It's 5:45am. Your alarm goes off and you check your watch before your feet hit the floor. Your HRV is down 12 points from your 7-day average. Your sleep score says 71. You slept six and a half hours because your flight landed late. Your training plan says tempo run: 8 kilometres at threshold pace.

What do you do?

If you're like most runners, you do one of three things. You ignore the data and run the session anyway, because the plan says so. You skip the session entirely and feel vaguely guilty about it. Or you stare at the numbers, wonder what they actually mean, and go for an easy jog because that feels like a reasonable compromise.

None of these is a particularly good decision. The first risks pushing through fatigue that a single night's sleep won't resolve. The second wastes a potential training opportunity if the HRV dip is just noise. The third is a coin flip dressed up as caution.

This is the fundamental problem with wearable data in 2026. We've never had more information about what's happening inside our bodies, and we've rarely been worse at knowing what to do with it.

What HRV actually measures

Heart rate variability is one of those metrics that sounds straightforward until you try to explain it. Your heart doesn't beat like a metronome. The interval between consecutive beats varies constantly: 820 milliseconds, then 790, then 850. HRV quantifies that variation.

The common misconception is that regularity equals health. The opposite is true. Greater variability generally reflects a more adaptable cardiovascular system. A heart that can modulate its rhythm rapidly in response to changing demands is a heart with headroom.

The mechanism behind this is the autonomic nervous system, specifically the interplay between its two branches. The parasympathetic branch, mediated by the vagus nerve, slows the heart and acts beat-to-beat. It's associated with rest, recovery, and what physiologists call vagal tone. The sympathetic branch speeds the heart and acts more slowly. It's associated with arousal, stress, and exertion.

HRV primarily reflects parasympathetic activity because vagal influence operates on each individual beat, while sympathetic effects are more diffuse. High HRV generally indicates strong vagal tone and healthy autonomic balance. Low HRV suggests the sympathetic system is dominant, which could mean stress, fatigue, illness, or simply that you ran hard yesterday.

This is established science. The European Society of Cardiology published standards for HRV measurement back in 1996, and the research base has only deepened since. Buchheit's work on monitoring training status through HR measures (Frontiers in Physiology, 2014) and Plews et al.'s research on HRV-guided training in endurance athletes (Sports Medicine, 2013) have built a robust framework for applying HRV in athletic contexts.

The science is not the problem. The interpretation is.

What your watch gets right (and what it doesn't)

The hardware has improved dramatically. A 2025 validation study published in Physiological Reports found that overnight HRV readings from consumer wearables like the Oura Ring achieved concordance correlation coefficients of 0.99 against chest-strap ECG, essentially clinical-grade accuracy in resting conditions. Apple Watch, Garmin, and WHOOP all produce reliable resting HRV data when measured overnight or immediately upon waking.

But there are important caveats that most runners never hear about.

Optical sensors (the green light on the back of your watch) are highly sensitive to noise during movement. The accuracy that holds during sleep degrades significantly during exercise, particularly at higher intensities. This is why chest straps remain the gold standard for session-level heart rate data, and why your watch's real-time HRV readings during a run should be treated with caution.

Timing matters enormously. Overnight recordings reflect what happened yesterday: how well you recovered, how much residual fatigue you're carrying. Morning spot measurements, taken in the first few minutes after waking, reflect your capacity for today. Neither is wrong, but they're measuring different things, and using them interchangeably is a common source of confusion.

And perhaps most importantly: HRV is profoundly individual. There is no meaningful population reference range. Two runners of the same age, sex, and fitness level can have wildly different absolute HRV values, largely due to genetics. Comparing your number to someone else's is meaningless. You need four to eight weeks of consistent personal data before your baseline is even established.

There's also seasonal noise to account for. HRV shifts with temperature, daylight exposure, and likely circadian rhythms, tending higher in summer and lower in winter. A declining HRV trend in autumn could be entirely environmental rather than a sign that your training or lifestyle is off. Without accounting for this, you could wrongly conclude that something needs fixing when nothing has changed.

