AI Running
010 Reading Readiness, HRV and Recovery Scores 1,722 words · 8 min

Your HRV Dropped. Here’s What That Does and Doesn’t Mean

You woke up, glanced at your wrist, and the number was wrong. 42ms, against a baseline that usually sits near 58. Garmin has printed “Unbalanced” where it normally says “Balanced.” Whoop is showing 31% recovery in red. Oura has decided your readiness is 68 and would like you to consider a rest day.

And you have 8 × 800m at 5K pace on the calendar this evening, because Runna put it there three weeks ago and you have been looking forward to it.

Here is the short version of what follows: one low HRV reading is not information you should act on. A four-day slide against your own baseline is. The distinction sounds pedantic and it is the single most useful thing you can learn about this metric, because getting it wrong in either direction will quietly wreck a training block. Bail on every red morning and you never accumulate the consecutive hard weeks that actually make you faster. Ignore a genuine trend and you spend six weeks digging a hole that takes another six to climb out of.

Why a single overnight reading is mostly noise

HRV is the variation in time between your heartbeats, measured overnight, and it reflects the balance between your sympathetic and parasympathetic nervous systems. That sounds authoritative. What it means practically is that HRV is a downstream signal of dozens of inputs, only one of which is your training.

The night-to-night coefficient of variation in HRV for a healthy adult is commonly in the range of 10 to 20%. If your rolling baseline is 58ms, a reading of 47ms sits inside normal biological scatter before we have discussed a single lifestyle factor. Nothing has gone wrong. That is just what the measurement does.

Then add the actual causes. Two glasses of wine at 9pm will suppress HRV that night by 20 to 30% in many people, and the effect is dose-dependent and largely gone within 48 hours. A big late meal, especially one eaten inside two hours of sleep, keeps digestion running through your early-night sleep cycles and drags the number down. Sleeping in a warm room does it. A short-haul flight does it. Your second coffee at 4pm does it. The first night of a head cold does it, days before you feel ill. Poor strap contact does it, which is why Whoop users see phantom dips after a night of tossing.

And the algorithm itself introduces variance. Garmin samples HRV during sleep and averages the stable portions. If your watch slipped, if you woke up at 3am and read your phone for twenty minutes, if the sleep stage detection misfired, you get a reading computed over less data and more noise. Tuesday’s 42ms and Wednesday’s 61ms can come from the same underlying physiology.

So when you see one low morning, the correct response is genuinely boring. Look at the previous evening. Did you drink? Eat late? Sleep badly or short? Fly? Feel a scratchy throat? If yes, you have your explanation and it is not your training plan. Do the session, and use your legs and your pace at a given effort as the real referee.

What a trend actually looks like

A trend is not two low days. It is a run of readings that sit below your baseline, ideally with the baseline itself starting to bend downward, and it usually comes with corroborating evidence you can see elsewhere in your data.

Here is a real-shaped example. A runner eight weeks into a 16-week marathon build, baseline HRV 58ms, resting HR 48.

Day  HRV   7-day avg   RHR   Sleep   Notes
Mon  61ms    58ms      48    7h20m   easy 8km
Tue  43ms    57ms      49    6h10m   beer + late dinner, 8x800m
Wed  59ms    57ms      48    7h45m   easy 10km
Thu  57ms    57ms      47    8h00m   rest
Fri  60ms    58ms      48    7h30m   tempo 3x10min

Tuesday looks alarming in isolation. It is a beer and a short night, and it resolves in one day. The 7-day average barely notices. Nothing here justifies changing the week.

Now the other pattern, same runner, four weeks later:

Day  HRV   7-day avg   RHR   Sleep   Notes
Mon  52ms    56ms      50    7h30m   easy 12km, legs heavy
Tue  49ms    54ms      51    7h40m   6x1km, missed pace by 8s/km
Wed  47ms    52ms      52    7h15m   easy 10km
Thu  45ms    50ms      52    8h10m   rest, still flat
Fri  44ms    48ms      53    7h50m   32km long run scheduled

Nothing here is dramatic on any single day. 52ms is not a scary number. But five consecutive readings below baseline, a baseline that has fallen from 58 to 48 (a 17% drop), resting HR up 5bpm, sleep duration fine, and a key session missed by 8 seconds per kilometre despite normal effort. That is a trend, and Friday’s 32km is a bad idea as written.

