What the evidence actually says

Two things are true at once, and most content about cycle-based training only tells you one of them.

First: the average effect is small. The most-cited meta-analysis (McNulty et al., 2020, pooling 78 studies) found exercise performance trivially reduced during the early follicular phase — the days of the period itself — and rated the overall quality of evidence as low. Across women on average, phase explains very little.

Second: individual variability is large. The same review says so explicitly. Plenty of athletes reliably notice differences in energy, sleep, and how sore the same session leaves them, at consistent points in their cycle. A small average effect with big individual spread means the textbook curve is useless for you personally — but your own pattern, if you have one, is worth knowing.

That’s the honest frame: cycle data is planning context, like sleep or stress. Not a rulebook.

The four phases in 90 seconds

Menstrual — roughly days 1–5, counting from the day the period starts.
Follicular — from the end of the period up to ovulation.
Ovulatory — a window of about ±2 days around ovulation.
Luteal — the last ~14 days before the next period.

One non-obvious detail: the luteal phase is the biologically fixed part, at about 14 days. So ovulation is estimated backwards — cycle length minus 14 — not forwards from the last period. A 31-day cycle puts ovulation around day 17, not “mid-cycle.” Repsheet uses exactly this model: it starts from a default 28-day cycle and, after two or three logged periods, switches to your personal average (gaps over 60 days are discarded as missed logs, not treated as 60-day cycles).

How to actually use it

Log the start of each period. In Repsheet that’s one tap, with ±1 day buttons if you log late.
Give it two or three cycles before trusting any pattern. The app shows a confidence level and is upfront while it’s still guessing from the 28-day default.
Then plan loosely: put the heaviest exposures in windows where you historically feel strong, and keep permission to go lighter in windows where you historically don’t. If the day says otherwise, believe the day.

What not to do: rearrange a whole program around a hormone chart you read once. If you feel great on paper-”bad” days, train hard. Individual variability is the headline finding, not the footnote.

What Repsheet does — and deliberately doesn’t

The feature is off by default. If you turn it on, the input is exactly one kind of data: dates a period started. No symptoms, no temperatures, no fertility windows.

The output is the current phase, the day of cycle, and a confidence signal. If you’ve connected an AI coach over MCP (Claude Desktop or Cursor), it can read the phase through get_cycle_phase and weigh it in intensity suggestions alongside everything else in your log.

What it is not: a fertility app, a contraception tool, or a medical device. The app labels the model a statistical heuristic right in the UI, because that’s what it is. And if you’re on hormonal contraception, the underlying hormonal cycle is mostly flattened — phase-based planning has little meaning there, and no app should pretend otherwise.

Privacy, concretely

Period data is about as sensitive as personal data gets, and the app category built around it has a poor record — in 2021 the FTC settled with Flo Health after the app shared users’ health data with third-party analytics SDKs despite promising not to.

How Repsheet handles it:

Cycle data stays on your device by default. Syncing it to your other devices is a separate opt-in toggle, independent of turning the feature on.
No ads, no third-party analytics, nothing sold. There’s no revenue attached to this data because the app has no revenue — it’s free, and that’s the whole model.
Any entry can be deleted at any time, and the feature switched off.

Bottom line

If you don’t want this in your training log, leave the toggle off and nothing changes. If you do: log start dates for three cycles, then compare how your sessions actually went per phase — working-set PRs are the cleanest signal to compare across weeks. Adjust only where your own data shows a pattern, and treat everything else as a normal training week.