Why does Apple Health show more time asleep than my other tracker?

Often because InBed is being counted as a stage. It is not — it is the interval containing the stages, so adding it to them counts the same minutes twice.

What does Fitbit’s "restless" mean?

Asleep but moving. It comes from the older classic vocabulary used on devices without staging and on short sleeps. Counting it as awake shortens every classic night; treating it as light sleep claims a staging the device did not perform.

Is a device with four stages more accurate than one with two?

It is finer, which is not the same thing. A two-state device is making a coarser claim, not a wrong one, and the honest way to combine them is to prefer the finer staging only when it covers a comparable share of the night.

Sleep stages are not the same across devices

Four vocabularies, one of which overlaps the others. Merging them naively double-counts sleep, and the usual fix — trusting whichever device you trust more — throws away the finer measurement about half the time.

What each source actually emits

Source	Vocabulary	Shape
Apple, iOS 16+	Core, Deep, REM, Awake, InBed	Four stages plus a containing interval
Apple, pre-iOS 16	InBed, Asleep	No staging at all
Oura	deep, light, REM, awake, unmeasurable	Fixed 5-minute grid
Fitbit `stages`	deep, light, rem, wake	Variable-length intervals
Fitbit `classic`	asleep, restless, awake	Coarser; older devices and short sleeps
WHOOP	Per-stage totals	Not a hypnogram — no ordering
FIT `sleep_level`	Five values defined	See below

Apple's InBed is the one that breaks arithmetic

InBed is not a fifth stage. It is the interval that contains the stages. Summing InBed with Core, Deep and REM counts the same minutes twice and produces a night longer than the time between going to bed and getting up.

It is modelled as a session boundary rather than a stage — which is also why an export from before iOS 16, containing only InBed and Asleep, contains no staging at all rather than a two-stage hypnogram.

"Restless" is asleep

Fitbit's classic vocabulary has no light stage. restless means asleep and moving. Two wrong readings are common: counting it as awake, which shortens every classic night; and mapping it to light sleep, which asserts a stage the device never determined. Asleep-unspecified, flagged as estimated, is what it actually is.

Disagreeing about resolution is not disagreeing about facts

This is the part that has no standard answer, so here is ours. When a four-stage source and a two-state source both cover one night, they are not two claims about the same thing where one is wrong. The coarse source is making a weaker claim that the fine one is consistent with.

Resolving that by source priority — "I trust my ring more than my watch" — discards the finer staging whenever the trusted device happens to be the coarser one, which is a real and silent loss. So the rule is: prefer the source with more distinct resolved stages, and fall back to priority only when they tie — with an override so that a fine source covering twenty minutes cannot beat a coarse one that staged the whole night. Unknown and InBed contribute no resolution.

The FIT corpus contains no hypnogram at all

Worth knowing before assuming a Garmin file will contest an Apple night. The FIT profile defines five sleep_level values. Across 19,068 files and 443,374 sleep records in our corpus, exactly 2 appear: awake (404,450) and unmeasurable (38,924). Light, deep and REM: zero.

So those files are daily-wellness recordings whose sleep message only ever says "awake" or "could not tell". Whether Garmin devices that do stage sleep write it in sleep_level or somewhere else is a question 68,068 files do not answer.

What this means for you

Do not add InBed to stage totals.
Do not compare a stage total from a four-stage device with one from a two-state device and conclude that either is broken.
Do check which generation your Apple export is — see what is in an export.zip.
Do treat WHOOP's per-stage totals as totals: there is no ordering in them to compare interval by interval.