Methodology

Two separate questions

BodySignal first looks for one repeated association in your existing Watch history: how later sleep compares with your typical sleep timing. It then offers a separate future experiment: whether avoiding caffeine after 2 PM is associated with different sleep outcomes over 14 assigned days. The caffeine experiment is not presented as the explanation for the historical sleep-timing pattern.

Health data used

The active sleep flow reads sleep analysis and timing, heart rate during sleep, and nightly HRV where available. Sleep duration, sleep start, and awake duration are derived from the imported sleep record. Analysis occurs on the iPhone.

Valid sleep nights

A night is excluded when the main sleep episode is missing, sleep start or end is missing, the interval is invalid, recorded sleep is under 3 hours or over 12 hours, the main episode spans more than 16 hours, or known Watch-wear coverage is below 50%.

Historical eligibility

BodySignal uses the latest 180 valid nights and requires at least 30. It also requires at least 10 nights in the later-sleep group and 10 nights in the typical-timing group with sleeping heart-rate data. If these gates are not met, the app shows progress instead of a finding.

Personal timing groups

Sleep starts are normalized across midnight. The later-sleep threshold is derived from the upper portion of your own sleep-timing distribution. The typical group is drawn from the middle range of your own distribution. This is a within-person comparison, not a universal bedtime target.

Historical effect

The primary comparison is the difference between median sleeping heart rate on later nights and typical nights. HRV, sleep duration, sleep start, or awake duration may appear as supporting comparisons when enough data exists. BodySignal shows no more than one supporting effect in the main reveal.

Evidence and uncertainty

Evidence grade is based on the available group sizes, effect magnitude, and uncertainty of the median difference. The finding always includes a limitation: it is an observed association, and other changes in routine, health, environment, or Watch wear may have contributed.

The 14-day caffeine protocol

Before the experiment begins, BodySignal creates a balanced on-device schedule with seven 2 PM caffeine-cutoff days and seven usual-routine days. Each adjacent pair contains one day of each condition. Usual-routine days do not require caffeine consumption.

Daily adherence

The user records whether the assignment was followed, not followed, or involved no caffeine that day. Illness and travel can be marked and excluded. The schedule, its original timezone, and every response are persisted on the iPhone.

Experiment eligibility

A final interpretation requires at least six valid adherent nights in each condition and sleeping heart-rate coverage on at least 70% of adherent nights. BodySignal waits for the final Health sync before calculating a result.

Possible results

• Possible benefit.
• Possible negative effect.
• No meaningful difference detected.
• Inconclusive.
• Not enough usable data.

A null or inconclusive result is not converted into a positive claim. The result includes group counts, primary effect, up to two supporting effects, evidence grade, limitations, and plain-language interpretation.

Source handling and duplicate protection

BodySignal resolves overlapping sleep records to avoid double counting and prefers Apple Watch sleep data when duplicate Apple Health sources describe the same episode. Third-party or incomplete records may be excluded when they cannot support a valid main sleep episode.

What BodySignal is not

BodySignal is not a medical device, diagnostic service, treatment tool, or causal inference system. Apple Watch measurements and BodySignal estimates can be incomplete or inaccurate. My result, not medical advice.