How the forecast works

Ciclo does not pick one “certain” date. It builds a probability distribution over upcoming days: how likely the next period is to start on each of them. Then it summarises that into a most-likely day, a few window probabilities, and an uncertainty label (low / medium / high).

The same path exists in Python (openciclo.predict) and in the browser. The web app does not send your dates to a forecast API.

What is predicted

V1 predicts:

1. Next period start
2. Cycle length

It does not predict ovulation, a fertile window for contraception, or pregnancy.

Where the “typical length” comes from

With little history, the model leans on a published population prior: aggregate numbers fitted on the mcPHASES study (42 people in Canada, timing only). That is a convenience typical length, not a worldwide accuracy claim.

As you log more completed cycles, the forecast mixes your median length with that prior. The jargon for this mix is shrinkage. With few cycles, the prior still weighs a lot. With many, your own history dominates.

Utah / Creighton cycle charts are an external check. They are not mixed into the file the app ships (released_model.json).

Uncertainty

Uncertainty stays high when you have fewer than about three completed cycles, or when the spread is wide. Low needs more history and a tighter spread. The label is a summary, not a second scientific claim.

A day with no log is unknown, not proof that the period did not happen.

When the expected day has passed

The Calendar home card and the first card on Patterns keep the original most-likely day. Once that day has passed without a logged period start, they count the days late, show how long the open cycle already is, and quote today's chance from the initial forecast. They do not move the date forward. The Patterns chart then shows that same initial calculation, with the original day highlighted. The “from today” summary is hidden in that case.

Phases vs the forecast

The engine’s job is the next start. The words Follicular / Ovulation / Luteal on the home card are a separate calendar heuristic. See Calendar phases and “estimated”.

Turning forecasts off

Settings can hide forecasts. Your logs and completed-cycle statistics stay. Observed bleeding does not depend on the model.

Limits worth repeating

• Forecasts can be wrong, especially with irregular cycles or sparse logs.
• Do not use Ciclo as birth control.
• Optional symptoms, temperature, and lab-style signals may be stored in the diary. This release does not feed them into the forecast.