According to your documentation the low sensitivity setting for Heart Rate Awake Detection is 1.25*median heart rate. Apparently my heart rate periodically spikes without really waking me, and SAA records this as 10 minutes awake. So it would be great to have two things:
Heart rate spikes of a single reading (so the readings either side are the sleeping heart rate) to result in shorter awake times (perhaps 1 minute, instead of 10)
A sensitivity settings below Low (perhaps called Very Low), which works at about 1.40*media heart rate.
In fact we already make sure to exclude any extreme readings. In 5 minute windows we collect HR with a high sampling and than do a median on them to mitigate any extreme values.
But the problem is that if the band of your smartwatch is not tight enough during sleep the watch will easily slip to a side for instance if you lay on it so there are positions where the overall figures may get high for the 5 minute period as a false positive…
Also HR may get very much elevated in intensive dreaming. We have tested our algorithm on PSG data with good results - especially we try to minimize false positive. But of course the quality of reading on as PSG is much batter than on consumer electronics smartwatches…