Prediction Methodology and Limitations

Current release methodology · Updated July 11, 2026

How Dela creates estimates today

Dela currently uses an interpretable estimator based on the cycle history and settings a person records. An experimental Bayesian predictor runs only in a private test environment for accounts that separately enable prediction improvement. It never chooses the dates shown in the app, and product analytics does not authorize its use. Dela will consider a newer predictor only after it has real outcomes from consenting users, reviewed calibration and safety, and a gradual monitored rollout.

Next period estimates

The estimator uses up to the recent cycle history available for that account. Depending on history depth, it combines recent values using robust measures such as a median, a trimmed mean, recency weighting, and a bounded trend. A median absolute deviation check reduces the influence of unusually long or short observations. With little or no history, Dela uses the cycle length the person reported and shows wider uncertainty.

The displayed range widens when history is sparse, variable, or marked irregular. More logs do not guarantee accuracy: illness, stress, medication, postpartum changes, perimenopause, hormonal contraception, missing entries, and ordinary biological variation can all move timing.

Estimated fertile days and ovulation

Fertile timing based only on a calendar is derived from an estimated ovulation day, not measured ovulation. The fallback assumes ovulation roughly 14 days before the next estimated period, bounded so it does not overlap implausible cycle edges. Dela marks a conservative window from five days before through one day after that estimated day. The boundary follows the American College of Obstetricians and Gynecologists' explanation that sperm may survive several days and an egg may survive about 24 hours after ovulation.

Logged cervical fluid and ovulation test results can add context about possible timing. A sustained shift in basal temperature is retrospective evidence only: it can suggest that ovulation may already have occurred, but temperature alone does not prove an exact day. Fever, sleep disruption, alcohol, travel, measurement time, and device differences can affect readings.

Dela pauses natural fertile window and ovulation estimates when pregnancy is active or when an active hormonal method makes the assumption about a natural cycle inappropriate. An unknown active method fails closed rather than guessing.

Pregnancy dating

If a person enters a due date from a clinician, Dela uses that date for its week counter. Otherwise, the general fallback adds 280 days to the first day of the entered last menstrual period. This does not account for every cycle pattern, assisted reproduction date, or ultrasound finding. A clinician's obstetric estimate should replace an app estimate; an ultrasound in the first trimester is generally more accurate for clinical dating.

What Dela has validated and what it has not

• Automated tests cover boundary dates, irregular and sparse histories, missing data, contraception suppression, overdue cycles, pregnancy mode, time zones, and consistency among the app, API, and shadow predictor.
• Synthetic fixtures and historical regression tests show that code behaves as designed. They are not evidence of clinical accuracy and are not presented as claims about sensitivity, specificity, pregnancy rates or accuracy across a population.
• Dela has not completed a prospective study to clinically validate the current release and does not claim that its fertile window or ovulation estimates are clinically confirmed.
• Accuracy will be evaluated using aggregate metrics only when users have consented and the minimum cohort size is met. Metrics include error in period start dates, interval coverage, abstention rate, fertile window overlap and performance across groups with different cycle variability. An experimental model will not be considered for use until this evaluation is complete.

References and further context

• ACOG: Fertility Awareness-Based Methods of Family Planning
• ACOG: Methods for Estimating the Due Date
• U.S. FDA: General Wellness Policy for Low Risk Devices

These references inform boundaries and limitations; they do not certify Dela or make its estimates medical advice. Questions about this methodology can be sent to hello@delacycle.com.