TrainTwice grades your week and builds your next one from a fixed set of rules. This page is all of them: what each rule assumes, what it does to your plan, how well the evidence supports the assumption, when it last changed and what changed. Where the evidence is thin, the row says so — a rules page that labels everything “proven” is not labelling anything.
What the evidence labels mean
StrongConsistent results across several studies, including meta-analyses. The direction and the rough size of the effect both hold up.
MixedThe direction holds, but the size is unsettled, the studies disagree, or the effect shows in one outcome and not another.
LimitedA principle or a coaching convention with no trial that isolates it. We say so here rather than dress it up with a citation it does not have.
ContestedCredible evidence points both ways. We pick the reading that is safer to be wrong about.
How we read the literature
Three places where the research does not say what a fitness app would like it to say. Each one is a choice, and a choice can be argued with.
Autoregulation is not better, so we only autoregulate one thing
The meta-analysis finds autoregulated loading roughly equivalent to a standard fixed prescription — not superior. So the app does not tell you what to load. You pick the weight, the app corrects what it means, and the thing that autoregulates is volume: targets climb when your own numbers climb and stall when they stall.
The frequency meta-analyses do not show two sessions beating one when weekly volume is held equal. Frequency matters because it decides how much weekly work you can fit and recover from. We hold the frequency constant at two and let the volume be the variable.
Gamification effects on activity are significant in studies under 12 weeks and gone at 12 weeks or more. The app has XP, levels, streaks, quests and badges, and uses them for adherence. No rule on this page consults them, and no part of your program depends on them — a covered week is a covered week whether or not it was celebrated.
How much work each muscle gets, how a set is counted, and what the app does with the effort you report.
Weekly volume is the lever
Strong
The claim
Adding weekly sets per muscle adds growth, and the return on each extra set falls away fast — the curve is closer to a square root than to a straight line, and it does not fall off a cliff at some threshold.
What it does to your plan
Sets the weekly target every muscle is programmed against and carries 70 of the 100 points in your program grade. Volume is counted in fractional sets and rewarded up to 31 a week.
The volume check carries 70 points of the grade; spacing and fatigue carry 15 each. Nothing outside these three moves a letter.
Reward follows the square root of volume up to the 31-set ceiling. The meta-regressions show the curve still rising there, so 31 is where our reward stops, not where the evidence says growth does.
Under ~4 sets a week is not training the muscle
Strong
The claim
A muscle given fewer than about 4 fractional sets a week sits below the range in which the dose-response work shows a meaningful effect.
What it does to your plan
Any grading-relevant muscle under 4 fractional sets a week is a critical issue: it costs volume points, headlines the week's summary, and caps the letter — one hole cannot grade above B−, two or more cannot clear C−. The builder will not ship a plan that leaves one.
S2 (2026-10-06) made every point the grade loses belong to a named driver, so a muscle under the floor now shows as a row with the threshold it crossed and what it costs, instead of a silent deduction.
The cap binds where the blended grade is computed — one hole clamps it to 0.78 and two or more to 0.65 — so it holds however efficient the plan is against its own constraints, and the week says plainly which letter it cannot pass.
How a set is counted
Mixed
The claim
A hard set counts fully for its primary muscle and partially for its secondary movers, and effort matters: a set finished far from failure is worth less than one finished close to it.
What it does to your plan
Converts every logged set into fractional sets per muscle before comparing it with a target — secondary movers earn a fraction, and the fraction shrinks as the reported reps-in-reserve grows. One definition is shared by the grade, the weekly contract and the analytics, so a set cannot count one way in the grade and another way on the contract.
Why it is labelled mixed
Counting direct and indirect sets this way is standard in the volume literature. The exact fraction assigned to secondary work is a judgement call, and we treat it as one.
One fractional-set model (effort-discounted, session-capped) feeds the grade, the weekly contract and the analytics. The contract adds no second opinion about what a set is worth.
Reported effort is corrected, not trusted
Mixed
The claim
Trainees overestimate how close to failure they are — 'two in reserve' is usually more than two. Load estimated from un-corrected effort is optimistic.
What it does to your plan
Shifts logged reps-in-reserve by a bias learned from your own sessions, and corrects estimated one-rep-maxes to a true-failure equivalent before a load is compared with last week or allowed to drive a progression.
Why it is labelled mixed
The direction is well established and accuracy improves with experience. How far any one person's numbers are off is not known, so the app learns yours from your logs instead of assuming it.
One correction is applied everywhere a load is compared or carried forward: progression, carryover estimates and the analytics all read the same corrected figure.
Rep ranges follow the movement, not the muscle
Mixed
The claim
Hypertrophy is driven mainly by how close a set finishes to failure rather than the rep range it is performed in — but very costly compounds are better kept in low reps, and small muscles tolerate higher ones.
