How Splash predicts
What the app does with your dates, how we test it, and where it stops being useful. No black box.
In short: Splash estimates your next period from the cycles you log, giving recent cycles more weight and showing a range rather than one date. It places ovulation about 14 days before the predicted period, or uses LH tests and temperature when you log them. In tests on synthetic cycles the real start fell inside the range 82% of the time for regular cycles. It is not contraception.
Last updated
On this page 9 sections
The short version
Splash forecasts how long your cycle will be. It learns from the cycles you log, gives recent cycles more weight, and shows a range rather than one date. When you log positive ovulation tests or a temperature rise, it uses those too.
Your next period
Splash starts from the cycle length you gave it, then updates that guess every time a period starts. A few things keep it honest:
- Recent cycles count more. A cycle from a year ago counts about a quarter as much as last month’s.
- One odd cycle doesn’t take over. If a cycle is far from your usual (illness, travel, stress), it gets less say, so one 45-day cycle doesn’t drag every future date.
- The range shows how sure it is. With little history or very variable cycles, the range is wide. It narrows as you log steady cycles. If you’ve told Splash you have PCOS or are in perimenopause, it starts wider and never counts you as “late”.
Ovulation and fertile days
The time from ovulation to your next period (the luteal phase) is steadier than the time before it. It’s around 14 days for many people, but anywhere from about 10 to 16 is common.
- With no tests logged, Splash places ovulation 14 days before your predicted period, and the fertile window as the 5 days before it, the day itself and the day after.
- When you log a positive LH test or a confirmed basal temperature rise, Splash uses it to place ovulation in that cycle, and moves your next period to match.
- Over several confirmed cycles, Splash learns your luteal length. It only moves away from 14 slowly: one or two readings nudge it, and it always stays between 10 and 16 days.
Watch and ring temperature
If you import night-time temperature from a watch or ring, Splash looks for a rise above your own recent baseline that lasts at least three nights. When it sees one, it can say “ovulation likely confirmed”, after the fact.
It never uses watch or ring temperature to predict fertile days ahead of time, or to move your next period. Skin temperature is affected by sleep, alcohol, illness and room temperature, and it only shows a rise once ovulation has already happened.
Patterns in your symptoms
Insights can show links like “headaches were about twice as likely the day after poor sleep”. Because it’s easy to find links that aren’t real, Splash is strict about what it shows:
- Same part of the cycle. It compares days in the same cycle phase. Cramps and poor sleep both happen around a period, but that doesn’t make one cause the other, so a link like that isn’t shown.
- Enough data. At least 30 days with both things logged, spread over 3 or more cycles.
- Correcting for chance. Splash checks many possible pairs, so some would look linked by luck. It applies a standard correction (Benjamini–Hochberg) across every pair it checked.
Even then, a pattern isn’t proof. Something else, like stress, can drive both.
Your own track record
Once you’ve logged 4 or more cycles, Insights replays Splash’s method on your past cycles, using only the cycles before each one, and tells you how often it landed within 2 days. For example: “Your last 6 period predictions were within ±2 days 5 times.” It describes how predictable your cycle has been, not a promise about the next one.
If you’re 40 or over, or in perimenopause mode, Splash may also point out when your cycle lengths have become more variable than before. That’s an observation worth mentioning to your GP, not a diagnosis.
How we test it
We test the method on synthetic cycle histories: computer-generated people whose “true” cycle lengths and ovulation days we know. That lets us measure errors exactly. It also means these numbers describe made-up data, not real people, and real cycles can be messier.
Next period (200 synthetic people per group, 14 cycles each, each cycle predicted from the ones before it). “Old” is a simple average of the last 6 cycles; “now” is the current method. Coverage is how often the real start fell inside the range shown.
| Cycle type | Average error, old → now (days) | Inside the range, old → now | Range width, old → now (days) |
|---|---|---|---|
| Regular | 1.98 → 1.76 | 59% → 82% | 3.3 → 5.8 |
| Irregular | 4.12 → 4.02 | 45% → 75% | 6.5 → 12.4 |
| PCOS-like | 11.35 → 11.44 | 30% → 77% | 8.8 → 32.6 |
| Perimenopause-like | 9.03 → 9.41 | 32% → 74% | 6.2 → 22.3 |
The current method is about as accurate on the day itself, and much more honest about uncertainty: its ranges contain the real date far more often, at the cost of being wider when cycles are irregular.
Ovulation (300 synthetic people per group, 12 cycles each, LH tests logged in about 60% of cycles). “14 days” always counts back 14 days from the predicted period; “learnt” uses each person’s learnt luteal length.
| True luteal phase | Average error, 14 days → learnt (days) | Fertile window held ovulation, 14 days → learnt |
|---|---|---|
| About 14 days | 1.93 → 1.93 | 71% → 71% |
| 11 to 16 days | 2.22 → 1.99 | 64% → 68% |
| 10 to 12 days (short) | 2.99 → 2.19 | 33% → 55% |
Learning the luteal length helps most for people whose luteal phase isn’t 14 days, and changes nothing for those whose is. Even so, the estimated window missed the true ovulation day in a third or more of cycles. That’s one of the reasons calendar estimates are not contraception.
What it can’t do
- It can’t know about things you haven’t logged, like a new medicine or a stressful month.
- It forecasts cycle length. It doesn’t detect pregnancy, conditions or menopause.
- After hormonal contraception, pregnancy or breastfeeding, cycles can take a while to settle. You can leave those cycles out of the averages in the app.
- If your periods suddenly change, stop, become very heavy or very painful, talk to your GP. Don’t wait for an app to tell you.
Questions people ask about Splash’s predictions
How does Splash predict my next period?
It starts from the cycle length you gave it, then learns from every period you log, giving recent cycles more weight and letting one unusual cycle count for less. It shows a most-likely day with a range around it, and the range narrows as you log steady cycles.
How accurate are Splash’s predictions?
We do not call them accurate. In tests on made-up cycle histories, the real start fell inside the range Splash showed 82% of the time for regular cycles, 75% for irregular, 77% for PCOS-like and 74% for perimenopause-like. Real cycles are messier, so treat every prediction as an estimate.
Does Splash use my watch or ring temperature to predict ovulation?
Only after the fact. If you import night-time temperature, Splash looks for a rise above your baseline lasting at least three nights and can say ovulation was likely confirmed. It never uses watch or ring temperature to predict fertile days ahead or to move your next period.
Can I use Splash’s predictions as contraception?
No. Splash is not contraception and not a medical device. In our tests the estimated fertile window missed the true ovulation day in a third or more of cycles for people with a short luteal phase. If you want to avoid pregnancy, use a method designed for that.
Test figures checked against the app’s own test run on 28 October 2026. See also: Predictions in Splash · Medical disclaimer · How we write health content · How accurate are period tracker predictions? · Best period tracker for irregular periods.