Loading fundamentals 02: Reps in reserve

Short answer. End most sets when you think one to three good reps are left. You are probably about a rep further from failure than it feels, so trust the target rather than grinding an extra rep on a heavy barbell. Taking every set to failure roughly doubled the drop in bar speed after a session, and in one trial of the most experienced lifters yet tested it grew no more muscle than stopping a rep or two short. Pooled data lean slightly toward closer to failure for size, so the last set of a machine or cable exercise is the cheap place to go all the way.

What it is

RIR (reps in reserve: how many more repetitions you could have completed when you stopped the set) became the default currency of prescription because it adjusts to the day you are having in a way percentage loading cannot. Zero means the next rep would not have gone up with good form. It is also a guess.

The proposed mechanism

The case for failure is that the last repetitions of a set are the ones that matter. As fatigue accumulates, motor unit recruitment rises to compensate, so the final repetitions are where the highest-threshold fibers get their exposure.

The case against is that a repetition taken at the absolute limit costs more fatigue than the one before it, and the sets that follow pay for it.

What has actually been tested

What failure costs, within the session. Twenty-four trained men and women did six sets of bench press at 75% of one-rep max in three separate trials: every set to failure, every set stopped at 1 RIR, or every set at 3 RIR. Bar speed four minutes afterward was down 25% after failure, 13% after 1 RIR and 8% after 3 RIR (Refalo et al., 2023b).

From the first set to the last, failure lost 22% of its speed, against 9% at 1 RIR and 6% at 3 RIR. A day later, failure and 1 RIR were both still 3% down while 3 RIR was 2% up, and by 48 hours the differences were gone. A meta-analysis of crossover studies from an unconnected group found the same direction for fatigue, effort and a blood marker of muscle damage (Vieira et al., 2022).

What failure buys, over eight weeks. Eighteen trained men and women trained one leg to momentary failure and the other to a perceived 2 RIR on leg press and 1 RIR on leg extension, twice a week. Quadriceps thickness rose by a similar amount in both legs, and total volume across the eight weeks was similar too, even though it was free to differ. The failure leg lost more bar speed and more reps within its sessions throughout the block (Refalo et al., 2024).

These participants averaged nearly eight years of training, and half had competed in strength or physique sports: the highest reported training experience of any study comparing failure against non-failure. This is the group where an advantage for failure should most plausibly show, and here it did not.

What the pooled evidence says. Meta-regressions across the literature found muscle growth increased as sets ended closer to failure, while strength gains were similar across a wide range of RIR (Robinson et al., 2024). Two meta-analyses found at most a small edge for failure on growth: a small advantage in trained lifters only (Grgic et al., 2022), and a trivial one with all definitions of failure pooled but none under the strict one (Refalo et al., 2023a). In single-set training, 42 trained lifters saw several measures of growth tend to favor failure by modest margins, with strength similar (Hermann et al., 2025).

How well you can read the dial. Across twelve studies and 414 participants, people underpredicted the repetitions they had left by 0.95 on average (95% confidence interval, CI, 0.17 to 1.73), with considerable variation between studies (Halperin et al., 2022). Training status did not detectably change accuracy. The clearest effect was set length: predictions were much better in sets of twelve repetitions or fewer, and only slightly better close to failure and in later sets. In a long, burning set, the guess gets much worse. Experienced lifters in the eight-week trial were within about half a rep of a 1 RIR target (Refalo et al., 2024).

Our reading

Our position, and it is a position rather than a conclusion. Failure costs measurably more fatigue, and its advantage for size looks small at most. The null result rests on one eighteen-person trial, while a 42-person trial and the pooled data lean slightly the other way, so treat this as a default, not a rule: end most sets 1 to 3 reps short of failure.

The volume argument is weaker than it looks. Stopping short should save capacity for more total work, and both trials left rep counts free to show it. Within one session, 1 RIR banked slightly more total volume than failure, and failure matched 3 RIR (Refalo et al., 2023b); over eight weeks, volume was similar (Refalo et al., 2024). Failure banks reps early and gives them back later. Whether the capacity you save can be spent on extra sets that add growth has not been tested.

Where failure helped, volume was low. The clearest edge for failure came in single-set training (Hermann et al., 2025). A lifter already training near the volume they can recover from pays for failure's extra fatigue in every later set, so our reading is that its case shrinks as volume rises.

The injury trade is lopsided. Whether failure raises injury risk has not been measured, but the possible gain is small and one injury can cost weeks of training. On heavy lifts that trade is not worth taking, tested or not.

The measurement error argues for a target, not for aiming easy. Aim at 2 and you will probably land within a rep of it, inside the range that grew as much muscle as failure. Aim at 4 to be safe and you may be at 5, further from failure than the growth trial tested.

Where the goals diverge

Building muscle with moderate loads. Around 1 to 3 RIR on most work.

Building muscle with light, high-repetition sets. The same number does not transfer. Our reading is that in a set of fifteen, more of the useful stimulus sits in the final repetitions than in a set of eight, because full recruitment arrives later. Bodyweight movements and small-muscle isolation work fall here by default, and both warrant a lower RIR than the loaded compound work in the same session. These are also the sets where your guess is worst, so take them close to failure on equipment where failing is safe. (→ Loading fundamentals 04: Load and repetition range) Most sessions in OMNID's hypertrophy programs set a lower RIR on isolation work than on the compound lifts, which fits this reading.

Heavy strength work. Strength gains looked similar across a wide range of RIR, so grinding heavy sets to failure buys little, and a missed heavy rep is the costliest kind. Your read can be off by one, so set pins or use a spotter anyway.

The last set of an exercise. Failure costs least when no later exercise in the session trains the same muscle. Take it on machines, cables and single-joint dumbbell work. On barbell squats and presses, only with safety pins set or a spotter, and not on deadlifts, where form gives out before the bar stops.

Calibrating the dial. Every few weeks, call your RIR on a machine or cable set, then keep going until a rep stalls. The gap is your correction. If you have never been to failure, start here: it is hard to judge the distance to a point you have never reached.

Related. Velocity loss tracks the same fatigue with a device instead of a self-report. (→ Loading fundamentals 09: Movement velocity and intent) When the same load and reps leave more in reserve than the day asks for, add weight or reps. (→ Loading fundamentals 05: Progressive overload) Set endpoint, rest and how volume is spread are the three levers on how much quality work a week holds. (→ Loading fundamentals 03: Training frequency; Loading fundamentals sub-hub)

The case against

The null growth result is one trial of eighteen people, well designed and unusually trained, on one muscle group.

The pooled evidence leans the other way for size, though the RIR values in those meta-regressions were estimates (Robinson et al., 2024). Read our position as consistent with a small advantage to closer proximity.

Most of these sources have overlapping authorship. The fatigue meta-analysis comes from an unconnected group and agrees on the cost side. The growth result needs the same kind of independent replication.

Injury is the reason people give, and it is the least studied. The documented costs are fatigue, effort and a muscle-damage marker, and the eight-week trial logged minor discomfort in eight participants (five failure, three RIR) and one strain on an RIR leg, too few to read (Refalo et al., 2024).

What is not known

Whether failure's extra fatigue, which stayed greater throughout an eight-week block, changes outcomes over longer blocks, at higher set volumes, in older lifters or on other movements.

Whether spending the capacity you save on extra sets adds growth. That trial has not been run.

Whether the accuracy problem matters: no trial has manipulated it and measured the muscle.