Recovery Math: How HRV, Sleep, and RPE Predict Tomorrow's Session

Recovery prediction is a market full of consumer wearables claiming to score readiness on proprietary algorithms. Underneath the marketing, the data sources are HRV, sleep duration, and self-reported exertion, and the academic literature on each is older and less consistent than the wearable industry implies. This article walks through what each signal actually measures, where the consumer interpretations break, and how to assemble a 3-input model that predicts tomorrow's session better than any single number.

Use the Sleep Calculator to plan duration targets against your wake time, the RPE to Percentage Converter to track yesterday's session at a comparable load metric, and the TDEE Calculator as a sanity-check on whether energy availability is supporting the recovery demand.

1. What HRV actually measures

Heart rate variability is the variation in millisecond intervals between consecutive heartbeats. The Task Force standard^[7] defines several time-domain and frequency-domain measurements:

Metric         Description                              Athlete-relevant
─────────────────────────────────────────────────────────────────────────
RMSSD          Root-mean-square of successive RR diffs   Yes (main signal)
SDNN           Standard deviation of normal RR intervals Yes
LF/HF ratio    Low-freq / high-freq power ratio          Disputed
pNN50          % of intervals differing >50ms            Yes (similar RMSSD)

RMSSD is the metric most consumer apps compute and report (Whoop, Oura, HRV4Training, Elite HRV). The autonomic-nervous-system interpretation is approximate: high RMSSD reflects parasympathetic dominance (rest-state, recovered); low RMSSD reflects sympathetic dominance (fight-or-flight, stressed or under-recovered). The translation to "fitness for training tomorrow" relies on Plews's^[2] framework that 7-day rolling averages correlate with training load while single days are too noisy.

2. The HRV measurement-noise problem

Same lifter, same recovery state, same morning, two different HRV readings can vary 5–15%:

• Time of measurement. RMSSD measured 30 seconds after waking is different from RMSSD measured 5 minutes after waking. Standardise.
• Posture. Lying vs sitting changes vagal tone and shifts RMSSD by 10–25%.
• Recent food/water. A morning measurement before vs after the first sip of coffee differs measurably.
• Hydration state. Dehydration drops RMSSD; over-hydration after training raises it artificially.
• Alcohol the night before. Two drinks suppresses HRV by 20–40% for 18–36 hours.
• Late meal. Eating within 3 hours of bed shifts RMSSD downward overnight independent of training.

Plews's argument^[1] is that the 7-day rolling average smooths most of these confounds. A single morning's RMSSD ±20% from your typical value is noise. A 7-day average that's 10%+ below your baseline for a week is signal.

3. The sleep variable: quantity vs "score"

Mah, Mah, Kezirian, and Dement's 2011 study^[6] took 11 collegiate basketball players and extended their sleep to 10 hours/night for 5–7 weeks. The performance changes:

Metric                   Baseline    Sleep-extended    Δ
─────────────────────────────────────────────────────────
Sprint time              16.2 s      15.5 s            -4.5%
Reaction time            ~280 ms     ~270 ms           ~-3.6%
Free-throw accuracy      54%         63%               +9pp
3-point accuracy         42%         51%               +9pp
Subjective fatigue       Higher      Lower             -

The study's framing is that sleep is plausibly the most impactful non-training variable in athletic preparation. Halson 2014^[3] reports that 6 hours/night for a week reduces strength output, anaerobic capacity, and cognitive function. Practical implication: a lifter sleeping 7 hours regularly and reporting "good sleep score" on a wearable is leaving meaningful performance on the table compared to 8.5–9 hours.

The "sleep score" wearable metric collapses duration, efficiency, time spent in each stage, and HRV during sleep into a single number. Empirically, the duration component dominates for athletic performance prediction. A 9-hour sleep with mediocre "score" outperforms a 6.5-hour sleep with excellent "score" on next-day metrics. The wearable algorithms over-emphasise stage-percentage components that the published literature doesn't strongly support as performance predictors.

4. RPE as a recovery signal

Rate of Perceived Exertion (Borg's 0–10 scale, modified by Tuchscherer^[4] for resistance training) provides a same-session readout of effort. Used cross-session as a recovery signal, the relevant comparison is RPE-at-fixed-load: yesterday's prescribed top-set squat at 80% 1RM should land at the same RPE today as it did 4 weeks ago, plus or minus a small drift.

Tuchscherer's RPE methodology^[4] treats RPE drift at fixed prescription as the primary autoregulation signal. If yesterday's 80% top-set landed at RPE 8 instead of the expected RPE 7, accumulated fatigue is unrecovered. The 2018 systematic review on RPE for resistance training^[5] validates that experienced lifters report RPE within 1 unit of intended after ~20 sessions of practice, accurate enough to use as a tracking metric.

