Recovery metrics explained: what readiness scores measure, and how to use them.

By Raji Shah · Updated 2026-10-24

Every major wearable now hands you a morning score that claims to say how ready you are to train. Under the branding, they are built from the same few signals: overnight heart-rate variability (HRV), resting heart rate, sleep, and your recent training. Each brand weights them differently. Once you know what goes into a score, and how reliable each input is, you can tell when it is worth acting on and when to ignore it.

The building blocks

Our guide to HRV for runners covers the first of these in depth: what HRV measures, why one morning's reading means little, and what moves it.

What each platform's score is built from

This is how each brand describes its main readiness score in its own documentation, as of October 2026.

The biggest difference is what a score leans on. Some scores, such as Garmin's Training Readiness and COROS Recovery, lean heavily on the training your watch has recorded, so they drop after a hard session and recover over hours or days. Others, such as WHOOP Recovery and Google's readiness score, are built mostly from how your body looked overnight. That is why two devices worn together often disagree; our guide to wearing WHOOP and a Garmin together explains what to trust when they do.

Costs differ too. Garmin, Polar, COROS and Apple include their scores with the device. WHOOP is a membership, and without an Oura membership, the Oura app shows only the three daily scores and basic data. Google's support pages list no Premium requirement for the readiness score itself.

How accurate are the inputs?

Some inputs are much more trustworthy than others.

• Resting heart rate is solid. In a 2025 study of 13 adults over 536 nights, average error against a chest-strap reference ranged from under 2% for Oura to 3% for WHOOP, with Polar in between; the study left Garmin out of that analysis.
• HRV is usable as a trend. In the same study, overnight RMSSD from the Oura Ring 4 was within about 6% of the reference on average, WHOOP 4.0 within about 8%, an older Garmin fenix 6 within about 11% and a Polar Grit X Pro within about 16%. The 53-adult sleep-lab study in the next point found that most devices underestimated HRV in people with high values. Compare a device only with itself.
• Total sleep varies by device. In a study of 53 adults at a sleep lab that receives WHOOP funding, using older models such as the WHOOP 3.0 and the second-generation Oura ring, total sleep time from Oura, Polar and WHOOP was within about 12 minutes of the lab measurement on average, while Apple and Garmin overestimated it by about 40 minutes. In a 2025 study of 62 adults, some with suspected sleep apnea, two Fitbit models overestimated total sleep by 6 to 11 minutes on average, the Apple Watch Series 8 by about 20, the WHOOP 4.0 by about 25 and a Garmin Vivosmart 4 by about 38. Even the closest devices in both studies were off by about half an hour on a typical night. Wearables generally detect sleep well but miss much of the time you lie awake.
• Sleep stages are the weakest link. In the 53-adult study, agreement with the lab on light, deep and REM sleep was only 50 to 65% across devices. In the 62-adult study, every device except the two Fitbits misjudged deep sleep: a Garmin Vivosmart 4 overestimated it by about 45 minutes and the WHOOP 4.0 by about 30, while the Apple Watch Series 8 underestimated it by about 25. Treat nightly stage minutes as rough estimates.
• Breathing rate and temperature are best read as changes. In company-backed studies, WHOOP's breathing rate averaged within about 2% of a sleep-lab reference, though individual nights varied by about 7%, and Fitbit's method had an average error of about 3%. Skin-temperature sensors have been validated mainly for relative changes, such as the rise in the second half of the menstrual cycle, rather than for their absolute readings.

Sleep scores: what they reward

Sleep scores mix quantity with quality, but the recipes differ. Apple's sleep score gives 50 of its 100 points to duration, 30 to bedtime consistency over the last 13 nights and 20 to interruptions. WHOOP's Sleep Performance compares your hours with a sleep need that accounts for your Strain and sleep debt, and adds consistency, efficiency and sleep stress. Oura uses seven contributors, including total sleep, efficiency, timing and stage minutes. Garmin combines duration with quality measures such as stages, time awake, restlessness and stress during sleep. Google says most of its users average a sleep score of 72 to 83.

Given how rough stage estimates are, the parts of a sleep score worth acting on are the reliable ones: how long you slept and how consistent your bedtime is. Our guide to recovery tools explains why sleep beats every gadget for recovery.

Training load: what the ratios can and can't tell you

Most platforms now compare your recent training with your usual training. Garmin's load ratio divides acute load by chronic load, and calls 0.8 to 1.4 optimal. Polar compares a 7-day Strain with a 28-day Tolerance. Apple compares the last 7 days of workouts with the previous 28. Google sets a weekly target load by comparing the last week with the last month. COROS compares a 7-day Load Impact with a six-week Base Fitness.

These ratios are useful for spotting sudden jumps, but the research on them as injury predictors is weak. A 2020 analysis called the acute-to-chronic workload ratio "an inaccurate metric" and said that adjusting it to prevent injuries "remains a conjecture," and a review the same year said its limitations "should discourage its use." In one study of 435 runners, a higher ratio even went with lower injury risk. A 2025 study of 5,205 runners and 588,071 recorded sessions pointed to something simpler: injury risk rose when a single run was more than 10% longer than the longest run of the previous 30 days, while the week-to-week ratio showed no link. So watch your longest run as closely as your weekly load, and build both gradually; our guide on when to increase running mileage covers how.

Using recovery data to spot illness

The clearest wins for recovery tracking come when several signals move together. In one study of Fitbit users who later tested positive for COVID-19, 81% showed changes in heart rate, steps or sleep, starting a median of four days before symptoms, and resting heart rate also rose in most other illnesses studied. Respiratory-rate and temperature changes have also flagged infections, sometimes before symptoms, in studies run or funded by WHOOP, Fitbit and Oura. Apple's Vitals app alerts you when two or more overnight measures fall outside your typical range. If several vitals are off and you feel unwell, rest; our guide to running with a cold covers when it's safe to keep running.

How to use a recovery score

1. Learn your own normal. Give the device its baseline period, then judge each score against your usual range, not against an ideal.
2. Trust trends over single days. One low score after a normal day is usually noise. Several low days in a row, with resting heart rate up and sleep down, are worth acting on.
3. Check the obvious causes. Alcohol, a late hard session, travel, heat and short sleep explain many bad mornings.
4. When the score and your body disagree, go easy. Scores are built mostly from heart data and sleep, so they can miss sore legs, poor fueling or a stressful day. A green score isn't an order to train hard, and a red one isn't a ban on an easy run.
5. Keep the plan in charge. Use the score to move hard sessions by a day or swap them for easy running, which is how HRV-guided training worked in trials with runners, not to add extra intensity.

Bottom line

Recovery scores are a weather report, not a verdict. They combine HRV, resting heart rate, sleep and training load in different recipes, and their most reliable ingredients are resting heart rate, HRV trends and how long you sleep. Read them against your own baseline, act on clusters of signals, and let how you feel break the tie.

FAQ