How the scores work

1. Inputs from Apple Health

With your permission, Pulse Coach reads these from Apple Health. They are recorded by your Apple Watch.

• Heart rate variability: RMSSD when Apple Health has at least 7 days of it in the last 60, otherwise SDNN
• Resting heart rate
• Sleep stages: asleep (core, deep, REM), awake and in bed
• Heart rate samples and workouts, including Apple's time-in-zone data when present
• Respiratory rate, blood oxygen and wrist temperature, used only to flag unusual readings

2. Your personal baseline

Each metric is compared with your own 60-day rolling baseline (mean and spread). The baseline needs at least 7 days of data before scores become personal. Until then, the app tells you it's still calibrating.

3. Recovery score (0–100)

HRV and resting heart rate are converted to z-scores (how many standard deviations you are from your baseline). Each component starts at a neutral 0.5 and moves 0.2 per standard deviation, clamped to 0–1, so a 2.5 SD swing spans the full range. A missing metric stays neutral. Score = 100 × (0.5 × HRV + 0.3 × RHR + 0.2 × sleep).

Bands: 67–100 high (push) · 34–66 moderate (train with intent) · 0–33 low (ease off). Rest mode overrides the call to recovery.

4. Exertion score (0–10)

1. Five heart-rate zones: your Apple Watch heart-rate zones where iOS provides them, otherwise estimated from max heart rate = 211 − 0.64 × age (Tanaka) by % of max.
2. Each heart-rate sample counts for the time until the next sample, up to 5 minutes, in its zone. Zone weights are 1, 2, 4, 8 and 16: each zone counts twice the one below.
3. Exertion = 10 × log10(1 + weighted minutes ÷ 90), capped at 10. Example: 45 minutes in zone 4 = 360 weighted minutes ≈ 7.0.

Target range from recovery: high 5–8 · moderate 3–6 · low 1–3. A Peak goal raises the ceiling by 1, and Recovering lowers both ends by 1.

5. Sleep score, debt and bedtime

• Sleep score: duration vs 8 h (40%) + restorative sleep, deep + REM vs a 40% target (35%) + efficiency, asleep ÷ in bed (25%).
• Consistency: bedtime spread of ≤30 min is best, ≥120 min is worst.
• Sleep debt: shortfall against 8 h over the last 14 nights, capped at 10 h.
• Bedtime target: repays 25% of debt tonight, at most 1 extra hour.

6. Habit impacts

For each journal tag with at least 3 occurrences, Pulse Coach compares average next-day recovery (and same-night sleep, when available) on days with and without the tag, expressed as a percentage. Impacts are correlations, not proof of cause.

7. Unusual readings

Resting heart rate, respiratory rate, blood oxygen and wrist temperature are flagged when the latest value is more than 2 standard deviations from your 60-day baseline. The flag is informational and never a diagnosis.

8. The AI coach

On iPhones with Apple Intelligence, a compact summary of your scores, baselines, recent workouts and top impacts (rounded numbers, label:value lines) is passed to Apple's on-device Foundation Models. The training call is fixed by your recovery score, and the model is told never to contradict it, never to diagnose, and to suggest seeing a doctor when something looks unusual. Without Apple Intelligence, or in a language it doesn't support, a rule-based coach uses the same scores.

9. Where it runs

Every step above runs on your iPhone. There is no account and no server, and the app makes no network requests with your data. Read the privacy policy.