Your watch is giving you a real signal. It's just not telling you what the signal means.

The single-number trap

This is where most runners get stuck, and where the current generation of wearable apps does them a disservice. The typical interface presents HRV as a single number with a colour: green (you're fine), yellow (be cautious), red (take it easy). It's intuitive. It's also dangerously oversimplified.

HRV is simply a proxy for your autonomic nervous system's stress response. It can't distinguish where the stress is coming from. A suppressed reading looks identical whether it's caused by a hard track session, a bad night's sleep, a stressful week at work, or the early stages of illness. The body's stress response is the same signal regardless of the source.

This is key to note, and a limitation. HRV is applied so heavily in training contexts not because training is necessarily the main driver of it, but because training is the one stressor you can easily do something about. If your HRV is low because of work stress or family pressure, the data is still valid, but you're largely powerless to change the cause. Adjusting training is simply the most actionable lever available.

For many runners, this creates a paradox. Training is actually a stress relief, meaning life stress may be chronically suppressing HRV in the background while training is barely moving the needle at all. If you're puzzled why your HRV doesn't respond the way you expect to training changes, chronic life stress could be the elephant in the room. The value of HRV is in spotting patterns and adjusting behaviour accordingly, not in diagnosing any single cause.

This matters because the appropriate response depends entirely on the source. If your HRV is low because yesterday's interval session was genuinely hard, you might be fine to do an easy run today. If it's low because you're fighting off an infection, that easy run could tip you into illness. If it's low because of chronic work stress, training might actually help.

There's another blind spot most runners don't realise: HRV primarily flags intensity, not volume. Easy training, even high-volume easy training, barely moves HRV. The disruption comes specifically from intensity above the first ventilatory threshold. This means HRV-guided training really only gives you meaningful feedback around your hard sessions, not your overall training load. You could be accumulating volume-related fatigue for weeks without your HRV registering a thing, right up until something breaks.

A single number can't make these distinctions. A traffic light can't either. What's needed is context.

Why trends matter more than snapshots

The real value of HRV data emerges over weeks and months, not on any single morning. Plews and colleagues demonstrated this in their work on HRV-guided training: the runners who benefited most weren't reacting to daily readings. They were tracking trends and adjusting training blocks accordingly.

Here's what that looks like across a training cycle for a runner building toward a marathon.

During a base phase, HRV should remain relatively stable or trend gradually upward as aerobic fitness develops. Drops lasting two or more days suggest inadequate recovery. A long run might suppress HRV for a day or two, which is normal, but it should bounce back before the next quality session.

During a build phase, eight to twelve weeks out from race day, you'd expect HRV to trend slightly downward as training load increases. This is productive fatigue accumulating and it's by design. The warning sign isn't the downward trend itself. It's whether the trend reverses during recovery weeks. If HRV drops and keeps dropping through a scheduled recovery week, something is wrong.

During a taper, HRV should rise as training load decreases. If it doesn't rebound, the taper may need to be extended or deepened. If you're two weeks out and your HRV hasn't moved, taper harder. Better to arrive at the start line slightly undertrained than overtrained. A study reviewed in a 2021 meta-analysis in the Journal of Sports Science and Medicine found that HRV-guided training, where intensity was adjusted based on daily HRV status, produced fewer negative responders and more positive responders than predefined training plans, even when absolute performance gains were similar. The benefit wasn't faster times. It was fewer athletes arriving at race day overtrained or under-recovered.

This longitudinal view transforms HRV from a confusing daily number into a genuine training tool. But building that view requires something most runners don't have: the ability to synthesise weeks of data across multiple signals and make nuanced training decisions every single day.

Beyond HRV: the multi-signal picture

Here's the thing about HRV that the apps don't emphasise: it's one signal. A useful one, but just one. And interpreting it in isolation is like diagnosing an engine problem by listening to the exhaust note alone.