What it does to your plan
Prescribes 5–8 reps for maximal-fatigue compounds (squats, hinges, heavy presses), 12–20 for calves and abs, and 8–12 for everything else.
Why it is labelled mixed
The proximity-to-failure meta-regression is strong evidence. The specific ranges are coaching practice layered on top of it.
Fatigue cost, not muscle size, picks the low-rep band, so the heaviest movements are never prescribed at 12 reps.
A session has a ceiling too
Mixed
The claim
A single session stops adding return for a muscle well before the day runs out of exercises to fit — somewhere around 11 fractional sets for that muscle — and a session can only hold what its time budget allows.
What it does to your plan
Sizes the day from your session length at 3 minutes a working set plus a 5-minute warm-up, at 2.5 sets per exercise slot, with 2–4 sets per exercise. Nothing past the per-session point of diminishing returns is credited to the grade, so one long session cannot buy a muscle its whole week.
Why it is labelled mixed
The per-session meta-regression behind the 11-set point is a preprint and has not been peer-reviewed. It is also the least load-bearing number here, because the builder rarely approaches it.
Slot budget is sized off the real 2-set floor rather than a flat 3 sets per slot. At 3 sets the sixth slot was the last one kept on a 60-minute day, which dropped the core and biceps work under the minimum effective dose.
Frequency and recovery
How the week's work is spread, and what the app treats as too close together to recover from.
Twice a week is a way to spread volume
Strong
The claim
Training a muscle twice a week does not by itself grow more muscle than training it once — the meta-analyses show frequency matters mainly through the weekly volume it lets you fit. Two sessions is how we distribute the volume, not a magic dose.
What it does to your plan
Programs every muscle on at least two days of the week, requires the two sessions for a muscle to sit at least two calendar days apart on a four-day week, and spreads core and calves the same way.
Holding every muscle to two sessions is what makes ordinary weekly volume reachable inside ordinary session lengths.
S1 (2026-10-06) put the distribution on the screen as the weekly coverage contract, so the claim is visible per muscle rather than asserted once.
Back-to-back days for the same muscle
Limited
The claim
A muscle trained on two consecutive days gets no recovery window between them, so the second session adds fatigue on top of work that has not been repaired yet.
What it does to your plan
Charges 34% of the spacing check for each pair of consecutive days that share a primary muscle, and the week screen flags the pair — both days, every shared muscle — offering the split that clears it. Nothing is ever reordered on your behalf.
Why it is labelled limited
No trial isolates back-to-back same-muscle sessions, so this is a recovery principle and practitioner consensus rather than a demonstrated effect. The page says so; the app still never applies a fix without you.
Last changed·No study isolates this rule, and none is cited for it.
Changelog (2)
S3 (2026-10-06): the collision became visible wherever a day can move, with the split that clears it offered as a choice. The check itself is unchanged — it can no longer be graded and never shown.
An earlier version only fired when one session trained a muscle past the ~11-set session point. No realistic programme reaches that — the largest session measured in the app was 9.03 sets — so two upper days back to back scored the same as a proper alternating split. That gate was removed rather than re-tuned.
Fatigue is systemic as well as local
Limited
The claim
A day carrying far more than its share of the week, two maximal lower-body sessions on consecutive days, and a maximal hinge left on an upper-body day all cost recovery even when no single muscle is trained twice.
What it does to your plan
A day above 1.6× the week's average fatigue units costs 25% of the fatigue check; two maximal lower-body days back to back cost another 25%; a maximal hinge on an upper-body day costs 30% and is treated as critical. Fatigue carries 15 of the 100 points in the grade.
Why it is labelled limited
The least-studied rule on this page: it encodes standard practice about spreading heavy work and keeping heavy hinges on the days that train the muscles they load.
Last changed·No study isolates this rule, and none is cited for it.
Changelog (1)
Kept deliberately separate from the spacing rule. A heavy squat day and a heavy deadlift day can share no primary muscle at all, so spacing sees nothing while the athlete is still squatting on top of a deadlift session.
Progression
How the targets move from one week to the next, and when they are deliberately held back.
The ceiling is discovered, not assumed
Mixed
The claim
The volume a trainee responds best to is individual, and part of that response depends on what they were doing before the block started — so the right ceiling cannot be read off the literature.
What it does to your plan
Starts every muscle at a floor — 6 fractional sets a week for the largest muscles, 4 for the rest — rather than at an assumed optimum, and climbs from there on your own numbers.
The block starts at the floor rather than at a target: the first roughly six sets a week deliver most of the return, so starting low costs little and buys recovery and adherence.
Ramp while the numbers move, stop when they don't
Mixed
The claim
Progressively adding sets across a block beats holding volume constant — though the trials are clearer about strength than about size.
