Transparency: Wellness Score

Wellness Score is a non-clinical index (0–100) designed for personal reflection. It combines physical signals (sleep/activity/recovery) with mental signals (mood + sentiment) to produce a single daily score and a transparent breakdown.

This feature is intentionally designed to run primarily on device (iOS), with explicit consent and user-configurable inputs.

1) Consent and user control (iOS)

Wellness Score is off by default. When you enable it, you choose which inputs to include: Apple Health summaries, sentiment from your text, or both. The app requires explicit acknowledgements: it’s not medical advice, results can be wrong, and the analysis runs on this device.

You can disable Wellness Score at any time. Apple Health permissions are managed by you in the iOS Health app. Wellness preferences are stored locally and scoped per account (so preferences don’t leak across sign-in states).

2) Inputs (what we use)

Physical inputs (optional) are derived from Apple Health (HealthKit), when authorized: sleep duration, steps, workout time, resting heart rate (RHR), and heart-rate variability (HRV SDNN).

Mental inputs (optional) come from data you explicitly create in the app: mood check-ins and your own text (journal entries, reflections, timeline notes, etc.). When enabled, the iOS app can also compute on-device sentiment from your authenticated chat/call history by downloading recent history through the same signed-in APIs used for history—then analyzing locally.

3) Sentiment analysis (on-device)

Sentiment is computed with Apple’s built-in NaturalLanguage sentiment scoring when available. Each text sample yields a sentiment value in [-1, +1], where -1 is most negative and +1 is most positive.

We aggregate multiple samples using a word-count-weighted average (bounded so very long texts don’t dominate). If language support is missing or text is too short/noisy, sentiment may be unavailable for that day.

4) Scoring model (transparent weights)

Wellness Score is composed as a weighted average of Physical and Mental subscores. Missing components are skipped rather than imputed; the app will show “low data” notes when needed.

5) Component details

Sleep is scored with a simple guideline-aligned curve: best around 7–9 hours (peak at 8), decreasing when sleep is very short, and slightly penalized when sleep is extremely long (often a proxy for poor recovery or fragmented sleep).

Activity combines steps and workout minutes with transparent targets (roughly 8k steps/day and ~30 minutes/day of exercise time). Either input may be missing; the score uses what is available.

Recovery combines RHR and HRV. When enough baseline data exists, we normalize to your personal baseline using a robust estimator (median + MAD) to compute a z-score, then map that to 0–100 with a smooth logistic transform. If baseline data is insufficient, we fall back to gentle absolute heuristics.

6) Privacy and data handling

Wellness Score is designed to be device-local: Apple Health data is read locally from HealthKit, and sentiment scoring is computed on the device. The feature does not require sending raw Apple Health metrics to the server.

If you enable conversation-based sentiment, the app may download your chat/call history (already stored on the server for normal product functionality) and compute sentiment locally. This is purely additive; it does not require sending any additional text to third-party model providers beyond the existing chat feature.

7) Limitations

HRV is not real-time and may update only when Apple Health records a new sample. Sleep and activity can be missing if a device isn’t worn or permissions are disabled. Sentiment can misread sarcasm and mixed emotions. Wellness Score is a personal reflection aid—not a medical device.