You Probably Have More Reps Left Than You Think: Why RIR Estimates Are Off — 3-5 Reps If You're New, 1-2 If You're Trained

Yes — lifters systematically underestimate how many reps they have left, and the error runs in one specific direction: you almost always have more in the tank than you think. How much more depends on who's doing the guessing. A lifter in their first few months who believes they've hit failure typically still has 3 to 5 clean reps left. A trained lifter in the same position is much closer to the edge — on the order of 1 to 2. Either way the mistake isn't random noise that cancels out. It's a systematic undershoot, and it's quietly capping the results of millions of hard-looking sets.

Picture the newer lifter who finishes a set of leg extensions, drops the load with a grimace, and says "that was everything." Fresh again 3 minutes later, he could have done 5 more. I've watched that happen more times than I can count. The set looked maximal. It wasn't. Let me show you exactly how big the gap is, who it's biggest for, why your own nervous system lies to you about it, and how an objective measurement quietly fixes the problem. If you need the term itself defined first, start with what reps in reserve means and come back.

Are beginners bad at estimating reps in reserve?

Yes — and in the one large dataset that ranks lifters by experience, much worse than veterans. The single cleanest study is Steele and colleagues (2017), who had 141 trainees predict the number of reps they could complete to momentary failure, then made them actually grind to failure to see who was right. Everyone underpredicted. The size of the miss is where the experience gradient shows up.

Averaged across the 6 exercises, the least experienced group — under 6 weeks of training — underpredicted their true capacity by about 4.7 repetitions. Beginners at 1.5 to 6 months were off by 3.6. Then the gradient keeps falling: 1.9 reps at 6 to 12 months, 1.5 at 1 to 3 years, and 0.8 for experts with more than 3 years under the bar. The whole-sample average was about 2.0. Note the direction throughout: under-prediction. Every group could do more than they guessed.

Precision tracked experience too, and this is where the study is most often misquoted. The frequently cited "2.6 to 3.4 reps" figures are standard errors of measurement for the combined sample, not anyone's average bias. Broken out by group they fall steadily with training age — on chest press, from 4.3 reps in the newest lifters down to 1.6 in the experts. Exercise mattered enormously as well: leg press was the worst lift in every single group, at 8.5 reps of underprediction in novices against 2.8 on elbow flexion. Any single "reps short" number that doesn't name its lift is hiding a 2-to-3-fold spread.

So the "3-5 reps short" figure that gets repeated everywhere is real — but it belongs to the people in that study who had trained for less than 6 months, 36 of the 141 participants. It is not a general fact about lifters, and applying it to someone with a few years of training overstates their error by roughly threefold. Pool 12 studies and 414 participants and the average underprediction is just 0.95 reps (confidence interval 0.17-1.73), with heterogeneity so extreme (I² = 97.9%) that even that number is a rough centre of gravity rather than a constant (Halperin et al., 2022). And here's the qualifier almost nobody carries over: in that meta-analysis, training status did not significantly moderate accuracy. Neither did it in Hackett's 81-lifter dataset, nor in Remmert's. Steele's experience gradient is the strongest single sample we have, but it hasn't reproduced between studies — treat it as suggestive, not settled.

Think you're at 2 RIR: how big is the gap really?

The gap is large enough to change what your set is actually doing — but it's a range, not a constant, and where you land in it depends mostly on training age and on how far from failure you're making the call.

Start with the closest thing we have to a behavioural datapoint. A trained lifter who stops at self-perceived failure — what feels like 0 RIR — still has reps left: Armes and colleagues (2020) found trained participants stopped about 2.0 reps short of true momentary failure on knee extensions even when they believed they'd reached their limit. That study is honest about its own uncertainty, and you should be too: the confidence interval runs from 0.0 to 4.0, across 38 lifters in two deception experiments on a single-joint machine. Read it as "roughly a couple," not as exactly 2. For someone in their first few months, Steele's data puts that same perceived-failure gap at 3 to 5.

Then there's the rule that replicates far better than any headline number: your error grows with distance from failure. Hackett and colleagues (2017), testing 81 lifters on chest press and leg press, found roughly 1 rep of error when 0-5 reps actually remained, but more than 2 when 7-10 remained. Halperin's meta-regression found the same slope, plus a second one — error is small in sets of 12 reps or fewer and balloons in higher-rep sets.

