Why Does My Recovery Score Drop in My Luteal Phase?

It arrives on a schedule. Roughly the same stretch of every cycle, recovery slides, readiness turns amber, the app suggests taking it easy, and it lifts again a few days later. Then it happens again next month.

The usual reading is that something is going wrong for a week and a half at a time. There is a better one available: the number is behaving correctly, and it is being compared against the wrong thing.

What actually changes after ovulation

The second half of a cycle is dominated by progesterone, and progesterone has effects that any wearable will pick up.

Core body temperature rises, typically by a few tenths of a degree, and stays up until the next period begins. Resting heart rate rises with it, commonly by a few beats per minute. Heart rate variability generally falls over the same window. Respiratory rate often ticks up slightly. Sleep tends to fragment a little more.

Now look at what a recovery or readiness score is built from. Almost universally: resting heart rate, heart rate variability, temperature deviation, and some measure of sleep quality.

Every single input moves in the direction the score reads as worse. Not because anything has gone wrong, but because that is what the luteal phase does to those four signals in a healthy body.

The temperature row is the one that gives the game away. That same rise is the signal used to confirm that ovulation happened. One tool treats it as evidence the cycle is working. The other treats it as evidence you are run down. Same measurement, opposite conclusions, and the difference is entirely in what each one was built to expect.

The comparison is the problem, not the sensor

A score compares today against a baseline. If that baseline is a rolling average of the last two or three weeks, it has quietly mixed both halves of your cycle into one number.

Against a blended baseline, a normal luteal week has to read as a decline. It is being measured against an average that includes follicular days, when those same signals sit at their best. The score is not malfunctioning. It is answering the question it was asked, and nobody asked it what week you were in.

This is the same failure as a population reference range, one level in. A population range compares you to strangers. A blended baseline compares you to a version of yourself from a different part of the month.

What this costs, in practice

The cost is not that the number is annoying. It is that people act on it.

The common pattern is deloading, skipping sessions, or reading the amber as a warning for a week and a half of every cycle, which over a year is a substantial amount of training removed on the advice of an instrument that did not know what it was looking at. Some people conclude they are chronically overtrained. Some conclude the device is broken and stop wearing it, which throws away the useful signal along with the confusing one.

There is a quieter cost too. If a predictable monthly dip is occupying your attention, an unpredictable one has to compete with it for notice, and the unpredictable one is the one that carried information.

Compare like with like

Compare this luteal phase to your last luteal phase. Not to last week. That is the single change that makes the number readable, and it requires the phase to be an input rather than something you mentally subtract afterwards.

Expect the shape, then read the deviation from the shape. Once you know your own cycle costs you a given amount of HRV in the back half, the question stops being why it dropped and becomes whether it dropped more than usual, which is a real question with a real answer.

Give it a few cycles. One cycle is an anecdote. Three or four is a pattern you can hold a new month against.

Note the confounders that travel with the phase. Sleep, alcohol, salt, heat and training load all shift across a month for reasons that are partly behavioural. A phase-matched baseline handles the physiology, not everything that co-occurs with it.

What would be worth attention

A luteal dip that is on time and the usual size is a working cycle, and it is not a finding.

Worth a closer look: a dip that is much deeper than your own previous cycles, one that does not lift when your period arrives, a pattern that changes markedly from what it has been, or any of this alongside symptoms that concern you. Those are questions for a clinician. They are not questions for a score, and they are not questions for an app.

Where NUVARD sits on this

NUVARD treats cycle phase as an input to the reading rather than an anomaly inside it. A luteal week is interpreted as expected, not flagged as a decline, because the model knows which week it is looking at.

It supports cycling, hormonal contraception, perimenopause, menopause, pregnancy and not tracking, because these signals behave differently across those states and one mode cannot cover them. Pregnancy mode is deliberately conservative and gives no dosing or diagnostic guidance. While the history is still short the predictions are hedged and say so, rather than projecting confidence the data has not earned.

Underneath, it reads from more than 300 devices and apps and holds 53 physiological variables across 15 body systems against your own history. Forecasts are scored in the open as Held, Still open, or Missed.

NUVARD is wellness and health intelligence. It does not diagnose, treat, cure, prevent or detect disease.

NUVARD releases in full on the app stores on 21 August 2026, and everyone gets the same access on the same day. The waitlist at nuvard.ai is the launch notification list, and it gets you the download link first.