FEWER MEASUREMENTS · MORE TRANSPARENCY
What can fewer
sensors tell us?
Explore a frozen calorie-label model.
See its estimate, uncertainty and the tradeoffs behind both.
A research estimate, not a verified calorie measurement. Source labels and units are unverified. Start with a fictional example; no personal information is needed.
Read the evidence
Awaiting estimateYour estimate belongs here.
Choose the measurements you have, then estimate.
We’ll show the uncertainty alongside the number.
ESTIMATED RECORDED CALORIES
This fixed interval was calibrated on separate records. It is not a personal guarantee.
The interval extends below zero. Bounds are shown without clipping; negative expenditure is not physically meaningful.
*kcal/session is the notebook’s interpreted unit. The source measurement protocol and calorie-label construction are unknown.
THE INPUT TRADEOFF
Less to measure. Different errors.
Frozen XGBoost models · same 3,000 test records
Mean absolute error (MAE). Lower is better.
What the interval can tell you
90% is the nominal coverage target across comparable records, not a 90% probability that one person’s true calorie expenditure lies inside the interval. The primary model covered 90.7% of test labels overall, but only 79.5% for 21–30 minute sessions.
What this experiment cannot establish
Low benchmark error does not validate real-world calorie measurement. Simulated heart-rate noise with SD 5 bpm raised development MAE from 1.08 to 7.78. Unknown collection, label construction and possible participant dependence limit the results. Staying inside input ranges does not remove those limits.