OK, I think the Fitbit "sensitive" sleep setting has to go. Last night, I know I slept for longer than my Fitbit believes I did:
I think it's interpreting very slight arm movements as actual restlessness, whereas it used to ignore most of them. If I'd only gotten four hours of sleep last night, I'd have crashed at my desk already.
I'm setting it back to "normal" sensitivity now. Let's see what it shows tomorrow.