The Data Was Always There

Here is a fact that will make you angry if you think about it for more than thirty seconds: until 1993, the National Institutes of Health did not require women to be included in clinical trials. This was thirty years ago. For most of modern medicine before that, the female body had been treated as a variation of the male one, too complicated, too cyclical, too hormonal to study properly. The assumption wasn't that women's bodies worked differently. The assumption was that the difference was noise.

Which is an interesting word, noise. It's what engineers call signals they don't understand yet.

Thirty years later, we still don't have continuous, noninvasive hormone tracking. We have pregnancy tests that tell you something already happened. We have ovulation strips that require you to pee on a stick every morning and squint at lines that may or may not be darker than yesterday's lines, a process that transforms hopeful adults into forensic analysts of their own urine. We have apps that predict your cycle based on averages, which is a bit like predicting tomorrow's weather based on the historical average temperature for your city, technically possible but missing the point entirely.

Real-time hormone data? That required blood draws. Lab visits. Waiting. Insurance negotiations. More waiting. Results that told you what your hormones were doing three days ago, which is useful the way a weather report from last Tuesday is useful.

The consequences of this exclusion ripple forward in ways that are both statistically significant and personally devastating. Seventy percent of PCOS cases go undiagnosed. The average woman with endometriosis waits seven years, seven years of being told her pain is "normal" or "stress" or "in her head", before receiving a diagnosis. Heart attacks in women are missed more frequently because the symptom checklist was developed on men. The list goes on, and on, and on.

We started with a different question: What if the data was already there, continuously broadcast by the body, and we just hadn't built anything capable of listening?

The Signal in the Noise

Wearables have been collecting biometric data for over a decade now. Heart rate. Skin temperature. Heart rate variability. Sleep stages. Motion patterns. Steps. Billions of data points streaming continuously from millions of wrists, generating datasets of a scale that would have seemed absurd to researchers even twenty years ago.

We've built an entire industry around quantifying the self, and yet the most fundamental biological rhythm in half the population, the menstrual cycle, the hormonal fluctuations that affect energy, mood, cognition, metabolism, immune function, and athletic performance, remains invisible. A woman's resting heart rate varies by two to four beats per minute across her cycle. Her body temperature shifts by half a degree. Her HRV patterns change. Her sleep architecture changes. Her recovery patterns change. These aren't subtle effects if you're looking for them. They're quite pronounced, actually.

Most wearables treat these variations as exactly what earlier medicine treated them as: noise. Background fluctuations to smooth over. Inconvenient variability that makes the analytics messier. The algorithms weren't built to see cyclical patterns. They were built to find trends, and when the trend is itself a cycle, the algorithm tends to get confused and average everything out.

We looked at the same data and saw something else. Those daily variations weren't obscuring a signal. They were the signal. The body was already broadcasting its hormonal state, continuously, through a dozen different channels at once. The problem wasn't the data. The problem was that nobody had built a receiver.

What the Body is Actually Saying

To understand what we're measuring, you have to understand a bit about what hormones actually do. Hormones are not abstract. They are physical molecules that do physical things to your body.

Progesterone is thermogenic. This is not a metaphor. When progesterone rises after ovulation, it binds to receptors in the hypothalamus, your body's thermostat, and physically turns up the set point. Body temperature increases by 0.3 to 0.5 degrees Celsius. This is why basal body temperature tracking works at all, and also why it only tells you ovulation happened after the fact, which is a bit like a smoke detector that alerts you to fires that have already burned out.

Estrogen affects the cardiovascular system. As estrogen rises in the first half of the cycle, it causes vasodilation, subtle changes in blood vessel tone that ripple through to heart rate patterns, heart rate variability, and blood pressure. Studies have documented heart rate variations of two to four beats per minute across the cycle, with distinct patterns that correlate with specific hormonal phases.

The autonomic nervous system responds to all of it. Heart rate variability shifts as the balance between estrogen and progesterone changes. Higher estrogen tends to correlate with higher HRV. The luteal phase, when progesterone dominates, has its own distinct HRV signature. Even electrodermal activity, the electrical conductance of your skin, follows hormonal patterns.

The body doesn't broadcast its hormonal state once a day when you take a temperature reading. It broadcasts continuously, through multiple channels, all the time. The question was never whether the signal existed. The question was whether you could build something sensitive enough to detect it, and smart enough to interpret it.

Building the Receiver

Most wearables capture a handful of metrics with consumer-grade sensors optimized for battery life and cost. That's perfectly adequate for step counting. It's less adequate for detecting the subtle physiological variations that correlate with hormonal changes, which exist at the edge of what current sensors can reliably measure.

We built a different kind of sensor stack. Ten distinct biosensors capturing simultaneously: proprietary optical sensors for hormone monitoring, skin temperature with clinical-grade precision, electrodermal activity, motion in multiple axes, and 100+ other mechanistic biomarkers. The goal wasn't to measure one thing really well. The goal was to measure many things well enough that their combination would reveal patterns invisible to any single metric.

