FDC entries are the most defensible single reference: they are publicly funded (no commercial COI), peer-curated (continuous quality control), and broadly used in the academic dietary-assessment literature (the standard reference).

Limitations

Ground truth via FDC is not perfect. Two limitations are worth flagging.

First, FDC values are computed from chemical analyses of representative samples. Individual food items vary around the database value (a particular banana may differ from the FDC banana entry by 5-10%). The variance is small relative to the variance introduced by tracker error but is not zero.

Second, FDC has US-skew. For non-US food cultures, ground truth via FDC is imperfect; we use peer-reviewed regional databases (UK COFID, Indian National Institute of Nutrition database, FAO compositions) where appropriate.

Both limitations are accepted in v3.2 because the alternatives are operationally infeasible (per-meal chemical analysis) or carry their own COI risks (vendor-curated databases). FDC remains the best available reference.

Related terms