How the Gravitus Strength Standards Are Computed

Where does the data come from?

Gravitus lifters log workouts in the app. For each lifter and each exercise, the app keeps one all-time best set, the set with the highest estimated one-rep max that lifter has ever recorded for that movement. Those best sets are the raw material for the standards.

Three rules limit what enters the pool. A workout the lifter marked private never contributes a best effort, so private training stays private. An account flagged for abuse is excluded entirely. A lifter with no recorded sex is excluded, because every table is published separately for men and women.

Which sets count?

One set counts per lifter per exercise: that lifter's all-time best. A lifter who benches twice a week for five years contributes exactly one number, the same as a lifter who benched once. This keeps the distribution a picture of lifters rather than a picture of training volume, and it stops the most active accounts from dominating a bracket.

Sets from other exercises never spill into a lift. A close-grip bench press belongs to the close-grip bench press, not to the bench press.

How does a set become a one-rep max?

Gravitus converts every set to an estimated one-rep max with the Brzycki formula:

estimated 1RM = weight × 36 / (37 − reps)

When the lifter logged an RPE for the set, the app first adds the reps that lifter left in reserve, so reps becomes reps + (10 − RPE). A set of 5 at RPE 8 is treated as a set of 7, because two more reps were available.

The rep count is then capped at 10. Brzycki's denominator shrinks as reps rise, so the formula inflates rapidly in high-rep territory and a set of 30 light reps would otherwise produce an absurd one-rep max. The cap makes a high-rep set estimate conservatively rather than wildly.

The one-rep max calculator uses the same formula, so a number you calculate there is directly comparable to the tables.

Which numbers get thrown out?

Logged data contains typos, unit mix-ups and joke entries. Two filters remove them.

First, hard plausibility bounds. A record is dropped when its estimated one-rep max falls outside 20 lb to 1,500 lb, when the lifter's bodyweight falls outside 60 lb to 500 lb, or when the set's rep count falls outside 1 to 500.

Second, a trim. Within each sex, Gravitus removes the lowest 0.5% and the highest 0.5% of the surviving one-rep maxes before computing anything. The trim is small on purpose: it clears the tails where mistakes concentrate without reshaping the distribution that real lifters produce.

How are lifters grouped by bodyweight?

Strength scales with bodyweight, so every table is published per bodyweight bracket. Brackets are 10 lb wide and each lifter is assigned to the nearest one, so a 176 lb lifter and a 183 lb lifter both land in the 180 lb bracket.

Published brackets run from 120 lb to 300 lb for men and from 100 lb to 200 lb for women. Lifters outside those ranges are pooled into the nearest end bracket rather than dropped.

How are levels defined?

Each of the five strength levels is a fixed percentile of the real distribution inside a bracket. The levels are not tiers someone chose; they are positions in the data.

The population is lifters who track their training in Gravitus, not the general public. Intermediate means average among people who log workouts.

How much data does a table need before it is published?

Thin data produces confident-looking numbers that mean nothing, so three floors govern what gets published.

• 50 lifters per bracket. A bodyweight bracket with fewer than 50 qualifying lifters is not measured directly. Its row is interpolated instead.
• 200 lifters per sex. An exercise needs at least 200 qualifying lifters of a given sex before Gravitus publishes a table for that sex at all.
• 500 lifters per sex. A continuous level curve, the smooth fit that lets the calculator place any bodyweight rather than only the bracket midpoints, is only fitted above 500 lifters.

An exercise that clears none of these floors gets no published standards page.

What happens at bodyweights with too few lifters?

Very light and very heavy brackets are always the thinnest. Rather than publish noise or leave gaps, Gravitus extends the measured curve to those brackets with allometric scaling, the standard method for comparing strength across body sizes:

strength = reference strength × (bodyweight / reference bodyweight)^0.67

The reference point is the nearest bodyweight where real data exists, and the exponent is 0.67, the two-thirds value the square-cube law predicts for how cross-sectional muscle area scales with body mass.

Every row produced this way is marked as interpolated and carries an asterisk on the standards tables, so you can always tell a measured number from an extended one.

How do the "average" pages differ?

The average lift pages report the measured median only, which is the 50th percentile of the same trimmed distribution. Nothing on those pages is interpolated or formula-extended: an average is published when the real data supports it and omitted when it does not. The by-bodyweight tables on those pages likewise use measured rows only.

How often do the numbers refresh?

A full incremental rebuild runs every Sunday at 08:00 UTC and refreshes every published table. Between those runs, an hourly job rebuilds exercises that have recorded new best efforts, with a seven-day cooldown per exercise so the popular lifts are not rebuilt continuously to move a percentile by nothing.

Each standards page shows the date its numbers were last computed.

What are the limits of this data?

Five limitations matter when you read or cite these numbers.

• The population selects itself. Everyone in the dataset chose to install an app and log their training. That group trains more deliberately than the general population, so these percentiles sit above what a random sample of gym-goers would produce.
• Only public workouts count. Lifters who keep their training private are absent, and nothing here establishes whether they differ from the lifters who share.
• The one-rep maxes are estimated, not tested. Every number comes from a submaximal set run through a formula. Formula estimates carry error, particularly for lifters whose rep strength and single-rep strength diverge.
• Sex is recorded as male or female only. The app offers no other option, so no other tables exist.
• There is no age adjustment. Gravitus does not require a birthday and too few accounts record one, so an age-split table would describe the accounts that volunteered a birthday rather than lifters of that age.

How do I cite the standards?

The standards are free to use and published under a Creative Commons Attribution license. Cite them as:

Gravitus strength standards, gravitus.com/strength-standards/, accessed <date>.

Please link to the specific page you took numbers from, and include the date, because the tables are rebuilt weekly and the figures move.

Ready to see where you stand? The strength standards calculator places any recent set on these tables in a few seconds.