Last month, my smart ring told me I was under high stress at 7 pm on a Tuesday. I wasn't feeling "high stress" at all, instead, I was just running up the stairs because I had forgotten my laptop charger and needed to get to my Zoom meeting in six minutes. The same night, an hour later, I really was feeling stressed as I looked at a full inbox that I didn't want to read. My ring gave that moment almost the exact same score.That's when it clicked for me. These sensors aren't measuring stress. They're measuring a body that's revved up, and they have no idea why.If you build wearable software, health apps, or anything that touches biometric data, this matters more than a fun personal observation. It's a design limitation and it is the key to understanding the confusing numbers in your readiness or stress level.What Your Sensors Are Actually ReadingEvery consumer wearable claiming to track stress is leaning on one or both of two signals.Heart Rate VariabilityHRV is the tiny variation in time between each heartbeat. A heart that beats like a metronome, with exactly the same gap every time, is actually a sign of a stressed or under-recovered body. A heart with more variability usually indicates that the nervous system is able to breathe and shift easily between rest mode and alert mode. Most smart rings and smartwatches measure this during sleep and translate it into a recovery score.Electrodermal ActivityOther devices, such as the Fitbit Sense, also monitor electrodermal activity (EDA) which is also known as galvanic skin response (GSR). EDA monitors your skin's electrical conductance as you sweat more as part of your physiological response to stress. Since the sweat glands are regulated by the sympathetic nervous system, EDA measures the activity of your sympathetic nervous system, giving an indirect measure of your body's physiological arousal.Research shows that EDA, combined with HRV is one of the most reliable physiological measurements in terms of stress detection and is commonly used in stress monitoring in commercial wearables.This is what the two measurements have in common. Both measure your autonomic nervous system, your sympathetic division, responsible for fight or flight response. And your sympathetic nervous system does not have a separate circuit for "I am afraid" or "I am excited" or "I just completed twenty burpees".Your Body Genuinely Can't Really Tell The Difference EitherThis isn't just a sensor limitation but a biological one first.The levels of cortisol and adrenaline will rise no matter whether the cause is a breakup or a deadline. The heart rate will increase whether you are scared, angry or merely just climbed a hill. Researchers have known for some time now that autonomic signals largely indicate the general state of thrill, which in turn is affected by physical activity, stress, caffeine, an argument and even a horror film.A 2026 study on the differentiation of types of stress, physiological states that differ considerably from each other but still cause an increase in the level of thrill, tends to be merged in wearables and the devices have serious problems distinguishing between psychological stress and physical activity.Therefore, when a red warning appears on your smartwatch as you deadlift, it's not broken. It's doing exactly what it was built to do. It just can't ask you what's actually happening.The Accuracy Numbers Are Better Than You'd Expect, And Also NotHere's a fun contradiction: machine learning algorithms fed on wearable sensor data reached an astonishing accuracy level of 99% in laboratory settings to distinguish between stressed and non-stressed states. This is impressive until we notice the word "controlled". These models are mostly developed and tested on the same small sample of individuals, in laboratory conditions, carrying out some pre-planned activity associated with stress, such as giving a speech or solving a math problem in under 2 minutes.The advantage of this is that researchers know in advance when the stress-inducing activity began and ended. Real life doesn't hand you clean labels.You are not performing a planned Trier Social Stress Test at your desk, you are receiving a passive-aggressive Slack message while being slightly hungover and dehydrated at the same time. A systematic review of this area of research noted a small sample size and lack of standardized protocol as the main reasons behind the inability of laboratory accuracy rates to be maintained outside of the lab. So yes, the technology is indeed effective in recognizing a thrill but much less in determining its cause.The Missing Piece: CortisolThe hormone that is most strongly correlated with actual stress is cortisol and there currently exists no wearable device that can actually measure it. Cortisol response is indicative of the way your brain assesses the situation rather than simply indicating how quickly your heart is beating. That is why studies have used both saliva testing and wearables: heart rate and skin electrical activity can't tell the difference between a stress-inducing meeting and a stress-inducing exercise session. Obviously, you can't take a saliva sample with your wrist device anytime soon.Context Is Doing All The Heavy LiftingThat's where things become intriguing for anyone building in this space. The raw data cannot sort this issue out. However, there is always something else surrounding the signal.Some factors that can really improve the accuracy:Motion data. When you combine HRV or EDA with accelerometer readings, your model could be able at least to exclude that the increase of stress was associated with your working out. This is the first thing that most wearables try to deal with, that's why your watch does not interpret a run as a stress event anymore.Time and location. A 2 AM reading and a 3 PM reading of the same number of HRV would give you completely different information about the nature of your stress.User's tags. Oura has this tool exactly because the sensor data cannot help here. By adding tags like "Coffee", "Flight" or "Argument", your app will be able to establish your individual baseline, not a population one.Longer window. The momentary value of the stress is almost irrelevant. What's relevant is the two to four-week window, where all the outliers will be leveled down.Why This Matters If You're Building With This DataIf you happen to be developing any wearable APIs, it's worth designing around this limitation instead of pretending it doesn't exist.Some concrete things I would advise:Don't label a raw HRV dip as "stress" in your UI without context as that would be misleading. Instead, refer to it in your terms: elevated sympathetic activity and leave it to the user or contextual information to figure out what it means.If you are building some sort of mental health system, take any single reading as a hypothesis, rather than a fact. The consistent finding from the research here is that accuracy of single-point detection is pretty poor, even if multi-day trends can be detected fairly accurately.Multimodal analysis wins hands-down. Models using multiple sensors (EDA, HRV, behavioral and movement data) have been consistently found to outperform any single-channel model.Personalization is essential as the "high stress" threshold based on a general population is bound to be useless for athletes, outliers and people with various forms of heart disease.So What Do You Actually Do With The NumberI'll admit that I still check my stress score daily. What's changed is that I'm no longer taking it as the absolute truth. Should it jump and I understand the cause (stairs, coffee, an upcoming argument), I just move on. Should it jump and I simply cannot pinpoint the reason, now that's when it starts getting my attention and that is typically my body telling me something that my mind isn't picking up on just yet.That sensor isn't deceiving you. It's simply telling you your nervous system has reacted, but for reasons known only to itself at this time. That's the part you have to find out, and truthfully, I think it's a good thing. I'd rather wear a device that couldn't read my mind than one that could.To all of the future engineers looking to make the next version of this technology, here's what you need to understand: there is no clever algorithm that you can code to give you more data about your stress levels from HRV and EDA alone. There is a limit to those signals and you've reached it. Where the innovation has to take place in the future is in tying the sensor data to the voluntary information about context supplied by the user- tags, journal entries, calendar data, sleep and nutrition information, anything that gets you to the "why" and not the "how much."