Hi Alan,
I’m Torey, and I lead the Data Science team here at WHOOP. I wanted to reach out personally to thank you for sharing your concerns and to apologize for the frustration these HRV spikes have caused in your Recovery scores.
You are absolutely right — extreme outliers like this should not have such a strong influence on your Recovery calculation. We also know that extremely high HRV readings that are atypical for a member are often not a sign of better recovery, and it’s a current limitation of the algorithm that these events can both give you an overly green Recovery score in the moment and then throw off your averages for weeks afterward. I’ll be taking this feedback directly to my team, and we’re looking closely at how we can better filter and handle these readings in the Recovery calculation.
I also took a closer look at your data from the dates you mentioned. That night, your HRV readings were indeed quite elevated. We examined the underlying beat-to-beat intervals across the night, and they showed unusually high variability throughout the entire sleep period. I’m honestly a bit surprised your other devices reported significantly lower values in comparison.
Just to clarify our method — WHOOP doesn’t use a single HRV measurement for the night. (The explanation of calculating it exclusively during SWS in the blog post is a bit outdated and I’m working with Evan to ensure that it’s updated to accurately reflect how the algorithm works.) Instead, we calculate HRV in overlapping 5-minute windows (stepping forward every 30 seconds) throughout your entire sleep. We remove periods of wake, then take a weighted average of the remaining windows. The weights are higher for intervals that occur during likely slow-wave sleep as well as in intervals that occur later in the night, as HRV can change by sleep stage, and later periods tend to better reflect your overnight recovery. This method is designed to capture a stable and representative HRV value for Recovery, though in rare cases like this, unusually high variability can still impact the score more than we’d like.
I appreciate you bringing this to our attention — feedback like yours helps us improve both the science and the experience for all members. While I can’t retroactively delete the spikes, I can assure you that addressing these outlier effects is something we are actively working on and giving a high degree of attention.
Best regards,
Torey