How the programme works

Ruleset 2026.09.27

Anybody can record a workout, and the free apps do it well. What none of them will tell you is why the third set is three sets and not five. This page is that answer, in full, including the parts where the engine deliberately does less than it could.

Nothing is made up at the point of use

Every number in your programme comes from a named rule, and every rule carries the evidence behind it. When the evidence moves, the rule moves and the version above changes — so any week this app has ever written can be traced back to the rules in force the day it was written. A generator that hard-codes “three sets of ten” somewhere in a loop cannot tell you why, and cannot be corrected when the field moves on.

Volume, counted honestly

Sets are counted per muscle, per week. Direct work counts as one set. Indirect work counts as half: a barbell row is a set for your back and half a set for your biceps, because that is closer to what actually happened than either counting it fully or ignoring it. Most apps count exercises, or credit a set to one muscle only — which is how somebody rows four times a week, watches their arms stall, and is told their biceps volume is zero.

Below the maintenance figure a muscle loses what it has. Above the ceiling more sets are not impossible, just a poor use of the same minutes — and the engine has to justify the time against an alternative, which it usually cannot.

Pelland et al. 2026 meta-regression: the volume slope exceeds zero with 100% probability for both hypertrophy and strength, with diminishing returns. Baz-Valle 2022: 12–20 sets is standard for trained lifters, with no reliable advantage above 20.

Hard where fatigue is cheap

Effort is prescribed per exercise, not per session. A set of lateral raises taken to failure costs almost nothing that matters. A set of squats taken to failure costs the rest of the session and some of the week. Heavy compounds are held at two to three reps in reserve; isolation work goes to nearly nothing. That one idea is what lets a 35-minute session be genuinely productive.

Schoenfeld 2017 and Lopez 2021: hypertrophy is comparable across loading ranges, while 1RM strength clearly favours heavy loads. Robinson et al. 2024: hypertrophy rises as sets approach failure, with the slope flattening past roughly 2 reps in reserve.

Your time is a hard constraint

Every set is costed in minutes before a session is written, and one that does not fit the time you have is rebuilt rather than handed over and hoped about. Somebody with 35 minutes three times a week is not running the same programme as somebody with 90 minutes five times a week, shortened. It is a different programme, because the first cannot afford long rests on heavy lifts and the second can.

Progression you can predict

Fill the rep range at the prescribed effort and the weight goes up. Judged on every working set, not the average, because one strong first set followed by three that fall apart is not a completed range — and averaging it hands you a weight you cannot hold next week. Grind three sessions running at the same load and the app says so and backs off, rather than pushing you into a wall.

Blocks and deloads

Five weeks, occasionally six. Volume climbs while effort holds roughly steady, then an easier week arrives before you need it. This is a deliberate departure from chasing the most volume you can recover from: that is right for a bodybuilder with a controlled life, and wrong for almost everybody else.

A deload is also pulled forward when two main lifts stall, when three readiness checks come back poor, or when you tell the app your next fortnight is difficult. That last one is the most predictive input there is, because you know a house move is coming and no algorithm does.

The session adjusts to the day

Three taps before you start: sleep, soreness, how you feel. A good day runs as written. A middling one holds the weights, cuts an accessory and eases the effort. A bad one becomes about twenty minutes of maintenance — and still counts as done, because an app that punishes you for training sensibly on a bad day is teaching you to skip instead.

Connect Health Connect or Apple Health and your sleep, steps and weight fill themselves in, so the readiness check has your week behind it rather than only the answers you gave this morning. What you tell it still decides: if you say you feel wrecked, you feel wrecked.

What it deliberately does not do

• It does not use AI to write programmes. The engine is deterministic and can be audited. A language model producing set counts cannot be checked, corrected systematically, or asked why.
• It does not diagnose anything. Every figure is compared with your own history and nobody else’s. None of it is a medical assessment and none of it is presented as one.
• It does not overrule a coach. If you train with a coach, nothing here changes your week without them asking. The automatic parts are for people training on their own.
• It does not guess when it does not know. Where your history is too thin to judge, it says so rather than producing a confident number from noise.