What it does to your plan
Raises a muscle's weekly target by one set while its corrected best lift is still improving. Two consecutive flat weeks retire the muscle from the ramp for the block, and nothing ramps past 22 sets a week.
Why it is labelled mixed
The trial behind the ramp added sets every two weeks and showed a strength dose-response; for hypertrophy the differences between groups were not statistically clear, and the authors read their own confidence intervals as plateauing at higher volumes. We use it to progress, and we do not claim it is the thing making you grow.
A muscle with too few logged sessions neither ramps nor stalls — an unknown trend is not treated as a flat one.
Two calibration weeks, then a step back every sixth
Limited
The claim
The first two weeks are better spent learning the trainee's numbers than chasing volume, and a block should step back deliberately rather than only when it collapses.
What it does to your plan
Holds every muscle at its floor for weeks 1–2. Every sixth week halves the targets, never below 2 sets, and restores them the week after.
Why it is labelled limited
Calibration is a data-quality argument, not a physiological one, and the evidence for scheduled deloads is thin. We keep the deload small and say the evidence is thin instead of implying otherwise.
Last changed·No study isolates this rule, and none is cited for it.
Changelog (1)
A deload week never overwrites the stored ceilings, so the week after restores exactly the volume the block had built.
Selection, ordering and estimates
Which exercises a day carries, in what order, and the numbers the app pre-fills for you.
How the exercises are chosen and ordered
Limited
The claim
Training a muscle from more than one angle, putting compounds before isolation, and not repeating one isolation exercise twice in a week make a plan more robust — as coaching convention, not as a demonstrated effect.
What it does to your plan
Shapes the day: compound-first ordering, a different angle for each of a muscle's two weekly sessions, core and calves spread across the week, and isolation variety wherever a second option exists.
Why it is labelled limited
Convention. The app labels it that way instead of dressing it in citations it does not have.
Last changed·No study isolates this rule, and none is cited for it.
Changelog (1)
Applied as a deterministic pass over the plan, so an AI-built week and a hand-built week are held to the same rules.
Starting loads are estimates from your own history
Mixed
The claim
Strength carries over between exercises that train the same muscle roughly in proportion to what each movement typically allows, so a sensible starting load can be estimated from lifts you have already done.
What it does to your plan
Pre-fills a new exercise's sheet from your carryover estimate for that muscle, scaled by the ratio of the two movements' load indexes and personalised by your own measured ratio whenever both movements appear in your history. Estimates are never written to your history — they fill a blank sheet, and every field stays overwritable.
Last changed·No study isolates this rule, and none is cited for it.
Changelog (1)
Estimates are never persisted: they pre-fill a sheet and are recomputed from history each time, so a corrected log corrects the estimate.
Not every muscle counts equally
Limited
The claim
Muscles differ in how much they matter to the outcome a trainee is chasing, so a week that leaves forearms short is a smaller problem than one that leaves quads short.
What it does to your plan
Gives each muscle a weight. A shortfall in a heavy muscle costs more of the grade than one in an incidental muscle, and an incidental muscle never headlines the week's summary.
Why it is labelled limited
This is our editorial judgement, not a finding. It is the one number on this page that no study supports, and it is labelled accordingly.
Last changed·No study isolates this rule, and none is cited for it.
Changelog (1)
Incidental muscles still contribute to the volume score; they are excluded from the headline sentence only.
What we do not claim
We do not claim this split is optimal. The evidence supports the volume distribution and the recovery rules; it does not rank arrangements of days, and the app only offers a reordering when a reordering clears a specific problem.
We do not claim the grade predicts your results. It grades the plan, not you, and it is not a forecast of how much muscle you will add.
We do not claim the numbers here are thresholds in the literature. Four sets, 31 sets, 11 sets, 22 sets and the 1.6× day are the shape of a curve applied to a programming problem — the studies report slopes and confidence intervals, not cut-offs.
We do not claim anything about injury, rehabilitation, nutrition or medical conditions. TrainTwice offers general training guidance, not medical advice.
We do not claim a citation for every row on this page. Where the evidence is convention or judgement, the row says so.
Peer-reviewed · The weekly volume curve: more sets per muscle a week produces more growth, with sharply diminishing returns; and that frequency is largely a means of distributing volume.
Preprint — not peer-reviewed · The per-session point of diminishing returns — the only reference on this page that has not been through peer review, which is why it carries no rule on its own.
Peer-reviewed · Why the app autoregulates volume and not load: autoregulated loading is equivalent to a fixed prescription, not superior to it.
If a rule here is wrong
Then the app is wrong with it, and the fix starts here rather than in a changelog nobody reads. Every rule above is applied by the same code the grade runs, and each row records what it changed and when. Send the argument to hello@traintwice.app.