The picture becomes dramatically more useful when HRV is combined with other data streams that most serious runners are already generating. Training load, both acute and chronic. Sleep duration and quality. Resting heart rate trends. Pace-to-heart-rate ratios on easy runs, which reveal cardiovascular drift before the runner feels it. And subjective feel, which remains one of the most underrated data points in endurance sport.

Consider the 5:45am scenario from the opening. HRV is down 12 points. In isolation, you don't know what to do. But layer in the context: you flew yesterday (travel fatigue), you slept six and a half hours (below your average of seven and a half), your resting heart rate is four beats above baseline (sympathetic activation), and your last two easy runs were 15 seconds per kilometre slower than your 30-day average (accumulated fatigue showing up in performance). Now the picture is clear: this isn't noise. This is an athlete carrying a real fatigue load, and today's tempo session needs to become an easy 40 minutes or a rest day.

A week later, your HRV is down again, but by only 6 points. You slept well. Your resting heart rate is normal. Your easy pace yesterday was right on your average. The dip is probably just day-to-day variability, and the tempo run is likely fine to execute. Same metric, different context, different decision.

No single data point drives the decision. The decision emerges from the pattern across all of them. This is exactly how a great human coach thinks: they weigh multiple inputs, apply experience, and make a judgment. The problem, as we've discussed in previous posts, is that most runners don't have a coach, and even the coaches who exist can't realistically synthesise this much data for thirty athletes every morning.

Where AI changes the equation

This is the gap that AI is uniquely positioned to fill, and it's worth being specific about why.

The challenge isn't computational. Any spreadsheet could flag an HRV reading below your 7-day average. The challenge is contextual interpretation: understanding what the combination of signals means for this particular athlete on this particular day, given their training history, their life circumstances, and their goals.

Large language models can carry context across weeks and months of interaction. They can hold the fact that you mentioned a sore left Achilles three weeks ago alongside the observation that your HRV has been trending down since your work trip, and cross-reference both with the fact that your marathon is eight weeks away and your long runs have been consistently undershooting target pace. From that, they can make the kind of judgment a great coach would make: this athlete needs a recovery block now, not next week, and the Achilles needs addressing before it becomes a real problem.

This isn't speculative. Research reviewing data from over 11,000 athletes found that AI-adapted training plans, where daily sessions were adjusted based on incoming physiological data, reduced injury rates by up to 23% and improved workout consistency by roughly 16% compared to static plans. The gains didn't come from better training science. They came from better daily decisions: the right session at the right time for the right athlete.

The multi-signal picture that makes HRV genuinely useful, HRV plus sleep plus load plus pace plus subjective feel, is also the picture that's too complex for a runner to interpret manually every morning. You'd need to cross-reference five or six data streams, weight them against your recent training history, factor in where you are in your macrocycle, and make a decision before your coffee gets cold. Every single day. For months.

That's not a realistic expectation. But it is a realistic description of what an AI coaching system can do continuously, consistently, and at scale.

The future is multi-signal, not single-metric

The wearable industry has spent the last decade getting very good at collecting data. Heart rate, HRV, sleep stages, blood oxygen, skin temperature, and on and on. The sensors are accurate. The data is abundant. What's been missing is the interpretation layer: something that takes all of these signals, understands them in context, and translates them into a concrete training decision.

That's the shift happening now. The future of personalised training isn't a better HRV algorithm or a more accurate sleep score. It's the ability to synthesise everything your body is telling you, combine it with everything you're telling your coach about how you feel, and produce daily training decisions that are genuinely individualised. Not individualised in the marketing sense, where your name is at the top of a generic plan. Individualised in the Bowerman sense, where the plan adapts because the coach knows you.

Your watch is already collecting the data. The question has always been whether anything was smart enough to use it.

Zepho is adaptive coaching for serious runners. It's live and available now on the App Store (desktop) & (mobile).