Training to Failure vs Reps in Reserve: What Works

Proximity to failure is how close a working set gets to the point where another clean repetition is impossible. Lifters describe it as reps in reserve (RIR): 0 RIR is failure, 3 RIR means three good reps were left in the tank. For two decades the field argued a binary question, train to failure or stop short, and meta-analyses kept returning a frustrating "it depends." The 2024 literature changed the question, and the answer sharpened.

The reframe matters because the binary framing hid a gradient. Two studies can both be labelled "non-failure" while one stopped at 1 RIR and another at 4 RIR. Averaging them flattens any dose-response signal. This article walks the Grgic 2022 binary meta-analysis, the Refalo 2023 proximity review, and the Robinson 2024 meta-regression that recast the whole thing as a continuous curve, then turns the evidence into programmable RIR targets.

The binary era: Grgic 2022

Grgic, Schoenfeld, Orazem, and Sabol pooled trials that pitted training to repetition failure against stopping short, and analysed strength and hypertrophy separately.^[3] The headline was a non-result: no statistically significant overall difference between failure and non-failure for either strength or muscle size.

• Strength: no significant overall difference between conditions. A subgroup of studies that did not equate training volume favoured non-failure, consistent with non-failure sets preserving bar speed and quality.
• Hypertrophy: no significant overall difference. A subgroup of resistance-trained participants showed a signal favouring training to failure, hinting that trained lifters may need to push closer.
• Reading: failure is a tool, not a requirement. The binary contrast could not separate "stopped at 1 RIR" from "stopped at 4 RIR," so the gradient stayed invisible.

This became the standard talking point: failure is optional. The conclusion was correct for the question asked. It was also incomplete, because the question collapsed a continuum into two bins.

The proximity review: Refalo 2023

Refalo, Helms, Trexler, Hamilton, and Fyfe ran a systematic review with meta-analysis aimed squarely at proximity to failure and hypertrophy.^[2] Rather than a clean binary, they examined how the distance from failure related to muscle growth across the included trials.

The pooled signal favoured training closer to failure (lower RIR) for hypertrophy, but the magnitude was small. The practical translation the authors stressed: the size benefit of pushing nearer to failure is real but modest, and adding a set is often a larger lever for growth than shaving a rep off the reserve. The review flagged the same limitation that haunts the field, that RIR was frequently estimated from how studies described their protocols rather than measured set by set.^[5]

The reframe: Robinson 2024 meta-regression

Robinson and colleagues stopped treating proximity as a category and modelled it as a continuous predictor, running a series of meta-regressions on estimated RIR against strength and hypertrophy outcomes.^[1] This is the methodological pivot that the headline of this article points at.

• Hypertrophy: in every best-fit model, the marginal slope for estimated RIR was negative, and its confidence interval did not contain zero. Muscle size increased as sets were terminated closer to failure. Growth is a graded function of proximity, not a step that switches on at failure.
• Strength: in every best-fit model, the RIR slope confidence interval contained zero. Strength gains were similar across a wide range of RIR. Proximity to failure is a weak lever for 1RM.
• Divergence: the dose-response curve for hypertrophy and the curve for strength point in different directions. One outcome cares about proximity; the other mostly does not.

The split is the actionable finding. If size is the goal, the curve says keep working sets near failure. If strength is the goal, the curve says proximity barely moves the needle, so the lifter is free to leave reps in reserve and bank the recovery for heavier loads.

Why the older verdict and the newer verdict both hold

There is no contradiction between Grgic 2022 and Robinson 2024. Grgic asked whether two bins differ on average and found they do not; the bins each spanned a range of RIR that smeared the gradient. Robinson kept the gradient intact and found a slope for hypertrophy. A binary test with low resolution and a continuous model on the same underlying data can honestly disagree, because they are measuring different things.^[1][3]

Refalo 2023 sits between the two: it pointed at proximity directly and saw a small hypertrophy advantage for lower RIR, foreshadowing the slope that the meta-regression later quantified as continuous.^[2]

The measurement problem

The biggest caveat across this literature is RIR estimation error. Lifters judge proximity well when a set is near failure and poorly when several reps remain; estimates taken at 4 or more RIR are noisy. Worse, many trials did not measure RIR at all, so reviewers assigned it from the written protocol.^[5] That introduces uncertainty into every dose-response point on the curve.

