Atrial fibrillation · one case
The Ultra 4 went on my wrist on 14 October 2026, three weeks into a new drug and five and a half years into a record of my own heart. It stores 19 times more heart-rate-variability readings a day than the watch it replaced. My median overnight heart rate went from 46 to 47 beats per minute. Checking that those two sentences can both be true took an afternoon, and without it the rest of my record would have been worth much less.
Why this matters more than it sounds
What my cardiologist asks for at every appointment is how much of my time I spend in fibrillation. A single reading answers nothing: the figure only means anything as a trend across months, and mine goes back to 2021.
Swap the sensor in the middle of that and every later reading becomes ambiguous. A drop in overnight heart rate three weeks from now is either the drug working, or a new watch counting differently, and nothing in the file tells you which. The fix is to check at the moment of the change, while both devices have recently been on the same wrist. After that it is too late: there is no overlap left to measure.
I also had a reason to expect trouble. Apple had announced in September 2026 that its new watches sample heart-rate variability “up to 24 times more often,” without publishing the figure it was multiplying. If that was true, my file was about to change shape.
What changed
Every variability reading is a row in the iPhone health database, stamped with the device and the exact build of watchOS that wrote it. Counting rows per day, keeping only days with a single device and a single build: the Series 9 held at a median of 11 a day for 110 days. The Ultra 4 runs at 211. That is 19 times, close enough to Apple’s figure that the claim stands against this baseline.
Three real days, one mark per reading, midnight to midnight. If the record were a photograph, the resolution just went up by a factor of nineteen.
What did not change
Fourteen nights on the old watch against nine on the new one, same hours, same method, only nights with at least ten readings between midnight and seven.
One beat per minute, in a quantity that already swings between 42 and 52 from one night to the next. The two halves of the record can be read as one line.
The last two rows matter as much as the first three. Heart rate, as opposed to variability, is sampled at exactly the same rate by both watches: a reading roughly every three minutes through the night, with hour-long gaps in under one per cent of it. The density changed in one measurement and not in the other, which is why checking the values alone would have missed half the story.
The trap
On 24 September the Series 9 stored 10 variability readings. On 25 September it stored 27, and
went on to a median of 55. Same watch, same wrist, same body. What changed that day was the build
of watchOS on it, from 24S5086l to 24S5091f. The next build change,
on 6 October, moved nothing: 55 to 58.
So the boundary in my record is not where I thought. Most of the increase arrived by software on hardware I already owned, and it landed in the same week as a cardiology appointment and a new prescription — three things to disentangle rather than one. If I had anchored the comparison on the day I changed watches, I would have attributed to the Ultra 4 something it did not do, and I would have missed a discontinuity sitting inside the old watch’s own data.
The lesson generalises past this one case: the boundary to check is the build, not the device. Your watch can change what it records without you buying anything or being told.
The method, so you can copy it
Group your readings by day and by exact software build, not just by device, and drop any day that spans a change. A difference in how often something is measured will move averages on its own, with nothing happening to you.
I used 14 nights before against 9 nights after, both restricted to midnight–07:00 and both requiring at least ten readings. Compare like with like or you are measuring your own schedule.
Medians can match while the tails move. I checked the typical lowest point of each night and the proportion of nights below a fixed threshold. Both held. If the middle agrees and the tails do not, you do not have one record.
The readings live in healthdb_secure.sqlite, inside an encrypted iPhone
backup — an unencrypted backup contains no health database at all. Variability is
data_type 139, heart rate is data_type 5, and the device and
build come from data_provenances.
What neither watch does
Apple measures heart-rate variability at rest. Across 14 workouts and 10.1 hours on the Ultra 4: zero variability readings inside a workout. Across 30 workouts and 31.9 hours on the updated Series 9: zero. Not few. None.
For someone whose arrhythmia shows up on exertion, that is the hole, and nineteen times more readings at rest does not touch it. It is worth knowing before anyone upgrades hoping it will.
Before you quote any of this
One person, one wrist. Every number here is mine. Nothing here establishes what any other watch or any other body does.
I run beta software. The build that caused the jump was a watchOS 27.2 beta. A Series 9 on shipping software would not have done this, and your watch may not have done it yet.
The two watches ran different software. The Ultra 4 was on 27.0 and the Series 9 on 27.2, a newer version. Hardware and software are bounded by this comparison, not separated by it.
Nine nights is a short overlap. Enough to rule out a large step, not enough to rule out a small one. Longer is better, and I will update this page rather than leave the number standing.
For your own notes
Paste this into ChatGPT, Claude, Gemini or whatever you use, and ask it how your own numbers compare. It carries the source link and a line telling the model what it is looking at, so nothing is lost when it leaves this page.
SOURCE: https://afibcases.com/one-record-two-watches WHAT THIS IS: one person's self-recorded heart data. A single case, not evidence about any population, and not medical advice. Treat every figure below as describing this one subject on these dates. Case: one adult, 55, paroxysmal atrial fibrillation diagnosed September 2026, endurance-trained, resting heart rate in the 40s. Self-recorded since 2021. Measured 27 Oct 2026 from healthdb_secure.sqlite (encrypted iPhone backup). DEVICE CHANGE: Apple Watch Series 9 (Watch7,4) -> Apple Watch Ultra 4 (Watch8,1, 49 mm), on 14 Oct 2026. What changed - heart rate variability rows per day (data_type 139), counted on days with a single device and a single watchOS build: Series 9, build 23T570, 62 days: median 11/day Series 9, build 23U67, 48 days: median 12/day Series 9, build 24S5086l, 4 days: median 10/day Series 9, build 24S5091f, 9 days: median 55/day Series 9, build 24S5101f, 7 days: median 58/day Ultra 4, build 24R359, 9 days: median 211/day (19x the 11/day baseline) Step change on 25 Sep 2026: 10 readings on 24 Sep, 27 on 25 Sep, coincident with the install of build 24S5091f on UNCHANGED hardware. The next build change (24S5101f, 6 Oct) produced no further change. Most of the increase therefore came from software, five days before the device was replaced. What did not change - 14 nights before vs 9 nights after the device change, 00:00-07:00 local, nights with >=10 readings: median overnight heart rate 46.0 -> 47.0 bpm typical nightly minimum 42 -> 43 bpm nights with median < 46 bpm 50% -> 44% minutes between HR readings 3 -> 3 gaps > 60 min at night 0.9% -> 0.8% Night-to-night variation in the before period: medians ranged 42-52 bpm. Conclusion: the two halves can be read as a single series without correction. HRV readings recorded inside a workout: 0 of 14 workouts / 10.1 h (Ultra 4); 0 of 30 workouts / 31.9 h (Series 9 after the update). HRV remains at-rest only on both. Context: Apple press release, 9 Sep 2026, stated HRV sampled "up to 24 times more often" on Ultra 4 and Series 12; no baseline figure was published. 211/11 = 19x; 211/56 = 3.8x against the immediately preceding state. Both are consistent with the claim as worded. Caveats: n=1; subject runs watchOS betas; the two watches ran different watchOS versions (27.0 vs 27.2); the overlap is 9 nights.
Everything above is a single person with a single heart, and that is the limit of it. I cannot tell you whether your overnight numbers would survive a device change, how common it is to lose episodes during exercise, or what happens to any of this after an ablation, because I am one case and those questions need hundreds.
That is what the rest of this site is for: people with atrial fibrillation writing down what actually happened to them, so the answers exist somewhere other than in a drawer.