But sensors alone don't solve the problem. The human body is complicated. Sleep affects your metrics. Stress affects your metrics. Exercise, caffeine, alcohol, illness, ambient temperature, whether you ate breakfast, how hydrated you are, what you did yesterday. Everything affects your baseline. A model that only looks at raw values would see noise everywhere, because in some sense it is noise everywhere. The signal is in how the noise changes.

This is where sensor fusion becomes essential. Not just collecting multiple data streams, but combining them in ways that extract signal from noise. The insight is that hormones affect multiple systems simultaneously. Your temperature might be elevated because you had wine last night. But if temperature, heart rate patterns, HRV, electrodermal activity, and over hundred biomarkers all shift together in specific ways, that's not wine. That's progesterone. The body leaves fingerprints across multiple channels at once, and the fingerprints are consistent enough to identify.

Think of it like recognizing a friend's voice in a crowded restaurant. Any individual word might be lost in ambient noise. But the overall pattern, the rhythm, the cadence, the particular way they emphasize certain syllables, that's distinctive enough to pick out even when you can't hear every word clearly. We're doing something similar with physiological data. The individual measurements are noisy. The pattern they form together is not.

Forty Women and Five Thousand Days

The Clair prototype was tested on over 40 women across 127 complete menstrual cycles, generating more than 5,000 days of continuous physiological data. Participants represented real-world diversity: 71% with regular cycles (25 to 35 days), 29% with irregular cycles; ages 18 to 45; BMI range of 18.5 to 35; and representation across Monk Skin Tone shades I through X, which matters because optical sensors can behave differently on different skin tones, and we needed to know if ours did.

Wearable predictions were compared against urine-based hormone measurements, giving us ground truth to validate what the models predicted against what was actually happening hormonally.

Worth noting: the sensor stack has evolved substantially since that prototype. What we validated was proof of concept, the demonstration that this approach works at all. What we're building now incorporates everything we learned, with improved hardware, refined algorithms, better signal processing, and the world's first female biology world model.

The Numbers

94.10%

cycle phase classification accuracy

The model achieved 94.10% accuracy in classifying cycle phase from wearable sensor data alone. No blood draws. No urine tests. No user input required. Just the continuous physiological signals the body produces anyway.

That's the headline number. But accuracy isn't uniform across the cycle. Some phases have more distinctive signatures than others:

Menstrual phase: 96% sensitivity, 98% specificity
Follicular phase: 92% sensitivity, 95% specificity
Ovulatory phase: 93% sensitivity, 97% specificity
Luteal phase: 94% sensitivity, 96% specificity

The consistency across phases matters. The model doesn't just work for the easy parts of the cycle. It maintains high performance throughout, including the transitions between phases where the hormonal picture is most complex.

For fertility tracking specifically, we focused on detecting the LH surge, the hormonal event that triggers ovulation. The model detected LH surges with 87% sensitivity and timing accuracy within 1.2 days. That's approaching the performance of daily urine testing, except there's no testing involved. No strips to buy, no remembering to test at the right time, no squinting at lines. The device just knows.

We also built models to track progesterone patterns, which confirm that ovulation actually occurred. This matters because you can have an LH surge without actually ovulating (anovulatory cycles are more common than most people realize, especially as you get older or during periods of stress). Tracking the progesterone rise provides confirmation that the egg was actually released, not just that your body tried to release one.

What This Actually Means

Numbers are abstract. Let me make them concrete.

It means that when you wake up exhausted after eight hours of sleep, and your fitness tracker says your recovery is "excellent," Clair can tell you: your progesterone rose significantly overnight. That's why you're tired.

If you're trying to conceive, you currently have two options. Option one: pee on ovulation strips every morning, starting several days before you think you might ovulate (which requires already having some sense of when you ovulate), continuing until you see a surge, then having sex within the next 24 to 36 hours and hoping you timed it right. Option two: track basal body temperature every morning before getting out of bed, which will tell you that you ovulated yesterday, which is useful for pattern recognition over multiple cycles but not particularly useful for this cycle.

With continuous hormone inference, you get a notification: "LH rising. Fertile window opening." You know your body is approaching ovulation before it happens. You don't have to remember to test. You don't have to interpret ambiguous results. You don't have to set an alarm to take your temperature before moving. You just live your life, and the device tells you what you need to know when you need to know it.

If you're an athlete, you've probably noticed that your performance varies across your cycle, but you may not know why or how to train around it. Estrogen affects muscle repair and glucose utilization. Progesterone affects thermoregulation and breathing. There are real physiological reasons why high-intensity training feels easier in your late follicular phase and recovery feels harder in your luteal phase. Continuous tracking lets you see these patterns in your own data and adjust your training accordingly, instead of pushing through days when your body is genuinely less capable and holding back on days when you could push harder.

