Product / decision room
Research and product logic · v0.1 · 23 July 2026

Estimate first.
Learn carefully.

Start with the strongest practical population estimate, expose uncertainty, validate the person’s data, and never silently change a calorie target.

Read the decision Full technical document Project index

Use the 2023 National Academies EER equations by default.

The adult EER equations predict total daily energy expenditure directly from the onboarding inputs and were developed from a large doubly labelled water dataset. Use Mifflin–St Jeor as a transparent cross-check, Cunningham only with credible fat-free mass, and measured resting metabolism as an expert override.

The calculation path.

Every stage either reduces uncertainty or protects the person from an unsupported recommendation.

01Check eligibility

Run safety boundaries before showing an automated target.

02Classify activity

Use Apple Health coverage or a confirmed manual category.

03Estimate maintenance

Calculate NASEM 2023 EER and a ±10% working range.

04Cross-check

Compare with Mifflin–St Jeor × representative PAL.

05Apply goal logic

Lose, gain, or recomposition with caps and floors.

06Explain and save

Store method, inputs, range, confidence, and version.

Five methods, different roles.

The scientific reference is not necessarily the right onboarding method. The product uses each method where its evidence is useful.

Reference

Doubly labelled water

Measures free-living total expenditure across roughly 7–14 days.

Validation benchmark
Override

Indirect calorimetry

Measures resting expenditure; still needs representative activity.

Expert input
Default

NASEM 2023 EER

Predicts the needed value—TDEE—directly from onboarding inputs.

Primary estimate
Cross-check

Mifflin–St Jeor

Transparent resting estimate multiplied by representative PAL.

Explainability
Athlete check

Cunningham

Useful only when fat-free mass comes from a credible measurement.

Optional context

Goal-specific starting logic.

Calculate from the unrounded maintenance estimate, then round the displayed target to the nearest 25 kcal.

Lose

Moderate deficit

Default pace: 0.5% body weight / week
  • Allow 0.25–1.0% per week.
  • Cap deficit at 20% of TDEE or 750 kcal/day.
  • Respect unsupervised intake floors.
  • Default protein: 1.6 g/kg goal weight.
Gain

Conservative surplus

Default: +5–10%, based on training experience
  • Allow 5–15% as the normal range.
  • Usually cap the start at +500 kcal/day.
  • Slow the target for advanced trainees.
  • Protein range: 1.6–2.2 g/kg/day.
Recomposition

Maintenance first

Default: estimated maintenance
  • Offer 5–10% deficit only when fat loss is explicit.
  • Require resistance-training context.
  • Judge measurements and performance, not scale alone.
  • Never promise simultaneous fat loss and muscle gain.

Safety before automation.

When automated prescription is inappropriate, neutral logging and manual targets remain available.

Do not auto-prescribe

Route to qualified care.

The algorithm supports generally healthy, nonpregnant, nonlactating adults aged 19 and above.

Automatic recommendation blocked by
  • age below 19;
  • pregnancy or breastfeeding;
  • current or target BMI below 18.5;
  • disclosed eating disorder or active treatment;
  • material renal, liver, thyroid, or surgical context;
  • medication-driven change needing monitoring;
  • a requested target below the product floor.

Adaptive calibration.

Do not adjust during the first 21 days. From day 22 onward, update only with a valid rolling window.

Minimum evidence

  • All 21 days complete or corrected with total intake.
  • At least 10 body-weight observations.
  • At least one weight in the first and final four days.
  • Suspicious and incomplete days reviewed.
  • No illness, travel, medication, or device gap making the period unrepresentative.
LowEquation plus manual activity, insufficient validated data, or cross-check gap above 15%.
MediumCredible health-data classification or one valid 21-day calibration window.
HighAt least two consistent valid windows with no major quality flags—not clinical certainty.

Recommendation contract.

Precision in calculation must never become false precision in the customer experience.

Every recommendation shows

  1. current and proposed targets;
  2. included and excluded dates;
  3. weight trend and selected pace;
  4. average intake on complete days;
  5. estimated TDEE and uncertainty;
  6. reason and confidence;
  7. accept, defer, reject, edit, and undo actions.

Launch validation gates

  • Zero formula fixture failures.
  • Zero changes from incomplete data.
  • Zero targets below the unsupervised floor.
  • Every recommendation exposes evidence and confidence.
  • Manual override and undo pass end-to-end testing.
  • Median maintenance error improves after two valid windows.