Thirty days of primary data on how emotion actually moves through the menstrual cycle — and an interface that changes shape depending on where you are in it.
Menstrual tracking apps treat the cycle as a data-entry problem: dates in, forecasts out. But it is one of the few repeating human experiences that continuously alters mood, energy and self-perception — and almost no interface treats it as emotional.
I couldn't design for that from secondary research, because the clinical stereotype and lived experience disagree. So I ran the study first: four structured check-ins over 30 days, each timed to land a participant in a different cycle phase, rating emotional state, physical comfort, energy and mental focus 1–10, with open-ended questions after every rating.
Only three responses landed in the ovulation window, so it is reported as a range rather than a mean. One participant rated her emotional state 5/10 while describing herself as "strong and emotional today" — a single answer that exposed the limit of measuring emotion on one axis, and directly produced the app's most-repeated line of microcopy: "you may feel both strong and emotional today — that's not a contradiction."
More than half the group showed signs of low mood, and not one participant had discussed their cycle with a partner. Both findings became features rather than footnotes.
The phase colour system is not decoration — it is the product. Colour, gradient, microcopy tone, CTA label, check-in length and card density all shift with the phase. The 83% retention drop set the hard rule: the menstrual phase asks exactly one question.
Tap a mood, select triggers, start a coping tool, flip Safe Mode. Every screen is currently rendering the mauve menstrual palette because the mock user is on day 3 of 5.
Phase banner, a single mood ask, and the day-3 insight: “that's real — not weakness.”
Trigger chips exist so a user can attribute a low day to work stress rather than her cycle — the stress-over-phase finding, made operable.
Tools filtered by phase. In menstruation, rest and breathing surface first; movement sinks.
The app can present itself as “Wellness”. Built because 80% refused one question — disclosure is never assumed.
Eight active participants clustered in one age band is a signal, not a finding — the research paper treats the ratings as descriptive and refuses inferential claims. Ovulation has three data points and is reported as a range. Two participants with PCOS and PMDD were screened but never checked in, so the most clinically affected voices are missing entirely.
Run a second wave with automated check-in reminders — the retention collapse was caused by my own instrument, and fixing it would tell me whether the pattern holds or was an artefact. Then usability-test the phase-adaptive shift itself: does a user actually notice the interface changing with her, and does it read as care or as surveillance?
Three roles on one LMS shell, without building three products.