Independent publication
We test what training watches claim about you.
Your watch tells you a race time, a readiness score, a recovery status. We check those numbers against what athletes actually did, and publish the result either way. Garmin is where our data runs deepest; Coros, Polar, Suunto, Whoop and Oura get the same standard.
Original research
375,000 days of training data, and what they contradict
We ran a Garmin analytics service for a year. Before shutting it down and deleting the personal data, we computed aggregate statistics from the full dataset. Five studies came out of it; three are below. Everyone in the cohort is a self-selected Garmin owner — almost certainly fitter and more data-curious than average, a limit that applies to every number on this page.
- athlete-days of daily health data
- 375,261
- athlete-days of daily health data
- recorded activities
- 81,447
- recorded activities
- resting heart-rate mornings
- 94,000+
- resting heart-rate mornings
- scored nights of sleep
- 40,000+
- scored nights of sleep
- Race predictor
Every athlete we tracked slowed down harder than the formula says.
The textbook Riegel exponent is 1.06. Every athlete in the cohort implied a steeper one from 5K to marathon, median 1.131. Where a real effort could be matched to a prediction, the 5K predictions ran 22.7% optimistic at the median.
22,422 prediction-days · 48 matched 5K efforts
Read the study → - Training Readiness
Readiness told us almost nothing about how the run went.
Median within-athlete correlation between the morning score and running efficiency that day: 0.056. The positive control worked — readiness tracked yesterday’s load at -0.272, as it should by construction — so the null is a finding, not a broken method.
5,136 paired steady runs
Read the findings → - Recovery signals
Body Battery and HRV are largely one signal.
Body Battery correlated with overnight HRV at a median 0.674, and with sleep score at 0.683. For 97.6% of athletes the agreement cleared 0.3. Three tiles on the watch face, much less than three independent readings of your morning.
2,400+ same-day metric pairs
Read the findings →
Two more studies — VO2 max drift and sleep score versus performance — are in the research index. Cite any of it freely with a link back.
Free calculators
Five calculators. No account, no catch.
Threshold, zones, pace, aerobic ceiling, load — the numbers most training plans are built on. Every one runs on figures you type in yourself, in your browser, which is why the brand on your wrist makes no difference to the answer.
Start with these
Everything we have published →Does Training Readiness predict anything?
A correlation of 0.056, a working positive control, and what the score is actually good for.
Recovery scores compared: Garmin, Whoop, Oura
Three companies, three proprietary scores, and how much daylight there really is between them.
Connect your Garmin to ChatGPT or Claude
All four working routes, what each one hands over, and the one we would actually use.
How to export your Garmin data
Getting your own history out, in a format you can still open in ten years.
Reader-funded, in development
What Your Watch Isn't Telling You — the handbook.
The studies, assembled into one practical read: what each metric measures, when it lies, and what to do instead. $29 when it ships, nothing today, and the research on this site stays free either way.
You can check our work, including the parts that make us look bad.
About this site
Written by an engineer who spent a year building a Garmin analytics platform, trains six to seven hours a week, and got tired of taking the watch's word for it. We do not review hardware. We review the claims the hardware makes about you.
The coaching app that produced this data has closed — what happened to it, and to your data.