Two practical consequences follow. First, prescribing a target like "2 RIR" assumes the lifter can hit it, which is more reliable close to failure than far from it. Second, the steepness of the hypertrophy curve is estimated, not measured directly, so treat the direction as solid and the exact slope as provisional.^[1]

A controlled check: Refalo 2024

A free-standing trial helps ground the meta-analytic curve. Refalo and colleagues ran an eight-week within-subject study in 18 resistance-trained adults, with each lifter's legs randomised so one trained the leg press and leg extension to momentary failure and the other left 1-2 reps in reserve.^[4] Quadriceps hypertrophy came out similar between the failure and reserve conditions. That is consistent with the meta-regression read: near failure and at failure land close together for size, while true failure adds fatigue without a proportional growth payoff.

Ruple 2023 reported a comparable pattern in 19 previously trained adults, comparing a low-RIR group cued to end sets at 0-1 RIR against a high-RIR group held at 4-6 RIR.^[6] Strength and hypertrophy were similar between the two, though the high-RIR group was further from failure than most "non-failure" prescriptions. The trials and the meta-regression tell a coherent story: the last few reps before failure do most of the work.

Programmable RIR targets

Translating the 2022 to 2024 evidence into set-by-set prescriptions:

Hypertrophy priority
  Target RIR:        0-2 on most working sets
  Last set of move:  push to 0-1 RIR (the steep end of the curve)
  Isolation work:    0-1 RIR (low systemic cost, safe to push)
  Heavy compounds:   1-2 RIR (technique + joint risk near failure)
  Volume note:       adding a set often beats shaving the last RIR

Strength priority
  Target RIR:        2-4 on most working sets
  Top singles/doubles: 1-3 RIR, never grind to a stall
  Rationale:         proximity is a weak lever for 1RM;
                     bar speed and load matter more
  Fatigue:           reserve recovery for heavier sessions

Mixed / general fitness
  Compounds:         2-3 RIR
  Accessories:       0-2 RIR
  Deload weeks:      raise RIR by ~2 across the board

Where the evidence still leaves room

• Exact slope. The hypertrophy direction holds across models; the precise growth per RIR is estimated from imperfectly measured proximity and should not be quoted as a fixed number.^[1]
• Training status. Grgic's trained subgroup hinted that experienced lifters may need to push closer to failure to keep the size stimulus high. The interaction is suggestive, not settled.^[3]
• Exercise type. Pushing a leg press to 0 RIR is low-risk; pushing a heavy squat or deadlift to 0 RIR raises technique and joint cost. The curve does not price that risk.
• Long horizons. Most trials run 6 to 12 weeks. The fatigue cost of chronic failure training over a full year is largely extrapolated.

Cross-link tools

• Junk Volume Detector flags sets too far from failure to drive growth, the practical inverse of the proximity question.
• Workout Volume Calculator manages the set count that often outweighs a single rep of reserve for hypertrophy.
• 1RM Calculator tracks the strength outcome that proximity to failure barely moves.
• Progressive Overload Planner sets the load progression that drives strength more than RIR does.

Related reading: Junk Volume: Reading Hypertrophy From Your Log for the volume side of the stimulus, The Schoenfeld Volume Meta and Junk Sets for how set quality is judged, and Evidence-Based Programming 2026 for the broader periodisation frame. For where proximity to failure ranks among the other growth levers, see the full framework in How to Build Muscle: The Evidence-Based Levers.

• Grgic 2022 found no overall failure vs non-failure difference for strength or size; the binary framing hid the gradient.
• Refalo 2023 saw a small hypertrophy advantage for training closer to failure.
• Robinson 2024 modelled proximity as continuous: hypertrophy rose as sets neared failure; strength was similar across a wide RIR range.
• The hypertrophy and strength dose-response curves diverge, which is the actionable result.
• RIR estimation error is the standing caveat; treat the direction as solid and the exact slope as provisional.