Every figure in that table is a group average, and the spread between individuals is itself a finding — Halperin puts the between-lifter standard deviation at about 1.45 reps, so your personal bias can sit well outside these bands in either direction. The table tells you which way to lean, not what your number is.

That bottom row matters most. Estimates get worse the farther you are from failure. Your sense of "reps left" is decent in the last rep or two and turns to fog above 3 reps out — which is exactly the deep-reserve zone where light, high-rep "pump" sets live. The number you trust the least is the one you lean on the most.

Why does effort feel maxed before it actually is?

Because your brain rates the cost of the current rep, not the count of reps remaining — and those two things come apart under fatigue. Perceived effort climbs steeply in the back half of a set: the burn, the breath, the slowing bar, a training partner watching. All of that floods the "I'm done" signal long before the muscle's force-producing machinery actually quits.

This is the core mix-up that wrecks most RIR discussions. Effort perception and reps-remaining prediction are two different measurements, and they have opposite biases. Newer lifters tend to overrate effort — a hard-feeling set reads as RPE 9 or 10 — which means they underestimate how many reps they have left and stop with reps to spare. Say it in the direction that can't be misread: the error is thinking you have fewer reps in the tank than you really do. Zourdos and colleagues (2016) saw a version of this in 29 squatters: the experienced group moved slower and reported higher RPE at a true 1RM than the novices did, and the link between how fast the bar moved and how hard the rep felt was tighter in the experienced lifters (r ≈ −0.88) than in the novices (r ≈ −0.77). Worth naming the limit of that study, since it's often stretched past it: it validated an RPE scale against bar velocity. It never measured anyone's reps-to-failure error, so it can't tell you how many reps a beginner leaves behind — only that their sense of effort maps more loosely onto what the bar is doing.

There's a useful corollary here, and it's the reason this article keeps splitting numbers by population instead of quoting one. Remmert, Laurson and Zourdos (2023) tested 58 trained and untrained men and women on curls, pushdowns and rows and found that training experience and sex didn't significantly change RIR accuracy — what mattered was proximity to failure and set number. Predictions sharpened the closer people got to failure and in later, more fatigued sets. That's the third independent dataset, alongside Halperin's meta-regression and Hackett's 81 lifters, to find no training-status effect. So the fix isn't just "lift for more years." It's getting genuinely close to failure often enough to recalibrate what the edge feels like. If you stop reading what muscle failure actually feels like as "this is uncomfortable" and start reading it as "the bar physically stopped," your estimates tighten fast.

What does chronically sandbagging your sets cost you?

It costs you stimulus — the exact thing the set was supposed to buy. Muscle growth is dose-dependent on proximity to failure: the last few reps before failure, where motor-unit recruitment is highest and bar speed is lowest, carry disproportionate hypertrophic signal. If you're newer and your "2 RIR" is really a 4 or a 5, you're routinely leaving the most productive reps of every set on the table.

Do that across a whole training block and the math gets ugly. A lifter running, say, 16 working sets a week who's secretly 2 or 3 reps shy of their target on most of them is training closer to a maintenance dose than a growth dose — without ever feeling like they're slacking, because the sets feel hard. This is one of the quietest reasons people stall out. It often looks like a programming problem when it's really a proximity problem, and it's a prime suspect any time you find yourself not building muscle despite training hard. The flip side is real too: chasing failure on everything to "be safe" buries you in fatigue you don't need. The goal isn't max effort everywhere — it's knowing your true distance from failure so you can actually hit the target your goal calls for.

What objective cues do coaches use to check proximity to failure?

The main one is bar speed — how much a rep slows down across a set — because, unlike feel, it doesn't inflate. As a muscle fatigues, the bar physically decelerates, and that slowdown tracks proximity to failure in a way your perception doesn't. This is the entire premise of velocity-based training, and good coaches have used it for years as a sanity check on a lifter's self-reported RIR.

The numbers are clean and verified. Refalo and colleagues (2023) took 24 trained lifters through bench press and measured how much velocity dropped at different stopping points. The acute velocity loss they measured climbed with proximity to failure: roughly −25% after sets taken to true failure, about −13% after stopping at 1 RIR, and about −8% at 3 RIR. Across the whole session, velocity loss from the first set to the last was −22% to failure versus −9% at 1-RIR and −6% at 3-RIR. In other words, the magnitude of the slowdown is a quantitative fingerprint of how close you pushed — one that doesn't care how the set felt. Zourdos and colleagues found the same logic at the rep level in their squatters: bar velocity correlated strongly with reported effort, with the relationship tighter in experienced lifters (r ≈ −0.88) than novices (r ≈ −0.77). If you want those cutoffs laid out in one place, I collected them in velocity loss thresholds explained.

