Every load, every session and every rebuilt week comes out of a deterministic rule engine. Same inputs, same plan. This page is the engine's contract with you: the rules, where they come from, and how much weight the evidence can carry.
Most of the strength and health benefit of resistance training arrives inside roughly two hours a week (BJSM 2026 · Momma et al. 2022). Beyond that the curve flattens; below it, it falls fast. Your days times your minutes are checked against the band, not against an ideal.
The bad-week banner in onboarding tells you which side of the band you're on. The plan never asks for more than the band.
Set time estimate plus prescribed rest plus transitions is at most your window, tolerance −2/+0 minutes (Planned-duration invariant). Change the window and the session is rebuilt from the same seed. Actual elapsed time feeds a per-member pace factor.
The window stepper on Today. "Built for 23 min" in every session card.
Two consecutive sessions at the top of the rep range with RPE at or below 8 trigger a step (Halperin et al. · Helms et al. · Zourdos et al.). Your push setting sets the size: +2.5%, +5%, or +7.5% upper / +10% lower. Tendon-loaded patterns cap at +5%. Two stalls hold or drop 5%. Never to failure; two reps in reserve is the floor. Reported reps-in-reserve are treated as under-counted by one.
Log a set with RPE. Profile → How hard to push.
Short sleep or low energy eases load by 5 to 10% and drops one movement (Saw et al. 2016 · Craven et al. 2022). A sore muscle group routes the pattern around it for the session, and for two weeks if it repeats. The comparison is to your own rolling baseline, never a population norm, and never a rest-day prescription.
Check in · 3 taps on Today, then the eased session.
Up to seven days away changes nothing (Comeback ladder · Bickel et al. 2011). Eight to 21 days: 5 to 10% lighter, one more rep in reserve, one session. 22 to 56 days: about 40% in week one, ramped over half the gap. Longer: re-onboard at 50 to 65%. The return week never asks for more than one fewer session than your bad week, and no load after a gap exceeds the last one you logged.
Welcome back → What changed this week.
One hard set per pattern, once or twice a week, with load held and two to three reps in reserve maintains strength for up to eight weeks (Spiering et al. 2021 · Bickel et al. 2011). Then a two-week ramp back at two thirds volume.
Profile → Life mode. A 1 to 8 week stepper and a four-row contract.
The substitution ladder runs lighter, slower tempo, shorter range, different implement, different exercise in the same pattern (Substitution ladder · coverage invariant). Pattern coverage holds through every swap. Sharp or radiating stops the movement, rests the pattern for two weeks, and exits to "see a professional". There is no pain scale and the word injury does not appear in the app.
The comfort sheet during any set.
The elapsed time. Built to fit at standard rests. Your pace owns the clock.
A readiness score. Three taps shape tonight against your own baseline. No calibration period, no number.
Injury prevention. The comfort sheet eases, swaps or stops a movement. Sharp or radiating means see a professional.
Body composition. We plan strength and conditioning minutes. Nothing here measures fat, weight or a "biological age".
A composite fitness score. The quarterly capability check stores four raw numbers and compares them to your last quarter only.
The week re-plan, progression, comeback and life mode are deterministic rule engines. Every rule above is covered by an invariant test that runs on every build. A language model is used in exactly one place: reading the free-text answer you type during onboarding. Nothing that produces a load, a session or a week goes through a model.
That is what lets the app say "nothing invented", and what lets this page exist at all: a rule can be written down and checked. A black box can't.
Read the research behind each rule ›