Reading Apple Health without fooling yourself
The database does not distinguish between a measurement and a guess. You have to.
Open Apple Health and you get a wall of numbers. Steps, resting heart rate, VO₂max, walking asymmetry, sleep stages, respiratory rate, cardio fitness notifications, headphone audio levels. They are presented identically: a name, a value, a chart.
They are not remotely equal. Some are counted, some are measured, and some are the output of a model several inferences deep. Learning to tell them apart is most of what it takes to use the thing well.
Three tiers
Counted. Steps, distance, flights climbed, stand hours. A sensor and a threshold. These are close to ground truth for their own definition — the caveat being that the definition may not be yours. Steps counted from a pocketed phone and from a wrist disagree, and neither is wrong.
Measured. Heart rate, heart rate variability, blood oxygen, skin temperature, weight from a connected scale. A physical quantity with a physical sensor. Accuracy varies with conditions — optical heart rate degrades badly during high-cadence or high-vibration activity — but there is a real measurement underneath.
Inferred. VO₂max, sleep stages, walking steadiness, cardio fitness levels, most “scores”. A model turning tier-one and tier-two data into something the sensor never saw. Useful, often well built, but carrying the assumptions of the population it was fitted on.
The single most common mistake is treating a tier-three number with tier-one confidence, because they look the same on screen.
Multiple writers, one database
HealthKit accepts data from every source you have connected. A watch, a phone, a scale, a third-party running app, a cycling computer, a sleep tracker. They overlap, they disagree, and Apple applies priority rules to decide what to show you.
This produces the two most common data mysteries. A step count that jumps when you install an app is usually double counting from two sources. A metric that goes silent for a fortnight is usually a device that stopped syncing, not a physiological change — and gaps quietly corrupt any average computed across them.
Before concluding anything from a trend, check that the collection was continuous over the window you are reading. It very often was not.
What to actually do
Pick a small number of metrics and follow them properly rather than glancing at thirty.
For most people the useful set is: resting heart rate, HRV, sleep duration and regularity, walking speed, and one load metric for whatever sport they do. Everything else is context you consult when one of those moves.
Then read them the same way each time. Establish a baseline over four to six weeks; compare against your own baseline rather than a published range; distrust any single day; and check the source and the continuity before you believe a change.
And keep the boundary firm, because the interface will not keep it for you. None of this is diagnosis. It is a record of how your body has been behaving, which is genuinely valuable to bring to a clinician and genuinely poor as a substitute for one.
The metrics library documents what each one is, how it is derived, and what the literature says it does and does not indicate.