Noise thresholds below are conservative estimates from published self-measurement studies. We replace them with our own repeatability study before launch and will say so here.
How we measure
A number without its error is just an opinion. Here is where every number on your screen comes from.
1. Your tape is the anchor
The Tensile Index (0–100) starts from your waist-to-height ratio — one of the most reliable numbers you can take at home — plus waist-to-hip when you add it, and how you train today. Fixed rules, no randomness, no AI deciding numbers (engines tape-v1 in the quiz, scan-v1 in the app). It is a training index for prioritizing zones. It is not a body-fat measurement and not a medical assessment.
2. The camera, on your phone only
If you choose to use the camera, an on-device model (MediaPipe Pose Landmarker, loaded from our own server) finds 33 body points. We never read the face points. The capture only starts when seven checks pass — full body, distance, centered, level, arms slightly out, light, one person — and it takes the median of about a second of good frames. Only proportions leave your phone: shoulder width relative to your height, hip width, and how level shoulders and hips are. No photo or video is uploaded or stored.
Landmarks sit at joint centers, not at the skin, so they describe your frame, not body fat. They refine a few zones (shoulders, chest and back on the men’s path; posture on the women’s) and those zones get a wider True Range: +1 point with a perfect capture, up to +4 with a poor one.
3. Same input, same score
Inputs are rounded the same way every time (0.5 cm on the tape) before scoring. The app keeps a fingerprint of those inputs — and, with the camera, of the exact frame’s pixels — sealed with a key unique to your account. Send the same inputs again and you get the stored scan back, not a new guess. The Re-run button recalculates from scratch and shows that the result matches.
4. True Range
Every Index carries a range of at least ±4 points, wider when a measurement is missing or capture quality is low. Body fat from a tape (calculator) uses the US Navy formula with ±4 percentage points, in line with validation against DXA; individual error can be larger.
5. Real Change vs within noise
We call a change real only when it beats the minimal detectable change at 95% confidence (1.96 × √2 × typical measurement error):
| Waist (tape) | ±1.9 cm |
|---|---|
| Hip (tape) | ±2.2 cm |
| Weight (7-day average) | ±0.7 kg |
| Tensile Index | ±3 pts |
| Zone Index | ±3.9 pts |
| Strength (est. 1-rep max) | ±5.5 % |
Smaller differences are shown as “within noise”, with no color drama. Trends beat single readings: charts show the average of your last three.
6. Your plan
Plans come from a rule engine (plan-v1) built on the ACSM 2026 position stand on resistance training: every major muscle group at least twice a week, progressive overload across weeks 1–3, an easier week 4, and most sets stopping 1–3 reps short of failure. After 40, the women’s path adds weekly volume for legs and glutes and low-impact power work. An AI only writes the short explanation, from the plan’s parameters — never your name, measurements or photos. Guidelines-based, not endorsed by ACSM, and not medical advice.
7. Form Check
Camera angles have an error of about 7–14° (published studies of the same model), so Form Check counts reps and reports range as a category — full or partial — never in degrees, and only for movements filmed from the side. Nothing is recorded; only the count can be saved with your set.
Limitations we won’t hide
- Self-measured tape varies with placement and breathing. Follow the same protocol each time.
- The Index has no external “gold standard” to validate against; it is consistent and calibrated to your tape, not a lab result.
- Formulas and landmarks are less accurate at the extremes (very lean, very muscular, very high BMI) and have little validation for women over 40.
- Rep counting is less reliable for one-sided and complex movements.