Rest is a prescription, not a pause

How long the recovery between reps should be, at what intensity, whether it should stay constant across the session, and how far apart two hard days belong. Four questions that are usually answered by copying the last workout.

REF-131 (B in the base; n=10 trained cyclists, VO2max 58.6, age 41 ± 10) fits W′ reconstitution as bi-exponential and the group fit is exact (adjusted R² = 0.999; AICc 30.2 vs 72.2 for mono-exponential): τ-fast ≈ 21.5 s returns roughly the first 50% of W′ (fast-component amplitude 50.7%); τ-slow ≈ 388 s returns the second 50%. Two further findings from the same paper matter for a session: after a second bout the slow time constant lengthened to ~716 s while the fast one did not change — the slow half was slower after a second exhaustive bout (later bouts were not tested) — and individual fits varied widely (τ-fast 22 ± 12 s, τ-slow 377 ± 100 s on the first recovery).

Read practically, that means recovery time buys capacity at wildly different rates depending on where you are on the curve (Inference from the two time constants; the paper reports the fit, not these bands):

So there are three defensible recovery lengths, and the space between them is where sessions quietly waste time (Not from a study; the base's own practice, reasoned from the curve):

A 3-minute recovery is a 2-minute recovery with a minute of standing around (about 69% vs 64% of W′ by the group time constants), unless the session is deliberately targeting the slow phase.

REF-246 (B+, controlled crossover, n=18 trained cyclists, FTP 258 W, on Zwift — whether ERG mode was used is not stated) compared 50 W, 50% FTP and self-selected recovery between 5 × 3-min severe-intensity intervals with 2-min recoveries: no difference in subsequent interval power (300.1 / 296.9 / 298.8 W).

The authors' reading: dropping recovery below 50% FTP failed to enhance the next interval, so a lower limit for useful W′ reconstitution had been reached; self-selected recovery (~122 W on average) drifted down through the session as riders adapted it to hold the work power.

The coaching consequence is direct: stop policing recovery watts and start prescribing recovery duration. Let the athlete pedal at whatever feels natural. (Honest limit: this is one B+-level study with 2-minute recoveries and recovery intensities at or below 50% FTP. Over longer recoveries, where the slow component is doing the work, recovery intensity may well matter — the base does not answer that.)

REF-130 (B in the base; n=12 amateur cyclists, Tier 2, crossover) — HIDIT holds the same work and recovery power and the same 3:2 work:recovery ratio (the entry gives ~117% CP / ~83% CP) while the work interval decreases (3 min → 30 s, recovery 2 min → 20 s) and beats both classic long (3/2) and short (30/20) intervals on time above 90% VO2peak: 312 s vs 179 and 182 s, with time to exhaustion, lactate and RPE not different.

That is the curve applied: as W′ depletes across the session, the work the athlete can actually complete shrinks, so shrinking it deliberately keeps every rep on target instead of watching the last two collapse (inference — the study measured time near VO2peak, not W′).

The same entry carries a second warning worth repeating: work-above-CP is not a valid measure of a HIT session's effectiveness (the dissociation is in the entry's full-text table, not the abstract). Do not judge a session by that number.