This is why velocity is the coach's lie detector. If a lifter says "2 RIR" but the bar already crawled to a near-stop, the velocity says they were at failure. If they say "0 RIR" and the bar is still flying, they stopped early. The bar slowing down is the thing your RIR is trying to perceive — measured directly instead of guessed.

Can velocity loss give you an unbiased RIR check?

It can give you an unbiased second opinion — which is exactly what a systematically biased guess needs. Velocity loss is the rare failure cue you can actually measure, and because a sensor has no ego, no burn, and no training partner to impress, it doesn't inflate effort the way your nervous system does. When your felt RIR and your rep-speed curve disagree, the velocity curve is usually the one telling the truth. That is the check you want: the set where you called "2 left" but moved like you hit failure, or stopped at "failure" with obvious speed to spare.

Here is where the watch on your wrist comes in. Riven is the Apple Watch app that scores muscle failure. It reads your wrist motion and stays quiet during the set; about three seconds after you rack the weight you get the exercise, the rep count and a 0-100 score. Reps slow down as a muscle nears failure, and that is what the score is built on. No barbell clip, no camera, no extra hardware. It shows you which set was the real one. If you want the deeper mechanics, I broke down why your reps slow down at the end of a set separately.

One honest caveat, because it is the whole point of trusting a number: velocity is complementary to feel, not a universal replacement. Paulsen and colleagues (2025) tracked 19 well-trained lifters through 6 weeks of squat and bench press — 2,972 measured reps — and found that mean bar velocity explained on average only about 30% of the variance in their perceived RIR, with individual correlations running the whole way from r = 0.1 to r = 0.9. Read that as a group average hiding very different lifters, not as a fact about you. The same study found perceived RIR was systematically pushed around by exercise type, load, velocity-loss threshold and set number, which is a stronger argument for measuring something objective than any single accuracy figure. One thing that study can't tell you, and is constantly misread as telling you: those lifters stopped at pre-set velocity-loss thresholds and never went to failure, so it measures how loosely perception tracks an objective signal — not whether anyone's RIR was correct. None of that makes velocity useless. It makes it an objective second opinion that beats guessing — and guessing, with an undershoot that runs 3 to 5 reps deep in your first months and 1 to 2 even after years under the bar, is what almost everyone is doing.

How to actually fix your RIR this week

You don't fix a systematic undershoot by trying harder to feel it. You fix it by recalibrating against reality. Here's the protocol I give people:

1. Take 2-3 sets to true failure on a safe machine. Leg extension, chest press, or a Smith-machine movement — somewhere a stuck rep can't hurt you. Go until the bar genuinely stops moving. This resets your reference for what 0 RIR actually feels like, and the "edge" almost always sits further out than you remembered.
2. Re-rate your normal "2 RIR" right after. Now that you've met real failure, do a working set, call your RIR, then keep going to see how many you actually had. If you're in your first months of training, expect to discover your old "2" was a 4 or 5. If you've trained for years, expect a smaller gap — but expect one.
3. Test it on the lifts you'd never suspect. Leg press produced 2 to 3 times the error of elbow flexion in Steele's data, in every experience group. Squats and presses deserve the audit more than curls do.
4. Trust feel most on heavy, low-rep sets. That's where self-estimates are sharpest, and it's the one condition the meta-analysis backs directly — error is small in sets of 12 reps or fewer. On 80%-plus work, your read is fairly reliable.
5. Distrust feel most on light, high-rep work. Deep-reserve, 15-plus-rep sets are where the guess collapses regardless of experience. Either push noticeably closer than feels necessary, or use an objective check.
6. Watch your rep speed on the last 2-3 reps. If they're still moving fast, you had more in the tank. A near-stalled rep means you're genuinely close. This is the manual version of what a velocity tool automates.
7. Cross-check the disagreements. When your felt RIR and an objective read part ways, log it. Over a few weeks you'll learn your personal bias — which, given a between-lifter spread of about 1.45 reps, is the only number that actually applies to you.

Do this and your "2 RIR" stops being a hopeful guess and starts being a calibrated target. If you want that turned into a week-by-week plan, the full 6-week version — anchor sets, predict-then-test drills, and rep slowdown as the objective check — is in how to calibrate your reps in reserve. And for the 3 ways of measuring it in the first place, I put together a complete guide to measuring your reps in reserve that builds on this.