A worked tracking example for a lifter on a 4-day upper/lower split:

Day        Lift           Load (% 1RM)    Expected RPE    Actual RPE
────────────────────────────────────────────────────────────────────
Mon        Squat top 5    80%             7               7    ✓
Tue        Bench top 5    77%             7               8    +1
Thu        Deadlift 3     85%             8               9    +1
Fri        OHP top 5      75%             7               9    +2

Pattern: RPE drifted +1 to +2 across week → accumulated fatigue
Action: drop volume next week 30%, deload

A single +1 RPE day is noise (sleep, nutrition, stress). A 4-day pattern of +1 to +2 across multiple lifts is a clean signal that the prescribed loads are now exceeding readiness. Tuchscherer would call this a planned-fatigue resolution: drop volume the next week, hold or reduce intensity, and let the trend normalise.

5. The three-signal combined model

No single signal predicts tomorrow's session reliably. HRV is noisy day-to-day; sleep duration captures only one recovery axis; RPE is a lagging indicator. Combined, they produce stronger inference. A practical 3-input model:

Signal               Green                    Yellow              Red
────────────────────────────────────────────────────────────────────────
HRV (7-day avg)      ≥ baseline -3%           -3% to -8%          < -8%
Sleep last night     ≥ 8.0 h                  6.5–8.0 h           < 6.5 h
Yesterday's RPE      = expected               +1 above            +2 or more

Decision rules:
  3 green:           full session, optionally ambitious
  2 green / 1 yellow: full session, conservative load
  1 green / 2 yellow: cap top set at -5%, drop accessory volume 20%
  3 yellow OR 1 red:  conservative session, no PR attempts
  2+ red:             active recovery only or full rest day

The combination outperforms any single signal because the signals capture different recovery axes. HRV captures cardiovascular and autonomic recovery. Sleep duration captures cognitive and hormonal recovery. Yesterday's RPE captures musculoskeletal and neural fatigue. A lifter green on all three has all three systems recovered. A lifter yellow on two has compounding deficits even if each alone is tolerable.

6. What the 7-day HRV pattern tells you

Single-morning HRV is noise; rolling-7-day-average HRV is signal. Plews's 2014 paper^[1] tracked elite rowers across training blocks and found:

• HRV stable inside ±5% of baseline: training load is being absorbed.
• HRV declining 5–10% across 5–7 days: functional overreach. The lifter is accumulating fatigue, but recovery is still occurring; one deload week typically restores it.
• HRV declining >10% for 10+ days: non-functional overreach territory. A single deload week may not be sufficient; a longer reduced-volume block is indicated.
• HRV elevated 10%+ above baseline: often misread as "super-recovered." Plews flags this as a parasympathetic-dominant state that can co-occur with under-recovery in masters athletes. Examine alongside other signals.

The implication is that HRV is most useful as a confirmation signal across a week, not as a single-day go/no-go. Wearables that report "today's HRV is low, take it easy" are over-reading the signal in 60% of cases.

7. Sleep duration anchors

The Sleep Calculator back-calculates required bedtime from wake time and target duration. Anchor targets for active lifters:

Population                         Target duration   Lower bound
──────────────────────────────────────────────────────────────────
Recreational lifter, day job        7.5–8.0 h        7.0 h
High-volume amateur (10+ sets/lift) 8.0–8.5 h        7.5 h
Competitive powerlifter, prep       8.5–9.0 h        8.0 h
Endurance athlete in volume block   8.5–9.5 h        8.0 h
Masters lifter (40+)                8.0–8.5 h        7.5 h
Adolescent athlete                  9.0–10.0 h       8.5 h

The "lower bound" column is the threshold below which Halson's^[3] review reports measurable performance decrement on the same training prescription. A lifter who routinely sleeps 6.5 hours is operating in chronic mild sleep debt, and the cumulative effect across a 12-week training block is a meaningful percentage of total adaptation lost.

8. Calibrating the model to your baseline

All three signals require a personal baseline. HRV varies dramatically between individuals: a 25 ms RMSSD baseline is normal for one lifter and a flag-state for another. Sleep tolerance varies: some lifters function near-optimally at 7 hours; others crash at anything under 8.

Two-week calibration protocol:

1. Measure morning HRV daily for 14 days at standardised conditions (same posture, same time after waking, no caffeine yet, no measurement on training-day mornings if you can't standardise post-session timing). Compute the 7-day rolling average from days 8–14.
2. Track sleep duration nightly. A wearable's duration estimate is fine; ignore the score component.
3. Log RPE on every working set at fixed prescriptions, comparing to the program's expected RPE.

After 14 days, you have a baseline HRV (the 7-day average), a sleep median (your typical duration), and an RPE-baseline drift pattern (typically near-zero for early-cycle weeks, drifting upward late-cycle). The thresholds in the 3-signal model become percentages off your baseline, not absolute numbers.

