How an AI coach adapts your training plan to HRV and sleep

A plan written twelve weeks out cannot know that you will sleep four hours on a Tuesday in week seven. A good human coach handles that in a text message: "Drop the intervals today, easy 30 minutes, we'll run them Thursday." The plan survives because one day moved, and the athlete does not dig a hole by hammering a hard session on no sleep.

An AI coach has to do the same thing, every morning, without being asked. This article explains what the readiness data can and cannot tell you, and how Loadout Solo turns it into a decision about today's session.

What the signals mean

Heart rate variability (HRV). The variation in time between heartbeats, measured overnight or on waking by most watches and rings. Higher HRV generally reflects a rested, parasympathetic state; a sharp drop below your own baseline often follows hard training, poor sleep, alcohol, illness or stress. The absolute number is not comparable between people. The comparison that matters is you against your own recent history.

Resting heart rate. Measured during sleep or on waking. A rise of several beats above your baseline is a simple, robust sign that you are not recovered, and it is less noisy than HRV in many people.

Sleep. Duration first, then quality. One short night is survivable; a string of them is the most common reason athletes break down in a long block.

Provider readiness scores. intervals.icu, Garmin and others publish their own readiness or body battery figures. They are useful inputs, computed from the same signals, and worth including rather than ignoring.

None of these is a verdict on its own. A low HRV with a normal resting heart rate and a full night's sleep is probably noise. A low HRV with a raised resting heart rate and five hours of sleep is a pattern.

Baselines, not thresholds

The mistake most apps make is a fixed threshold: HRV below 50, red day. Your HRV may live at 35 or at 110. Loadout Solo compares today against your own 28-day baseline for HRV and resting heart rate, imported from your watch history before the plan is even written, so the first morning already has context. A reading is a deviation from your normal, not a number on someone else's scale.

How the readiness score is built

Each morning the score is assembled from whatever signals are available:

• HRV against the 28-day baseline, weighted most heavily when a baseline exists.
• Resting heart rate against its baseline.
• Last night's sleep duration.
• The provider's own readiness score when present.

Missing signals are skipped rather than guessed. If you have no baseline yet, the HRV signal is left out until there is one. If the only signal is a provider readiness score, that score is used directly. The result is a 0 to 100 figure shown on your Today screen with the inputs beside it, so you can see why it landed where it did.

What the plan does with a low score

When the score is below 50 and today is a real training day (not rest or mobility), the AI proposes one adjustment. It never silently rewrites the plan; you accept it or keep the day as written.

1. Scale. The session stays, the loads come down. Strength sessions are rewritten with each exercise at 80 to 90 percent of the prescribed load and re-parsed into structured sets, so your watch and your log still show real numbers.
2. Swap. Today's hard session trades places with the nearest easier session in the next three days. The week keeps its total work; the hard day lands when you are more likely to be recovered.
3. Rest. The session moves to the next free day. Used when sleep and resting heart rate both say stop.

Which one it chooses depends on the session and on what is around it. A single bad night before an interval session usually gets a swap. A third short night in a row with a raised resting heart rate gets a rest.

Why only one change a day

An adjustment layer that touches everything becomes noise, and athletes stop reading it. One proposal, one reason, one tap. If you decline it, nothing else is changed. If you accept it, the change is made through the same path a human coach would use, so the session on your calendar, on your watch and in your log all agree.

What it will not do

• It does not diagnose illness or injury. A persistent drop in HRV with symptoms is a reason to see a doctor, not to accept a scaled session.
• It does not reward a high score with extra work. A good morning means the plan runs as written. The progression is handled week to week, not day to day.
• It does not need a wearable. With no HRV or sleep data the plan simply runs as written and the manual check-in (sleep quality, soreness, mood) still records how you felt.

Getting the data in

Loadout Solo reads HRV, resting heart rate and sleep from intervals.icu, Garmin, Apple Health and Health Connect, and activities from Strava as well. The history import on first connection fills the baselines; after that, the morning readiness check runs on whatever arrived overnight. Nothing is typed in.

FAQ

Is HRV accurate enough to base training on?

Night-time HRV from a watch or ring is consistent enough to track your own trend. It is not useful for comparing yourself with anyone else, and single readings are noisy, which is why the score weights HRV against a 28-day baseline and combines it with resting heart rate and sleep.

What readiness score triggers an adjustment?

Below 50 on a day with a real training session scheduled. Rest and mobility days are never adjusted.

Can I ignore the suggestion?

Yes. Every adjustment is a proposal. Keep the plan and the day runs as written.

Does a high score add training?

No. Progression is decided week to week from what you actually logged, not from one good morning.