Basis Peak sleep tracking: a multi-signal design a decade early

The most underrated thing the Basis Peak did, and the thing that almost no contemporary reviewer wrote about with any depth, was the way it tracked sleep. In late 2014, while the Peak was being benchmarked against the Fitbit Charge HR on steps and heart rate, Intel was quietly shipping a wrist device that inferred sleep stages from four different physiological signals running in parallel. Everyone else was using motion.

That is the entire story, in one sentence. The rest of this piece is what those four signals were, how the Basis app turned them into stages, what Basis Science did and did not publish to back any of it up, and how the approach reads now against the modern multi-sensor crowd — Oura, Whoop, Fitbit’s Sleep Profile, and Apple’s sleep staging in watchOS.

The four signals

The Peak’s caseback held an unusually dense sensor stack for a consumer wearable in 2014. The Basis B1, the company’s first-generation tracker from 2013, had pioneered the layout; the Peak inherited and tightened it.

Optical heart rate came from a green-LED photoplethysmography array on the caseback. Sampling was continuous, not duty-cycled like the early Fitbit and Jawbone bands of the same era. From the raw HR trace, the Basis firmware derived a beat-to-beat variability signal — proto heart-rate variability, though Basis never used the term — which gave the algorithm something more than mean rate to work with.

Skin temperature was measured by a thermistor in direct contact with the wrist. The output ran in roughly 0.1 degree Celsius resolution and was logged minute-by-minute. This is the signal that wrist trackers spent the next decade re-discovering: Oura built its Gen 1 ring (2015) and Gen 2 ring (2017) around it, the Apple Watch Series 8 added it in 2022, the Fitbit Sense added wrist temperature in 2020. The Peak had it on the wrist seven years before Apple.

Galvanic skin response, the third channel, used two stainless-steel contact pins on the caseback to measure skin conductance, effectively a perspiration proxy. This is the signal that has not made it into the modern mainstream wrist crowd. Whoop, Oura, and Garmin do not use GSR. The reasons are well-understood: wrist GSR is noisy, motion-prone, and very sensitive to contact pressure, and the inference value it adds over HRV plus temperature is debatable. Basis used it anyway, and the patent filings from 2012–2014 imply the GSR trace was important for detecting sleep-onset latency in particular.

Motion was a 3-axis accelerometer, the same sensor every wearable on the market had. The Peak fed accelerometer-derived movement into the sleep model alongside the other three; it did not, like every other 2014 wristband, rely on motion alone.

This is the core claim worth sitting with. The Peak was the first consumer wrist device to ship multi-signal sleep staging. Not the first device, period; sleep labs had been doing it on chest leads and electrodes for fifty years. The Peak was the first one you could buy at Best Buy and wear to bed without thinking. That is a meaningful first. We covered the broader sensor-quality picture in our wearable sensor accuracy explainer.

What the app actually showed you

The Basis iOS and Android apps displayed sleep as a stacked vertical hypnogram. Time ran top-to-bottom, the bars on the right were color-coded by stage, and a sidebar showed totals: total time in bed, total time asleep, time in each stage, and a count of “interruptions.”

The reported stages were five:

• Light sleep, the largest band by total time, the orange-ish bar.
• Deep sleep, in the darkest blue, typically clustered in the first third of the night.
• REM, in a lighter blue, the bar that grew toward morning.
• Awake, plain gray, for periods the algorithm thought you were lying still but conscious.
• Toss and turn, the fifth bucket, for fragmented motion inside what was otherwise an asleep period.

That fifth category is the part we have not seen anyone copy since. Modern trackers tend to collapse fragmented motion into “wake” or “restless” rather than expose it as a distinct band. The Basis approach, for what it is worth, matched our subjective sense of restless nights better than any single-axis “sleep score” we have used since. Whether that is because the algorithm was actually picking up something real or because the label was more honest about uncertainty is hard to say.

The web dashboard added a feature the mobile app did not: you could overlay the four sensor traces against the hypnogram. Pull up a night, and you could see the temperature curve dropping in the deep-sleep windows, the HR sagging into the low fifties, the GSR rising and falling with the temperature inverse. Whether or not the staging was correct (more on that below), the dashboard let you see the underlying physiology, which is something Oura, Whoop, and Fitbit still mostly do not.

The validation problem

Here is where the Peak’s story gets uncomfortable.

Basis Science, and later Intel, never published peer-reviewed validation of the Peak’s sleep staging. We have looked. There is no paper in Sleep, no paper in the Journal of Clinical Sleep Medicine, no paper in JMIR mHealth, no paper in Sensors from anyone with a Basis affiliation that compared Peak-derived stages against polysomnography in a controlled cohort. The closest the company came was a 2014 conference poster at the SLEEP annual meeting referencing an internal comparison study, the underlying data of which was never published.

What there was, at the time, was a stack of consumer reviews where the writer wore the Peak for a week and compared the hypnogram to their own sense of how rested they felt. That is, with respect, not a validation method. Engadget’s review noted the stages “felt accurate.” Wired’s coverage said the data “matched what I expected.” The Verge ran with it. None of those reviewers had a PSG, an EEG headband, or even an Apple Watch with concurrent ECG to compare against, because none of those tools existed in consumer form yet.

