Make sleep algorithm accurate again

[burgermind]

The v4.68 Public Preview algorithm is aggressively over-smoothing data, masking wake events as "light sleep." Comparing one night: New Fitbit (9h, 84 score), Legacy Fitbit (8.5h, 77 score), and Garmin (8h, 48 score). The update ignores conscious wakefulness and physiological stress, resulting in artificially high scores that deviate from objective

[donnadean8]

It’s doing the opposite for me - making average sleeping HR 70, although it was only that for a brief moment and below 65 for the rest of the night. As a result sleep score has dropped about 10 for a similar nights sleep. Makes no sense and I no longer trust it. 

[merigens6]

I’ve noticed the same when wake periods get blended into light sleep, the score looks cleaner but feels less honest, and that kind of over-smoothing can make the data less useful, a bit like when HDO Box hides the small details that actually matter.