If you have irregular cycles, calendar-based predictions are essentially useless. Your cycle might be 26 days this month and 42 days next month, and no app averaging your historical data is going to predict that. But continuous monitoring doesn't predict based on averages. It tracks your actual cycle as it actually unfolds, adapting to whatever your body is actually doing rather than assuming you're the statistical mean.

And if you've ever just wondered why you feel the way you feel, why some days your energy is high and other days you're exhausted for no apparent reason, why your mood shifts in patterns you can almost but not quite predict, continuous hormone data provides answers. Hormones affect mood, energy, sleep quality, cognitive function, appetite, pain sensitivity, and a dozen other things you experience every day. Knowing your patterns turns mysterious variability into understandable variability. That's not nothing. That's actually quite a lot.

What Comes Next

Our internal validation proved the concept. Continuous hormone inference from wearable sensors is not a theoretical possibility. It works. The data was always there, and we built something that can read it.

But internal validation is just the beginning. We believe the science should speak for itself, which means independent validation, not just us testing our own work. Starting in May 2026, we're launching an independent clinical trial through the Stanford Gladstone BeeHive program. This study will validate our production hardware and models under rigorous clinical conditions, with independent oversight and peer-reviewed publication of results. Not because we have to. Because the bar should be high for claims about understanding the human body, and meeting a high bar requires letting others check your work.

I want to be honest about what Clair is and isn't. Clair is not a diagnostic device. It doesn't tell you that you have PCOS or endometriosis or luteal phase deficiency. What it does is give you information, continuous, personalized, previously invisible information, that you can bring to conversations with your healthcare provider. It closes the data gap, not the care gap.

The care gap requires systemic change: more research on women's health, better training for providers, healthcare systems that take female-specific conditions seriously. Clair can't fix that. But Clair can make sure that when you walk into a doctor's office, you're not walking in with nothing but symptoms and frustration. You're walking in with data.

I keep coming back to this phrase because it captures something important about what we're doing.

We didn't invent the signals that Clair reads. Your body has always produced them. The temperature shifts were always there. The heart rate changes were always there. The patterns were always there, cycling and fluctuating and telling a story about your hormones to anyone who cared to listen.

What was missing was the technology to listen. The sensors sophisticated enough to capture the subtle changes. The models smart enough to fuse multiple signals into coherent predictions. The willingness to take women's health seriously enough to build it.

We built it.

The data was always there. We just learned how to listen.

Clair is building the first noninvasive continuous hormone tracker. Reserve your spot and get 50% off at wearclair.com.

References

NIH Revitalization Act of 1993, Public Law 103-43. https://www.congress.gov/bill/103rd-congress/senate-bill/1
Yasmin A, Roychoudhury S, Paul Choudhary A, Ahmed ABF, Dutta S, Mottola F, Verma V, Kalita JC, Kumar D, Sengupta P, Kolesarova A. Polycystic Ovary Syndrome: An Updated Overview Foregrounding Impacts of Ethnicities and Geographic Variations. Life (Basel). 2022 Nov 25;12(12):1974. doi: 10.3390/life12121974. PMID: 36556340; PMCID: PMC9785838.
Nnoaham KE, Hummelshoj L, Webster P, d'Hooghe T, de Cicco Nardone F, de Cicco Nardone C, Jenkinson C, Kennedy SH, Zondervan KT. Impact of endometriosis on quality of life and work productivity: a multicenter study across ten countries. Fertil Steril. 2011 Aug;96(2):366-373.e8. doi: 10.1016/j.fertnstert.2011.05.090. PMID: 21718982; PMCID: PMC3679489.
Mehta LS, Beckie TM, DeVon HA, Grines CL, Krumholz HM, Johnson MN, Lindley KJ, Vaccarino V, Wang TY, Watson KE, Wenger NK. Acute Myocardial Infarction in Women: A Scientific Statement From the American Heart Association. Circulation. 2016 Mar 1;133(9):916-947. doi: 10.1161/CIR.0000000000000351. PMID: 26811316.
Brar TK, Singh KD, Kumar A. Effect of Different Phases of Menstrual Cycle on Heart Rate Variability (HRV). J Clin Diagn Res. 2015 Oct;9(10):CC01-4. doi: 10.7860/JCDR/2015/13795.6592. PMID: 26557512; PMCID: PMC4625231.
Barron ML, Fehring RJ. Basal body temperature assessment: is it useful to couples seeking pregnancy? MCN Am J Matern Child Nurs. 2005 Sep-Oct;30(5):290-296. doi: 10.1097/00005721-200509000-00004. PMID: 16132004.
Schmalenberger KM, Eisenlohr-Moul TA, Wurth L, Schneider E, Thayer JF, Ditzen B, Jarczok MN. A systematic review and meta-analysis of within-person changes in cardiac vagal activity across the menstrual cycle: Implications for female health and future studies. J Clin Med. 2019;8(11):1946. doi: 10.3390/jcm8111946. PMID: 31766119; PMCID: PMC6912292.
Stanford Gladstone BeeHive Clinical Trial Program. https://www.stanfordgladstonebeehive.com/