From WHOOP data to decisions: a Sunday workflow

WHOOP will give you a recovery score every morning. Hybrid life will give you a reason to argue with it. A number without a decision is just another notification. Zignoli described response-guided sports science: look at how the athlete is responding, then choose the next dose (Zignoli, 2024). The wearable is the sensor. The model can be the clerk. The human is still the coach.

Miller, Sargent, and Roach found WHOOP overnight heart rate and HRV tracked reasonably against ECG in healthy adults, while four-stage sleep staging was weaker. Trust overnight cardiac trends more than stage pie charts (Miller, 2022). Panchawagh and colleagues reviewed WHOOP across wellness and performance papers: some metrics earn more confidence than others (Panchawagh, 2024).

Bellenger and colleagues showed that even Olympic athletes have real day-to-day bounce in WHOOP HRV. If you let Monday's red score rewrite a well-built week, you are coaching noise. Sunday is for slopes (Bellenger, 2022).

This is a suggested process, not a diary of my weeks. Pick one hour that is not after a race and not before a 5 a.m. alarm. Export or screenshot the last seven days from WHOOP: recovery, HRV, resting heart rate, sleep duration, and strain. Put the training log next to it: what you lifted, how far you ran, session RPE, and anything that was not training (travel, alcohol, heat, illness in the house).

Paste both into an AI chat with a boring instruction: summarize weekly patterns, not daily verdicts. Ask it not to diagnose injury or prescribe medicine. Read the summary like a training partner: useful if it matches the log, disposable if it invents a story the numbers do not hold. Biosensing plus machine learning is already discussed as coaching support, not as a replacement coach (Biosensors, 2026). The value is compression: seven mornings become two sentences about HRV, strain, sleep, and RPE.

Vague prompts produce vague mysticism. Specific questions produce a brief you can accept or reject. Keep the human veto. If the model tells you to skip a session your tendon already forbade, the tendon wins. If it tells you to hammer intervals because one recovery score was green, the plan wins.

A single recovery number is a mood. A seven-day recovery average is a weather report. Strain works the same way. This week's mean strain divided by the last four weeks' mean strain is an ACWR-style flag built from internal load. It will not match your running-kilometre ratio, and that disagreement is useful. Maybe the heart-rate model thought the week was mild while the downhill kilometres were not.

When strain ACWR sits high and the recovery rolling average is sliding, the Sunday decision is usually subtract, not add: drop a finisher, move the heavy lower-body day away from the long run, or turn an interval into aerobic work. When strain ACWR is modest and recovery is stable, you may have room to progress one quality, not both on the same afternoon.

Do not let the flag become a religion. Ratios inherit every limitation of the metric underneath. WHOOP strain is not kilometres and not tonnage. Combine it with the log or you are steering with one eye closed.

Kiviniemi and colleagues showed endurance training guided by daily HRV could be individualized: when HRV was down, the hard session waited (Kiviniemi, 2007). Vesterinen and colleagues found HRV-based prescription could match or beat a predetermined endurance plan (Vesterinen, 2016). Those papers are about swapping intensity when the overnight signal is suppressed, not skipping every session that is not lime green.

For a hybrid athlete the translation is conservative. If HRV and recovery have been low for several days and the log shows stacked intensity, move the quality session. Keep easy aerobic work and technique lifting. Do not invent a third hard day because a model wrote encouraging prose. If HRV is low and you also have pain, fever, or performance that has been down for weeks, stop treating it as a programming puzzle. That is a rest-and-clinician problem.

Sunday ends with one sentence you write yourself. Example: protect the squat; keep the long run easy; no intervals until sleep is back above the recent average. The export and the model exist to make that sentence shorter to find. They do not exist to outsource judgment to a paragraph that cannot feel your knee. This workflow is educational. Chest pain, fainting, disordered breathing at night, and unexplained collapse in capacity need care, not a cleverer prompt.