Insights & Prevention — how it works

1. No AI or machine learning

Nothing on the Insights or Prevention tabs is produced by a language model, a prediction model, or anything trained on data. Every chart, comparison and percentage is a fixed arithmetic formula — counts, rates, averages and simple two-group comparisons — applied to your own entries, recalculated fresh each time you view the tab. The phrase used throughout the tab is deliberate: these are correlations in your own logged history, not a clinical prediction and not a claim about what causes what.

2. The headache-rate trend and its shaded runs

Entries are grouped into consecutive 14-day periods starting from your first logged day. For each period, Migraine Today computes the percentage of days with a headache and the average intensity on those days. A 3-point moving average (this period and the one either side) smooths the line so a single unusual period doesn't read as a trend on its own.

The shaded “improving”/“worsening” stretches are detected the same way every time: a run starts wherever the smoothed rate moves by 20 percentage points or more from one period to the next, continues while it doesn't reverse by more than 2 points, and is only shown if the periods involved add up to at least 6 logged days — short of that, it's treated as noise, not a pattern.

Each run is labeled automatically: a worsening run is labeled with whichever of your trigger tags appeared most often during it (only if it appeared at least twice); an improving run is labeled with exercise or supplement logging if either was present on more than 30% of days in that stretch, and falls back to a generic label otherwise.

3. “What's worked for you” comparisons

Four fixed comparisons are run against your history: exercise days vs. rest days, 6+ hours of sleep vs. under 6, days you took a preventive supplement vs. days you didn't, and low-stress (2 or below out of 5) vs. high-stress (4 or above) days.

A comparison only appears once both groups have at least 3 logged days. It's only called out as meaningful if the headache rate between the two groups differs by 15 percentage points or more, or — failing that — if average intensity on headache days differs by at least 0.3 on the mild/moderate/severe (1–3) scale. Below those thresholds, nothing is shown for that comparison rather than reporting a difference too small to trust.

4. Which supplement helped which symptom

For every supplement category you've logged (magnesium, hydration/electrolytes, vitamin B2, and so on), Migraine Today compares the rate of each individual symptom on days you took something in that category against days you didn't — again requiring at least 3 days on each side. A pairing is only surfaced when the without-it rate is at least 15 percentage points higher than the with-it rate, i.e. only cases where your own history suggests it may have helped are shown; it doesn't report the reverse.

5. The symptom correlation matrix

This grid reads as same-day co-occurrence, not cause and effect. Each row is a cycle/menopause symptom you've actually logged at least once; each column is a migraine symptom you've actually logged at least once — nothing you've never recorded appears in the grid. A cell's percentage is: of the days you logged the row's symptom, what share of those same days also had the column's symptom. It says nothing about which one happened first or whether one caused the other.

6. Time-of-day pattern

For every entry with a recorded start time, the hour is converted into your account's home timezone and sorted into one of four six-hour bands (night, morning, afternoon, evening). The chart is a straightforward count per band — which part of the day your attacks have actually started most often, from your own history only.

7. Why some insights don't show up yet

Every comparison above has a minimum-sample rule for a reason: with only one or two days in a group, almost any percentage looks dramatic and almost none of it is reliable. The Insights tab stays mostly empty until you have roughly two weeks of entries, and individual comparisons or correlations appear only once they clear their own threshold — not because the feature is broken, but because there genuinely isn't enough of your own data yet to say anything trustworthy.

8. What's stored, and where

Nothing extra is stored for this feature — every chart is calculated on the fly from the same entries already saved in your diary, scoped to your account only, and nothing is sent to a third party or an AI service to produce it. See the Privacy Policy for retention and your rights under GDPR.