FAQ

Do beginners underestimate or overestimate their reps in reserve?

They underestimate their capacity — they stop too early. In Steele and colleagues' 141-person sample, the least experienced lifters underpredicted their reps to failure by about 4.7 reps and beginners under 6 months by about 3.6, while advanced lifters were off by roughly 1.5 and experts by about 0.8. The direction is a consistent undershoot at every level: you almost always have more reps left than you think. The size of the gap is where experience shows — though see the last question, because that experience effect is less settled than it looks.

If a set feels like failure, am I at failure?

Usually not, but how far off you are depends on your training age. Trained lifters who stop at self-perceived failure averaged about 2 real reps left in Armes and colleagues (2020) — with a confidence interval from 0.0 to 4.0 reps, on knee extension, so treat "about a couple" as the honest reading. For lifters in their first 6 months, Steele's data puts that gap at 3 to 5. Perceived effort spikes before the muscle's actual force-producing capacity quits, so "felt maximal" and "was maximal" are different events, and the gap widens on high-rep sets.

How much do reps slow down near failure?

A lot, and predictably. In Refalo and colleagues' bench-press data, acute velocity loss was about 25% after sets to true failure, roughly 13% after stopping at 1 RIR, and about 8% at 3 RIR. That measurable slowdown is why bar speed works as an objective proximity-to-failure cue when your feel is unreliable.

Can an Apple Watch actually check my RIR?

It can give you an objective second opinion, which is what a biased guess needs. Riven is the Apple Watch app that scores muscle failure; it scores each set the moment you rack it and shows you which set was the real one. The caveat is real: in 19 well-trained lifters, bar velocity explained on average only about 30% of the variance in perceived RIR, so a rep-speed read is a complement to feel, not a hard cutoff. But against a guess that runs 3 to 5 reps short in newer lifters and 1 to 2 in trained ones, an unbiased check is a clear upgrade.

Will my RIR accuracy improve if I just train for years?

Probably somewhat, but the evidence is weaker than people assume. Experience helped inside Steele's sample, where experts averaged under 1 rep of error against nearly 5 in the newest lifters. But 3 other datasets — Halperin's meta-analysis of 414 participants, Hackett's 81 lifters, and Remmert's 58 — each found training status did not significantly change accuracy. What did replicate in all of them is proximity: predictions sharpen as you close on failure. Years may help; deliberate calibration definitely does.

Sources

• Steele et al. (2017), Ability to predict repetitions to momentary failure is not perfectly accurate, though improves with resistance training experience, PeerJ — https://pmc.ncbi.nlm.nih.gov/articles/PMC5712461/
• Halperin et al. (2022), Accuracy in Predicting Repetitions to Task Failure in Resistance Exercise: A Scoping Review and Exploratory Meta-analysis, Sports Medicine — https://pubmed.ncbi.nlm.nih.gov/34542869/
• Hackett et al. (2017), Accuracy in Estimating Repetitions to Failure During Resistance Exercise, Journal of Strength and Conditioning Research — https://pubmed.ncbi.nlm.nih.gov/27787474/
• Armes et al. (2020), Ability to Predict Proximity to Task Failure in Resistance Trained Persons, Frontiers in Psychology — https://pmc.ncbi.nlm.nih.gov/articles/PMC7785525/
• Refalo et al. (2023), Influence of Resistance Training Proximity-to-Failure on Neuromuscular Fatigue in Resistance-Trained Males and Females, Sports Medicine - Open — https://pmc.ncbi.nlm.nih.gov/articles/PMC9908800/
• Zourdos et al. (2016), Novel Resistance Training-Specific Rating of Perceived Exertion Scale Measuring Repetitions in Reserve, Journal of Strength and Conditioning Research — https://pubmed.ncbi.nlm.nih.gov/26049792/
• Remmert, Laurson & Zourdos (2023), Accuracy of Predicted Intraset Repetitions in Reserve (RIR) in Single- and Multi-Joint Resistance Exercises Among Trained and Untrained Men and Women, Perceptual and Motor Skills — https://pubmed.ncbi.nlm.nih.gov/37036795/
• Paulsen et al. (2025), Exercise type, training load, velocity loss threshold, and sets affect the relationship between lifting velocity and perceived repetitions in reserve, PeerJ — https://pmc.ncbi.nlm.nih.gov/articles/PMC12360324/