When sleep score and HRV disagree on the same night

Last night's sleep score read 63, 10.9% below the 7-day average of 70.7. Total sleep was 6.3 hours, with 0.2 hours of deep sleep and 1.2 hours of REM. By the usual read, that is a disappointing number to wake up to.

Overnight HRV told a different story. It measured 94.0ms, 9.4% above its own 7-day average of 85.9ms, and the third-highest reading of the last 14 nights. Resting heart rate held at 49bpm, close to its own 14-day average of 49.3bpm. Two of the three overnight numbers that get the most attention moved in opposite directions on the same night. None of these three numbers, on their own, settles whether last night was good or bad for recovery.

Two numbers, two different jobs

Sleep score and HRV measure different things, and a single night can move them apart. Put on the same 14-day scale, the gap shows clearly. Sleep score's z-score for last night was -0.46, ranking 11th of the 14 nights, inside the usual spread for this window. HRV's z-score was 0.82, ranking 3rd of 14, among the highest readings of the window. The correlation between the two series over those 14 nights was weak (r = 0.2, r squared = 0.04), so a move in one does not predict a move in the other.

What HRV-guided training studies found

The idea of training by a personal baseline has been tested directly. A 2007 study split 26 healthy, moderately fit men into two 4-week running programs¹. One group trained by daily HRV: easier sessions or rest were scheduled only when HRV fell below a reference value built from the 10-day mean and standard deviation, or when it showed a two-day falling trend. That group's VO2peak rose from 56 to 60 ml/kg/min. The other group followed a predefined plan and did not change significantly. The authors report that the change in VO2peak did not differ significantly between the two groups, nine against eight men, so this one trial does not by itself prove the baseline method beats a fixed plan. The first group acted on a baseline built from ten days, never one reading.

A larger trial points the same way, with a smaller claim attached. Forty recreational runners trained for eight weeks, split between HRV-guided and predefined schedules². The HRV-guided group ran fewer moderate and high-intensity sessions, 13.2 against 17.7, and still improved a 3000 m time by 2.1%, a result the predefined group's 1.1% change did not reach. Both groups raised their VO2max, by 3.7% and 5.0%. The paper itself describes the gap between the two training styles as small.

A 2020 meta-analysis pooled randomized trials of HRV-guided training in endurance athletes³. Both HRV-guided and control training raised VO2max. The HRV-guided effect was significantly larger, though the overall size of that advantage was small, and it depended on the athlete's level and sex: amateur and female subgroups showed the better results. The paper does not report confidence intervals in its abstract, so that detail is left out here.

A 2021 meta-analysis went further and pooled eight studies, 199 people, comparing HRV-guided training with predefined plans across several outcomes⁴. HRV-guided training came out ahead for vagal-related HRV measures. For resting heart rate, maximal aerobic capacity, aerobic capacity at the second ventilatory threshold, and endurance performance, the difference between the two training styles was not significant. The authors describe any fitness or performance edge, on the evidence to date, as small, and say how to set the baseline itself, which index, which recording position, fixed or rolling averages, is still being worked out.

Across these four papers, decisions were driven by a multi-day baseline, and the fitness or performance advantage over a fixed plan was usually small and sometimes not significant. That pattern sets the backdrop for last night's split between a dipped sleep score and a strong HRV reading.

Turning this into a decision

1. Read last night's number against your own 14-day average.
2. See one low sleep score as a single data point; a two or three-night trend in the same direction carries more weight.
3. Check sleep score and HRV side by side. With a weak correlation between them over 14 nights, each one still needs to be read on its own.
4. Save judgment calls about training load for the days when several numbers point the same way.

Does one low sleep score mean today's session should be easier?

On its own, one reading does not settle it. The 2007 and 2016 trials above built their decisions from a 10-day or similar baseline and still reported real gains from that approach.

What was scheduled for this day?

The plan called for a 43-minute recovery spin at an easy intensity, zone 1 at a 0.5 intensity factor. Whether that session took place is not confirmed in this review.