Bayesian Cycle Prediction

Most cycle apps default to a 28-day assumption and either ignore variance or correct after the fact. Soulwise uses Bayesian prediction instead — a statistical approach where the estimate of "what phase are you in today" is the posterior of two things: your historical cycle length (the prior) and the daily check-ins you have logged this cycle (the evidence).

Why Bayesian, not fixed

A fixed 28-day cycle is confidently wrong for most people. The clinical normal range is 21 to 35 days, individual variance month-to-month is large, and life events (stress, travel, sleep, training load) shift cycle length more than most people expect. A fixed-length estimator gives you a precise phase label when the underlying uncertainty is high.

Bayesian estimation gives you a phase label with an uncertainty band. When the band is wide (early in your data, or after a cycle that ran long), Soulwise says so on the phase card. When the band is narrow (after several stable cycles), the phase label is trustworthy.

What goes into the prior

• Your last six cycles' lengths, if available.
• A weakly-informative starting prior (28 ± 5) before any data exists.
• A drift term so older cycles weigh less than recent ones.

What goes into the evidence

• Your daily check-in chips and energy slider.
• Any logged bleed or spot.
• Optional symptom marks (header, cramp) if you log them.

The system does not use:

• Basal body temperature (we are not a fertility tracker).
• Symptom-based ovulation guesses (out of scope).
• Cervical mucus reports (out of scope).

What it does not do

It does not predict ovulation for contraception or conception. See non-medical-cycle-tracking for the full anti-claim posture.

It does not give a single-day pinpoint when uncertainty is high. The phase card shows a 2- to 4-day window with confidence shading.

It does not learn from anyone else's data. Each user has their own Bayesian model, run on-device, against their own encrypted history.

Where it shows up

In the phase card in the daily ritual, in the recap letter, and as the input to cycle-weather — the short-term variance read compared to your stable baseline.