Loading fundamentals 11: Estimated and tested maxes

Short answer. A tested one-rep max is a reliable measurement. The case for estimating it from your working sets is that a tested number describes one day, cannot tell you how hard a percentage will feel, and has not been shown to produce better strength gains than loading by effort. Our position is to run on the estimate, show its error, and test a true max only when you need the real number.

What it is

Your one-repetition maximum, or 1RM, is the heaviest load you can lift once with full range and sound technique. Most loading schemes are written as percentages of it, which means every one of them depends on a number somebody has to find.

There are two ways to find it. Test it: work up to a single all-out attempt. Estimate it: take a set you have already done, with its load, its repetitions and how close it came to failure, and calculate the single that set implies. A number found the second way is usually called an e1RM, an estimated one-repetition maximum. (→ Loading fundamentals 04: Load and repetition range)

This entry is about which of the two a program should run on, and what each one gets wrong.

The proposed mechanism

A max is a moving target. It changes as you adapt, which is the point of training, and it changes from day to day with fatigue, sleep and everything else that affects readiness. A tested number is a measurement of one day. The further you get from that day, the less a percentage of it means. (→ Loading fundamentals 05: Progressive overload)

A percentage does not mean the same effort for everyone. Two people lifting the same fraction of their max can be very different distances from failure, because the number of repetitions a given percentage allows differs between people and between exercises. A percentage prescription is an effort prescription with an unknown error attached.

An estimate trades one error for another. It needs no maximal attempt and can be recalculated every session, but it depends on an equation and, where proximity to failure is part of the input, on a self-report. (→ Loading fundamentals 02: Reps in reserve)

What has actually been tested

A tested 1RM is a reliable measurement. A systematic review of 32 test-retest studies, 1,595 participants in all, found a median intra-class correlation of 0.97, a reliability score on which 1 is perfect agreement (Grgic et al., 2020). The median coefficient of variation, the typical test-to-test difference as a share of the mean, was 4.2%. Reliability was high regardless of training experience, familiarization, exercise, body region, sex or age. Whatever the case against testing is, it is not that the test gives a different answer each time.

The repetitions a percentage allows vary between people, most at lighter loads. A meta-regression of 952 repetitions-to-failure tests from 269 studies estimated not only the average number of repetitions at each percentage of 1RM but how widely individuals spread around it. At 80% of 1RM the between-individual standard deviation was about two and a half repetitions; at 60% it was over four. Sex, age and training status did not clearly change the relationship, but the exercise did: the bench press and the leg press each needed a table of their own (Nuzzo et al., 2024). The authors note the spread may partly be measurement error. Either way, a loading table describes an average person.

Estimates from bar speed tend to run high. An individual participant data meta-analysis of 20 studies and 434 participants found moderate predictive validity for load-velocity estimates, with a standard error of about 10% of the tested 1RM (95% confidence interval, CI, 7.4 to 12.2%). Whichever modeling approach was used, predictions overestimated the tested 1RM by about 4.5 kg on average (95% CI 1.5 to 7.4 kg). Its authors recommend direct assessment wherever possible, and see the method as useful for tracking trends across a training cycle (Greig et al., 2023). (→ Loading fundamentals 09: Movement velocity and intent)

Adjusting loads to the day has not been shown to beat fixed percentages. Across six trials in resistance-trained people, autoregulated load prescription by effort ratings or by velocity produced about 2 kg more 1RM strength than percentage-based prescription, with an interval that crosses zero (95% CI -0.32 to 4.46 kg) (Hickmott et al., 2022). The authors read that as similar improvements, and so do we. It is an absence of a difference, not a demonstration that the two are equal.

Maximal loads are handled routinely in one population: its injuries are rare per hour and common per person. A systematic review put injury incidence in powerlifting at 1.0 to 4.4 per 1,000 hours of training, similar to other non-contact strength and power sports and low against contact sports (Aasa et al., 2017). It noted that most of the nine included studies were of low methodological quality. Its update confirmed the incidence and added that, when an injury is defined as pain that impairs training, one report found 70% of powerlifters injured at the time they were asked (Tung et al., 2024). Neither review separates maximal attempts from the rest of the training.

Our reading

The case for estimating is not that testing is dangerous or inaccurate. The test is reliable, and injuries per hour among people who handle maximal loads as their sport are low. The prevalence figure is a reminder that low per hour is not the same as rare per lifter (Tung et al., 2024).

The case is what a tested number cannot do. It describes the day you tested, and you are training to make it wrong. It cannot tell you how hard a percentage will feel, because that differs from person to person and exercise to exercise (Nuzzo et al., 2024). And the one thing it buys, a more exact number, has not been shown to buy better strength gains than loading by effort (Hickmott et al., 2022).

The part we are least sure of is the part people argue about most: injury. A maximal attempt is the set with the least margin for error. We could find no study that compares the injury risk of a maximal attempt with a submaximal set, lift for lift. The low incidence comes from experienced competitive lifters, and a person training alone without a spotter is not one. So we treat the added risk of a solo max as a reason, not as a finding.

So the estimate is the default, and its error is shown rather than hidden. An e1RM is not more accurate than a tested 1RM. It is available every session, costs nothing, and moves with you. Its error is managed two ways. Estimates come only from sets that ended close to failure, with the margin widened the more repetitions an estimate is stretched across. And the day's load is set against an effort target, not against the number alone.

This is one of the reasons OMNID estimates maxima from logged sets, draws each estimate with its margin, and tests a true single only in a program built to peak for one.

Where the goals diverge

Competition. If you compete, you need the real number on the day. An estimate is how you get there, not a substitute for the platform.

Hypertrophy. The max barely matters: what counts is how close to failure the sets are, and that can be prescribed without one. (→ Loading fundamentals 04: Load and repetition range)

Newer lifters. A tested max should go out of date fastest here, and technique at a maximal load is least settled. We have not seen this tested, but both point toward estimating.

The case against

The strongest argument against us comes from the most rigorous source here. The individual participant data meta-analysis recommends direct 1RM testing wherever possible (Greig et al., 2023). It was written about bar-speed estimates, but its logic applies to any estimate: if you want the number, measure the number.

The test is more reliable than its reputation (Grgic et al., 2020). The grounds for avoiding it are practical rather than physiological.

Loading by effort has not been shown to be superior (Hickmott et al., 2022). If our only argument were better gains, it would fail.

Three of these six sources come from one overlapping author network in resistance training research, so they are less independent of each other than six separate groups would be.

Self-reported effort has its own error, and an estimate built on a report of how many repetitions were left inherits it. (→ Loading fundamentals 02: Reps in reserve)

The margin we draw around an estimate is our model, built from how far the equation is reaching and an allowance for a one-repetition reporting error. It is not a validated interval from a study.

What is not known

Whether a maximal attempt carries more acute injury risk than a submaximal set, lift for lift. We found injury rates by sport and by population, not by the intensity of the attempt.

How quickly a tested 1RM stops describing a trained lifter who keeps training.

Whether estimates from sets adjusted for reps in reserve are more accurate than classic repetitions-to-failure equations when the lifter is unsupervised.