First of all, thanks @2e3s for the great work!

These gaps (1-3 seconds) amount to an error of about 30 minutes per day, depending on session duration and window switches. It makes it hard to track accurate times. If you set poll-time-window-seconds to higher values, the error per day quickly is in the range of hours.
Ideally, an event would never have length 0. For example, if poll-time-window-seconds=10 and we discover a window for just one heart beat, the event length should be 10s.
Thank you for considering a fix 😄