The whole algorithm, written out

The loop

That loop is the whole coach. A rough check-in can hold step 3 back for a day — skipping the check-in never does.

When it takes weight off

A deload week sits above all of it and freezes the weight on its own.

And how many sets you do

That is the whole thing. Everything below is one of those steps in detail.

01Double progression

Every lift carries a rep range. You keep the same weight until you can hit the top of that range on every working set — then the weight goes up and the reps fall back toward the bottom. That drop is the mechanism working, not a bad session, and the engine is built so it never reads as one.

This is the part every tracker leaves to you. The set is stored either way; what changes is whether anything reads four outings of it back and says which of the two things just happened.

02The gate

Two conditions, both required, no exceptions:

A high score with one set short of the range holds the load. The score can never open the gate on its own — it is advisory, and the gate is binding. That order matters more than anything else on this page: a model that can talk itself into a load increase is a model that will, on the week you slept five hours a night.

03Rep ratio

Where your average set sat inside the range — 1.00 at the top, 0.90 at the bottom. Not distance from the ceiling: the bottom of the range is exactly where the last load increase put you, so it has to score as normal work. On a 6–10 lift:

Anchored the other way — measured against the ceiling — every class scores under the floor just by sitting at the bottom of its own range, and the engine prescribes a load cut on a session that did exactly what it asked for. That is not hypothetical: it is what this one did before it was fixed, on 300 of 598 evaluations of a real four-year log.

04Recovery

The optional check-in, and the only other term in the score. Each answer pulls the multiplier by a fixed amount, then the whole thing is clamped so no single bad night can wipe out a session:

Skipping is free. An unanswered check-in returns exactly 1.000 rather than a guess, so it can never quietly cost you an increase — which is why “Skip and start” is a peer of “Start workout” and not a link underneath it.

05What it decides

The steps at the top, in the order they are actually tested, with what triggers each one. The heading is the word the app will put on the lift.

Adding weight is never counted as going backwards. Both back-off lines require the load to have stayed the same or dropped, so a session where you added a plate and lost a rep reads as progress. Without that guard the engine punishes the exact session it just asked for — a machine row went 176 lb × 23 reps, then 190 lb × 22, and was told to back off to 176.

06Ranges, rest and the step up

Every movement belongs to a class, and the class sets the defaults before any of your own overrides:

Rest is a floor, not a target. Under about a minute costs you growth; past 90 seconds nothing further is bought — the longer numbers exist to keep your later sets from falling apart. All of it is per exercise: 3:30 after a heavy squat and 1:30 after leg extensions is one setting, made once, on the exercise rather than on the session.

The step is what the equipment actually allows, not a percentage, and it depends on the unit as much as on the machine — a pound gym has no 2.5 kg step, so prescribing one tells a 225 lb bench to load 231.5 lb. An increase moves one increment, once. Override it per exercise if your stack is pinned differently or you own microplates.

07Effort

How hard each set is meant to be, in reps left in the tank:

RPE 10 is 0 RIR, not failure. A rep you missed is logged as a miss and is a different thing. Calling 10 “fail” is how you end up with a log full of 9s and a coach quietly under-loading you for a month.

Self-reported effort is also the least reliable number in the model — trained lifters routinely think they are two reps further from failure than they are — so it corroborates a decision and never triggers one alone. That is why the −5% cut needs the reps to agree with the RIR before it fires.

08Volume

Load is the weakest of the levers. Hard sets per muscle per week is the one that moves growth most, and yours are measured against three landmarks — the middle one being a band rather than a line:

At around a dozen sets a week each further set is still buying growth, with the return shrinking as you climb. Every muscle is listed, zeros included — a muscle you have not trained cannot show up in a count of what you trained, which is the whole reason calves go missing for a year.

Then it names the muscle with room and the exercise that should absorb the next set: always one the coach is already progressing, never one it is backing off, with ties broken toward whatever carries the fewest sets.

09Training days and time off

Your plan is a rotation, not a week — an ordered list of workouts with no weekdays attached, so a missed day costs you nothing but the day. Training days is the one number that turns a rotation into a week, and every weekly figure above scales off it:

Left on Auto it is counted from your last four weeks rather than guessed, and the app says which of the two it used. Read the rotation total straight and you understate every muscle by up to 40%, which is enough to hide a group sitting past MRV.

Come back from a real break and the prescription waiting for you is the one written the day you last trained. It is adjusted when you open it — never written into your history — and it takes two reasons, because on a long rotation a ten-day gap between outings is simply your normal cadence:

Nothing happens under two weeks. Bodyweight lifts have no load to take off, so they only ever hold.

10Deload

Fatigue markers trigger a deload; the calendar only ever asks. Two reasons have to stack and at most one of them can be the date, so a deload always carries a real signal:

A deload dropped into the middle of a block purely because it was due produced no growth benefit and measurably cost strength. That is why a due date alone will suggest one and never announce one — and why the week itself holds the load and cuts the volume rather than the other way round. Intensity is what maintains the strength you are trying to keep; volume is what is making you tired.

11What it does not know

The list every app in this category leaves off its own page:

• Technique. Nothing collects it, so it is not in the score. It used to be, sitting at a constant and pretending to measure something.
• Your weekly volume, per lift. The per-exercise decision is made in isolation, so it can add load to a muscle already past MRV. The volume screen is where the two get reconciled, by you.
• Assisted movements. An assisted pull-up logs help, so progress runs down — the engine reads a heavier set as a better one and coaches those two lifts backwards.
• Whether you took the rest. Nothing records it, so no decision can depend on it.
• Your actual one-rep max. An estimated 1RM fits a curve to a set you did. Three formulas disagree by up to 8% — only compare figures from the same one, which is why the app names which it used.

12Where it runs

Your prescriptions are computed on the server, against your whole training log rather than the slice your phone is carrying. A second copy of the same engine ships inside the app and is what coaches you with no signal — the two are held to the same rules on purpose, and the set you log in a basement is queued on the device and settled later.

Weight is stored in kilograms end to end, which is why switching a lift to pounds changes what you read and never what is recorded. Four years of history follows the switch, including the halves.