Most cycle apps describe their predictions with a single unexplained number, or no number at all. We think you deserve the same standard a scientist would demand: the metric defined, the dataset named, the comparison shown, and the weaknesses admitted. This page is that standard, and we update it whenever the engine's published numbers change.
1. The headline numbers
| Metric | Result |
|---|---|
| Median error (next period) | 1 day |
| Mean error | 1.82 days |
| Within 2 days | 75.6% |
| Within 1 day | 56.8% |
| Mean error, 3 days out | 1.55 days |
| Calendar method, mean error (same data) | 2.08 days |
2. How we measured it
We replayed Ferrabelle's real production prediction engine against a public research dataset: the Fehring/Marquette menstrual cycle study (159 women, 1,665 cycles, published by Marquette University). For each woman we predicted every period as of several days before it actually arrived, using only the data that existed at that moment — a method called leave-future-out walk-forward validation. 120 women and 3,282 predictions qualified for scoring. There is no way for the engine to "peek" at the answer.
Two things make this number honest. First, the engine was never trained or tuned on this dataset — it saw each cycle cold, the same way it sees yours. Second, the evaluation is deterministic and repeatable: the harness lives in our codebase and reruns on demand, so this page can always be regenerated from scratch.
3. Where the engine's assumptions come from
Before you have logged anything, Ferrabelle starts from assumptions grounded in published population research covering over 600,000 cycles (including Bull et al. 2019 in npj Digital Medicine and the Apple Women's Health Study), conditioned on age. From your first logged cycle onward, the engine personalizes to you — your own data always outweighs the population as it accumulates. If you connect a wearable, its nightly temperature and heart-rate signals feed ovulation detection where your device provides them; validated wearable-anchored accuracy numbers are coming (see Section 4).
4. Where we are weaker (told straight)
- Long and irregular cycles are harder. On the same public dataset, cycles in the regular range predicted at 1.79 days mean error; long cycles (over 35 days) at about 2.8 days. If your cycles are irregular, expect wider, honest confidence windows rather than falsely precise dates.
- The public dataset has no wearable temperature data, so the numbers above reflect the engine without its strongest signal. Wearable temperature is where ovulation detection sharpens most; we will publish wearable-anchored numbers when we have a hormone-confirmed validation set.
- Hormonal birth control changes what prediction means. If you are on hormonal birth control, ovulation-based prediction does not apply the same way; we are building an honest mode for that rather than pretending otherwise.
5. How predictions improve over time
Your predictions personalize from your first logged cycle and keep sharpening as your history accumulates. Separately, if you choose to turn on “Help improve predictions for everyone” in Settings (off by default), your cycle patterns contribute to anonymous group averages — computed only from groups of 20 or more consenting users, containing no personal identifiers — that improve first-time predictions for new users. The program activates once at least 20 users have consented; until then your choice is recorded and no data is pooled. Your individual data never leaves your account either way.
6. Privacy receipts (each one checkable)
- Zero advertising or third-party analytics scripts on this site or in the app. View source; check the network tab.
- No large-language-model or third-party AI service ever receives your health data. The prediction engine is transparent statistics running on our own infrastructure.
- Your data is encrypted in transit and at rest, and exports for free from Account → Export.
- Deletion is real: Account → Delete removes your data from our systems, with backups cycling out on a fixed schedule. See the Consumer Health Data Privacy Policy.
- Built to Washington's My Health My Data Act, one of the strictest health-privacy laws in the country.
7. What these numbers are not
Ferrabelle is not a medical device and not a contraceptive, and its predictions must never be used to prevent pregnancy or to diagnose or treat any condition. Prediction error is a population statistic: your cycles are your own, and any app can be wrong about them. If your cycle changes unexpectedly or worries you, talk to a clinician.
Questions about our methodology: hello@ferrabelle.com.