The corroboration matters as much as the HRV. A falling HRV with rising resting heart rate and degrading performance at a known pace is three independent signals agreeing. A falling HRV on its own, with resting HR flat and your intervals hitting target pace comfortably, is much weaker evidence and often resolves on its own. If you want a deeper walkthrough of how readiness scores, HRV and sleep data combine (and where each one is unreliable), the pillar on reading readiness, HRV and recovery scores goes through the mechanics.

The three-of-five rule, and how to check it in about two minutes

You do not need software for this. Use a simple threshold: three or more of the last five mornings below your 30-day average, plus at least one corroborating signal. That is your trigger.

In Garmin Connect, open HRV Status and switch the view to the four-week chart. Garmin draws your personal balanced range as a shaded band, typically something like 45 to 68ms for our example runner. One dot below the band means nothing. Four of five dots hugging the bottom of the band or sitting under it, with the trend line sloping down, is the picture you are looking for. Garmin will flag “Low” after seven days outside the range, which is honest but slow; you can usually see it by day four.

Whoop users should ignore the daily recovery percentage entirely for this purpose and go to the monthly performance assessment, where HRV is plotted as a 30-day trend. Oura’s readiness contributors screen shows “HRV balance” as its own line, which is the trend-aware version of the number and much more useful than the headline score.

Then cross-check in Strava. Pull up three easy runs from the last two weeks at similar routes and conditions, and compare heart rate at matched pace. If you ran 5:30/km at 142bpm a fortnight ago and you are running 5:30/km at 152bpm now on the same loop in similar weather, your body is confirming what your HRV said. Ten beats of decoupling at an easy pace, with no heat or altitude explanation, is real. Two or three beats is noise, the same as one low HRV morning.

What to change, and what most people get wrong

If you have a genuine multi-day trend, do not delete the week. The mistake runners make is going from a full plan to complete rest, which loses fitness and does not address what caused the problem.

Cut intensity, keep frequency, reduce volume modestly. Concretely, for our runner staring down that 32km long run on a five-day downward trend:

  • Replace the 32km with 19 to 22km, all easy, no marathon-pace segments.
  • Keep the easy runs on the calendar but drop them by 10 to 15% and run them by feel rather than pace.
  • Delete the interval session entirely. Do not “just do four reps instead of six,” because the neuromuscular and sympathetic load of hard running is what you are trying to reduce.
  • Hold this for four or five days, then reassess against the same three-of-five test.

Sleep is the highest-leverage lever, so protect it: nine days out of ten, a genuine HRV trend in a recreational runner is training load plus a sleep debt plus under-eating, not training load alone. Runners cutting carbohydrate while adding mileage see exactly this pattern, and no amount of rest fixes it while the fuelling stays wrong.

There is also a version of this that is not about recovery at all. A sustained HRV drop with an unexplained resting HR rise of 7bpm or more, or a low-grade fever, or swollen glands, is a reason to stop running and not a reason to modify a plan. Illness reads as overtraining in every app on the market, because the apps cannot tell the difference.

Where AI plans make this harder

The specific failure of Runna, Garmin Coach and ChatGPT-generated plans is that they are largely blind to this signal, but in different ways.

Runna does not read your HRV at all. It adapts based on completed sessions and the paces you hit, which means a five-day downward trend shows up only after you have already run a bad workout, and possibly two. Garmin Coach sees HRV Status but responds conservatively and late, and its adaptations tend to be single-session rather than structural. A ChatGPT plan is a static document: it has no idea what today’s data says unless you tell it, and if you paste in one bad morning it will cheerfully restructure your entire week around noise, because it will treat whatever you told it as the most important fact available.

That asymmetry is the whole practical point. These tools under-react to trends and, when prompted, over-react to single readings. You are the component that fixes both. Check the five-day picture once a week, on the same day, before your hardest session of the week rather than on the morning of it. That gives you time to move things rather than abandon them.

One more thing worth knowing: HRV should drop during a hard training block. A 5 to 8% dip in your rolling average during a genuine overload week, which recovers within three or four days of the following easier week, is the signal that the block is working. Runners who chase a permanently high HRV are training too easily, and their plans, whoever built them, will progress them more slowly than the data suggests they could handle.