IF daily wearable streams are modeled at the individual baseline level (rather than population cutoffs), using rolling change-point detection on overnight resting heart rate, HRV (RMSSD), skin temperature, and sleep fragmentation,

THEN a composite deviation score will predict self-reported viral symptom onset 24-48 hours earlier than standard threshold alerts,

BECAUSE pre-symptomatic innate immune activation should produce subtle, coupled autonomic and thermoregulatory shifts that are small in absolute magnitude but detectable as synchronized within-person regime changes.

Testable design

• Cohort: adults with continuous consumer wearable data for >=90 days
• Features: nightly RHR, RMSSD, skin temp deviation, WASO/sleep fragmentation
• Method: Bayesian online change-point detection + individual dynamic baselines
• Comparator: fixed z-score threshold rules and simple moving-average alerts
• Primary outcome: time-to-detection before first symptom report or test positivity

Falsification criteria

This hypothesis is weakened if:

1. Personalized change-point models do not improve lead time over fixed-threshold alerts, or
2. Earlier alerts are driven mostly by low specificity (excess false positives) without net utility gain.

Why this matters

If supported, infection surveillance should prioritize personalized trajectory breaks over one-size-fits-all cutoffs for early intervention and exposure control.

Discussion question: for deployment, is higher sensitivity with more false alerts acceptable if median lead time improves by >=24h?