9. Where the consumer wearables go wrong

Whoop, Oura, Garmin, and similar devices market readiness scores that promise to predict training fitness. The literature support is mixed:

• HRV measurement accuracy: chest-strap and finger-PPG sensors agree well; wrist-PPG (Apple Watch, Garmin wrist) has known artefacts during sleep that bias HRV estimates 10–15%.
• Sleep stage estimates: consumer wearables claim REM/deep/light staging, but validated polysomnography comparisons show 70–80% agreement at best. Stage-based scoring is on shaky ground.
• Proprietary score weighting: the formulas combining HRV, sleep, and recent training load into a "recovery score" are not published and not independently validated. The output is a scalar with unknown calibration.
• Cross-device disagreement: the same morning, Whoop and Oura frequently produce recovery scores that differ by 20–30%. Both can't be right.

A lifter who tracks HRV (RMSSD, 7-day rolling), sleep duration (just minutes), and yesterday's RPE explicitly is doing what the wearable is doing, but with transparent inputs, personal baselines, and visibility into which signal is driving which decision. Coach-tracked rowers with this discipline outperform the wearable's recommendation in case studies cited by Plews^[2].

10. The under-recovery cascade

Sustained under-recovery follows a predictable signal sequence:

Week    Signal                                                  Stage
──────────────────────────────────────────────────────────────────────
1       Yesterday's RPE +1 across 2–3 sessions                  Early
2       7-day HRV dropped 4%, sleep okay                        Functional
                                                                 overreach
3       7-day HRV dropped 8%, RPE +2 on top sets                Late functional
                                                                 overreach
4       Sleep quality degrades, harder to fall asleep           Borderline
5       Sleep duration shortens despite same bedtime,           Non-functional
        morning HR elevated 5+ bpm                              overreach
6+      Mood, libido, motivation decline; lifts regress         Overtraining
        across training cycles                                  (true OT rare)

Catching the cascade at stage 2 (HRV drift confirmed, RPE drift confirmed) is the cheap intervention: one deload week resolves it. Catching it at stage 5 is expensive: 2–4 weeks of reduced training, often with regression of the prior cycle's adaptations. The 3-signal model's value is catching the signal at stage 2 rather than at stage 5.

11. Energy availability as a recovery floor

No HRV-or-sleep readiness model fixes under-eating. Plews and Buchheit both flag^[1][2] that endurance athletes in deficit show suppressed HRV that doesn't recover until calorie intake is restored. The same observation extends to lifters. The TDEE Calculator output isn't directly a recovery signal, but a chronic 500+ cal/day deficit during a training block sets a low ceiling on how recovered the lifter can become regardless of sleep and HRV inputs.

Practically: if HRV is consistently low and sleep is at target, check whether energy availability has dropped. A lifter who entered a cut while maintaining the bulk's training volume often exhibits HRV suppression that the 3-signal model interprets as accumulated training fatigue. The actual root cause is undereating relative to the ongoing training demand.

12. The honest scope of this model

The 3-signal HRV + sleep + RPE model has been validated in coaching practice but not in head-to-head RCTs against alternatives. The Plews papers^[1][2] establish HRV as a useful longitudinal monitor in elite endurance populations. The Mah sleep paper^[6] establishes sleep duration's performance-prediction value in collegiate basketball. Tuchscherer's RPE work^[4] validates RPE as a fatigue signal in resistance training, and the 2018 systematic review^[5] confirms experienced-lifter RPE accuracy.

Combining the three signals into a decision rule is informed extrapolation. The thresholds in the green/yellow/red table are reasonable starting points. After 8 weeks of calibration against your own training data, the personal thresholds will be more accurate than any generic table. The model's value is in giving structure to recovery monitoring; the execution depends on consistent measurement and honest reading of the signals against an actual baseline.

13. Population caveats

The literature anchoring this model:

• HRV literature is mostly endurance. Plews's rowers, Buchheit's runners, Kiviniemi's cyclists. Resistance-training-specific HRV-readiness data is much thinner. Treat the HRV-fatigue translation in lifters as informed extrapolation.
• Sleep literature spans broader populations. Mah's basketball players, Halson's general athlete reviews. The duration thresholds generalise reasonably across sports.
• RPE literature is well-grounded for trained lifters. Less reliable for novices, who systematically under-rate exertion and over-estimate reps in reserve in their first 4–8 months.
• Sex differences: HRV baselines run 10–20% different between sexes; the percentage-from-personal-baseline framing handles this. Sleep needs are similar.
• Wearable cross-validation: none of the published threshold percentages were derived from consumer-wearable measurements specifically. If your HRV source is wrist-PPG, expect more noise than this article assumes.