Which Wearable Should You Trust When Two Disagree? Apple's Answer Now Changes Moment by Moment

A user lifts weights wearing an Apple Watch and AirPods Pro 3. Both measure heart rate. As the wrist bends under a bar, the watch’s optical sensor loses contact; the earbuds don’t. Apple doesn’t pick one device and stick with it. Its support page says “the highest confidence source in the moment will be used automatically” [1], and Apple has described the software as choosing whichever device produced the most accurate data over the preceding five minutes [2].

Most health apps, and the platforms they read from, still work the other way. Apple Health and Health Connect resolve conflicts between devices with a priority list that the user, or the app, sets once per data type [6][9]. Whatever sits at the top wins, all day, in every context. On 13 October Apple was granted a patent that describes the alternative in general form: score every reading for confidence, then choose or blend by that score, across watches, earbuds, rings, chest straps, scales and in-bed sensors [3]. The gap between those two approaches is where much of the inaccuracy in multi-device health data comes from.

How do the platforms choose today?

• Apple Health keeps a source order per data type, set in the Health app. HealthKit’s statistics queries use it to merge overlapping cumulative data such as steps, so totals match the Health app. Raw sample queries return everything, from every source, and leave the choice to you.
• Health Connect keeps a user-controlled app priority list. Its aggregate() API keeps only the highest-priority app’s data where sources overlap; readRecords() returns all of it.

We covered these mechanics in detail in how to deduplicate health data across wearables. The point here is what they share: the ranking is fixed per metric. If the watch is first for heart rate, it is first during sleep, during a run and during a deadlift, whether or not it is the best device in the room at that moment.

Which device should win, and when?

Accuracy is not a property of a device. It is a property of a device, a measurement and a moment. The evidence makes that concrete:

Read down the second column and no single device wins every row. A fixed priority list therefore guarantees that some readings come from the wrong device: put the chest strap first and its gaps during the day go unfilled; put the watch first and you overwrite the strap during the one session it was bought for.

What Apple’s patent changes

The patent, US 12,758,110, “Dynamic multi-device physiological measurement systems and methods”, was reported as granted on 13 October 2026 from a provisional application filed in July 2024 [3]. As described in coverage of the grant, it does three things [3][4]:

• Scores each reading, not each device. Each device’s measurement gets a confidence based on signal quality and motion at that moment.
• Chooses or blends. The system either takes the highest-confidence reading or weights several readings by confidence, detected activity or motion, for heart rate, heart rate variability, blood oxygen and breathing rate.
• Turns off what it doesn’t need. A sensor whose reading isn’t being used can be switched off to save battery.

A patent is not a product, and most never become a named feature. But this one describes software linking devices Apple already sells, and it extends a behaviour Apple already ships with AirPods Pro 3 [1][2]. The direction is clear: the question is moving from “which device do you trust for heart rate?” to “which reading do you trust right now?”

How an app can do this today

Third-party apps don’t get Apple’s internal signal-quality scores. They do get enough context to replace a single list with rules that change by metric and moment:

What the platforms expose

• HealthKit: every sample has an HKSourceRevision (the app that saved it) and, usually, an HKDevice (name, model, manufacturer). Apple Watch heart rate samples carry HKMetadataKeyHeartRateMotionContext, which is sedentary, active or not set [8].
• Health Connect: every record has metadata.dataOrigin (the writing app), a recording method (manually entered, automatically recorded or actively recorded) and, for automatic and active records, a device type: watch, phone, ring, scale, chest strap, fitness band and others [9].

A context-aware priority, as a starting point:

for each metric and time window:
  candidates = samples from all sources in the window
  drop manual entries if any automatic or active sample exists
  if metric is heart rate:
      if window is inside an active workout:  prefer chest strap, then earbuds, then watch
      if window is overnight:                 prefer ring, then watch
      otherwise:                              prefer samples marked sedentary for resting values
  if metric is steps:
      prefer the wrist device; use the phone only where no wearable has data
  choose one source per window; never average two devices' sleep stages
  store the chosen source with the value

Three rules matter more than the exact ordering:

1. Choose per window, not per day. A day stitched from the best source for each window beats a day from the best device on average.
2. Choose, don’t average, unless you know why. Averaging two heart rate streams can help when both are good. Averaging a good one with one that has lost contact makes the result worse than either. Apple’s patent blends by confidence, which is exactly the information an average lacks [3].
3. Keep the provenance. Store which source produced each value, so a user, a clinician or your own support team can see why a number is what it is.

What still isn’t possible

Apps can’t see a confidence score for an individual reading; Apple keeps its own signal-quality assessment internal [1]. Much of the accuracy evidence comes from small studies, and some from the device makers themselves, such as the Oura study above [7]. And per-device rules need maintaining as devices change. None of that is a reason to keep a single list. Using the context already available gets most of the benefit of what Apple is building, today.

When two wearables disagree, the right answer is rarely “always trust this one”. It depends on what is being measured and what the user is doing. Apple’s own products have started answering the question moment by moment. Apps that read the data can, and should, do the same.