How we calculate variation loads

The model turns a competition 1RM into the load for a variation by multiplying four things: the RPE percentage for your target reps, a stance coefficient for the variation itself, an execution coefficient covering the pauses and the tempo, and a personal multiplier calibrated on what you actually lift.

This page describes the structure and the reasoning. It does not publish the individual coefficients: the app applies them, calibrates them against your own lifts, and the typical ranges are public. The structure is the part worth arguing about, and it is the part that is easy to get wrong.

The shape of the calculation

Four independent factors, applied in order:

1. The RPE percentage. Standard, and not ours — the widely circulated RTS-style table that maps reps × RPE to a percentage of your max.
2. The stance. What the variation costs before you move: a front squat, a deficit pull or a board press changes where the bar sits and how far it travels. This factor already contains the range-of-motion difference, which is why there is no separate ROM term — that would count the same effect twice, and it is the most common error in models of this kind.
3. The execution. The pauses, the descent and the ascent, combined into one number. Everything about how you perform the rep lives here.
4. Your own multiplier. A per-lifter, per-variation correction, learned from a set you actually completed.

Why the factors multiply instead of adding

Each factor acts on what is left after the previous one, not on the original bar weight. Adding the penalties would double-count them, and with enough of them stacked the load would fall to zero or below — which is obviously wrong, because a front squat with a three-second pause is hard, not impossible.

Concretely: if a variation costs you 20% and a pause costs 10%, the pause does not remove 10% of your competition max — it removes 10% of the weight you can already handle in that variation. Multiplying keeps every factor proportional to the load that actually reaches it. Adding treats them as independent taxes on a number that no longer exists.

Why the cost of a pause is concave

The first second of a pause costs the most, and every second after it costs progressively less. The cost rises with the square root of the duration rather than in a straight line.

The reason is that a pause and a long pause take away different things. Almost everything a pause removes is gone the moment the bar settles: the stretch reflex, the elastic rebound, the momentum you would otherwise carry out of the bottom. What the following seconds add is fatigue, and fatigue accumulates far more slowly than the rebound disappears.

Every lifter recognises this from under the bar: the step from no pause to a one-second pause is enormous, and the step from two seconds to three is noticeable but small. A linear model gets that backwards, and gets absurd quickly — at five seconds it would be predicting a weight nobody could fail to lift.

Where you stop matters as much as for how long

Pause cost depends on the position in the range of motion, not only on the duration. Each position carries its own difficulty, measured relative to the classic pause for that lift — the hole in a squat, the chest in a bench, just off the floor in a deadlift.

Two consequences that a duration-only model cannot produce:

• A suspended pause is harder than a supported one. Stopping a centimetre above the chest costs more than stopping on it. On the chest some load passes into the ribcage and the position is stable; suspended, the muscles hold everything in the worst leverage of the entire range. The same is true below the knee in a deadlift, which for most lifters is the real sticking point.
• A pause near lockout is nearly free. Stopping with the arms almost straight removes very little, because there was very little left to remove.

This is also why the model supports more than one pause in the same rep. "One second a centimetre off the chest, then one second on the chest" is a real prescription, and it is not the same as a two-second pause at the chest — the two stops cost separately, each with its own position and its own concave curve.

The competition bench is already a paused bench

In competition the bench press is judged with a visible pause at the chest of roughly one second. So a competition bench 1RM already includes that pause, and charging for it again would price the competition lift as harder than itself.

The model therefore treats the first second of a chest pause as free — but only that one, and only there. A second second costs normally, and a pause anywhere else costs from the first moment. This is the kind of detail that decides whether a model is describing powerlifting or describing arithmetic.

It has a consequence in the other direction too: a bench with no pause at all is a touch-and-go bench, which is easier than a competition bench because it gets the rebound back. The model has no way to represent that gain, so rather than produce a confidently wrong number it does not allow that combination to be built.

The descent and the ascent are not the same

A slow ascent is charged more per second than a slow descent. A slow descent is control — you resist while gravity does the work. A slow ascent is a refusal to accelerate exactly where the lift is hardest, so you spend the whole time you imposed on yourself inside the sticking point.

Both follow the same concave shape as a pause, and both cost less than a genuine stop of the same length: lowering slowly drains you and bleeds off elastic energy, but it does not eliminate the rebound the way a pause does, and it does not change your leverages.

This is the weakest claim on the page, and it is flagged as such below.

Why there is a personal multiplier at all

The ratio between a variation and the competition lift is highly individual. A lifter who fails at the chest and a lifter who fails at lockout will not have the same board press percentage, no matter how good the average is.

So the coefficients are honest starting points, not constants, and the model says so inside the app. What makes the number yours is calibration: you tell the app what you actually lifted on a variation, and it works backwards through the same chain to derive your own multiplier, then remembers it for that variation. After one honest set the average stops mattering.

This is also the answer to "my board is a different height" or "my box is lower than yours". Every gym's equipment is different, and no published table can fix that. Calibration can.

The four places this can be wrong

Published because a model that only lists its strengths is marketing. If you can break one of these, it is worth more to us than agreement.

1. The ascent-to-descent ratio is reasoning, not measurement. That a slow ascent costs more per second than a slow descent is defensible. How much more is a number we argued our way to, not one we measured. It is the first thing we would revise given real data.
2. The suspended pause is the least certain coefficient. That stopping a centimetre off the chest is harder than stopping on it matches what lifters report, but it is exactly the sort of claim that could be an artefact of who reports it. The same doubt applies below the knee in the deadlift.
3. Overload variations rest on thin evidence. Board presses and rack pulls sit above 100% of the competition lift, and the exact amount depends almost entirely on your lockout strength — the single most individual quality in the list. These are the coefficients where calibration matters most and the default matters least.
4. Beyond six reps the whole thing gets shakier. The RPE table is least reliable at high reps, and a long pause at rep eight is a fatigue event that the model's structure does not really describe. The app warns you when you enter that territory rather than pretending otherwise.

There is also a whole category the model does not attempt: anything where the variation changes the skill rather than the load. A first-time front squat is limited by upper-back position and wrist mobility, not by the coefficient. No multiplier fixes that, and none pretends to.

If you have real numbers

The most useful thing anyone can send is a set that actually happened: the lift, the variation, the pauses and tempo, the weight, the reps and an honest RPE — plus the competition max of the same lift. One of those is worth more than a page of opinion, this page included.

Coaches in particular see the cases we cannot: the same variation across twenty athletes, where the individual noise starts to cancel out. If you program variations and you have those numbers, we would like to be told where the model is wrong.

Frequently asked questions