The few quantitative third-party comparisons that did exist used consumer-grade actigraphy and, in one case, a Zeo Sleep Manager headband, not full PSG. The directional findings, where they existed, suggested the Peak overcounted REM and undercounted deep sleep, which is the same direction of error that almost every modern consumer wearable still shows.

How it compares, then and now

Below is a short comparison of the Peak’s sleep stack against its 2015 direct competitor (the Fitbit Charge HR) and the modern multi-sensor leader at the wrist or finger (the Oura Gen 3, which launched October 2021 and is still in production alongside the Gen 4). The Oura Gen 4 changed the sensor housing and refined the algorithm but did not add or remove a sensor channel.

A few notes on the table. The Fitbit Charge HR did add sleep stages later — Fitbit rolled out “Sleep Stages” in firmware in March 2017, two and a half years after the Peak shipped them. Sleep Stages eventually became “Sleep Score” and then, on premium-tier devices, “Sleep Profile” with monthly animal-archetype reports. The Peak got there first, with more sensors, and got essentially no credit for it. We dug into the head-to-head more broadly in the full Basis Peak review.

The Oura column matters in a different way. Oura’s marketing for the Gen 1 ring in 2015 was almost word-for-word what Basis had been saying about the Peak the year before: heart rate, temperature, motion, multi-signal staging, recovery scoring. The Gen 2 in 2017 doubled down on temperature. The Gen 3 in 2021 added SpO2. By the Gen 4 in 2024, the formula was, broadly, “what the Peak was trying to do, on a finger instead of a wrist, with a working business model.” If you are picking between Oura and the alternatives today, our Oura Ring vs Whoop comparison walks through the modern equivalents in detail; the broader landscape is in our best sleep trackers guide.

What we think the Peak got right

The four-signal architecture was right. A decade of subsequent product development has confirmed it. The Apple Watch Series 8 and Ultra added wrist temperature in 2022. The Pixel Watch 3 added on-wrist skin temperature in 2024. Withings has used temperature on the ScanWatch line. Garmin uses HRV plus motion. Whoop uses HRV, temperature, and motion. Every one of these devices is now doing some flavor of what the Peak did in 2014, minus GSR.

The exposure of raw sensor traces was also right, and we wish more companies had copied it. Modern devices typically show you a derived score and hide the underlying physiology behind a paywall. The Peak’s web dashboard, for all its other flaws, let you see why the algorithm thought what it thought. That is the only way to develop a real sense of the device’s failure modes, and it is exactly the affordance that the modern consumer wearable category has moved away from.

The five-stage taxonomy, including the “toss and turn” bucket, was a small but honest concession to uncertainty. Modern sleep scores tend to round off the fragmented-but-mostly-asleep state into one of the cleaner buckets, which makes the scores look more decisive than the underlying data justifies. The full breakdown of what each stage actually is, biologically, is in our sleep stages explainer.

What we think the Peak got wrong

The biggest single mistake was the absence of any validation literature. Basis had the data internally and had the academic relationships to publish it. The company chose not to, possibly because the results were not clean enough, possibly because the engineering team was small and consumer-facing reviews were getting them all the press they thought they needed. Either way, it left the entire claim, that multi-signal staging was better than motion-alone, unsupported in the literature, and that gap has never been filled because the device is dead.

The second mistake, smaller but related: the Peak buried its sleep tracking inside the broader Habits scoring system in the app. Sleep was not the headline feature. Recovery, the metric Oura and Whoop later built businesses on, was a sub-element of a “Pattern points” system that was hard to read. Basis built the engine and then put it in a car nobody could find the keys to.

The third mistake is structural and not really Basis’s fault: the data lived on Intel’s cloud. When the cloud was shut down on December 31, 2016, every Peak owner lost their longitudinal sleep history, including the multi-year datasets that, in retrospect, would have been the most valuable thing any consumer wearable had ever generated. The data is gone. We have a small folder of CSV exports from one of our office Peaks. As far as we know, that is more sleep data from the Peak than any researcher outside Intel has access to in 2026.

Why this still matters

Consumer sleep staging in 2026 is not solved. Chinoy 2021 said it was poor; the follow-up literature in 2023 and 2024 has shown incremental gains but no breakthrough. Apple’s sleep staging in watchOS 9 and later is, by Apple’s own published methodology document, trained on a few thousand PSG-aligned nights and validated to perform “comparably to other consumer wearables,” which is a careful way of saying it is not better than the category. Whoop publishes its validation; the agreement numbers are honest but not great. Oura publishes its validation; same story. The Peak, with arguably the richest sensor stack of any of them, never published anything.

The category is a decade in and still stuck around 60–70% epoch-level stage agreement with PSG. The Basis Peak did not solve sleep staging. Nothing has, yet. What it did was show what a wrist-worn device with a serious sensor stack and an honest UI could look like, and then disappear before anyone built on it.

That is the version of the Peak’s story we keep coming back to on this site. The hardware was ahead of the science. The science is still trying to catch up.