How it decides.
The rules, written down.

• Calculated in code

  The minimums scale to your week

  The table above is for four lifting days. Lift three days and every number is multiplied by 0.75, five by 1.25, six or more by 1.3, and never by less than 0.65, then rounded. Beginners get 70% of that. Abs and obliques have minimums only if you train core, which is the default.

• Rule for the AI

  Tagging the muscles

  The AI tags each exercise with up to three muscles, main muscle first, from a fixed pattern list: a press is chest, front delts and triceps; a row is upper back, lats and biceps; a pulldown is lats and biceps.

• Enforced in code

  Counting the sets

  Code does the counting. A set is worth 1 to the first muscle tagged and 0.5 to each of the others. Tags outside the list of 18 muscle names are dropped, and an exercise left with no recognized tag fails the check.

• Calculated in code

  Sharing the week out

  Each muscle's minimum is shared across the days that train it. Every day the AI gets a table: for each muscle, what earlier days this week delivered, what later days hold, and the least and the most today may add.

• Enforced in code

  Today's minimum

  After the AI writes, code adds the session up muscle by muscle. Any muscle more than half a set short of today's minimum fails the check.

• Enforced in code

  The weekly ceiling

  A muscle whose week would go past 22 effective sets (35 for abs and obliques) fails the check. Most lifters keep gaining up to roughly 20 hard sets a week; past that, returns shrink and recovery costs climb.

• Rule for the AI

  About 11 per muscle per session

  Past about 11 effective sets for one muscle in one session (10 for calves), extra sets add little. Code uses this cap to share out the week, and each day's table gives the AI its maximum for every muscle. The audit checks it; code checks the weekly ceiling instead, except on the day before a class.

• Calculated in code

  A missed day is not a debt

  Earlier days this week count as done: their logged sets, or the prescription where nothing was logged, or the day's planned share if it was never built. Missed work is never crammed into the days that remain.

• Calculated in code

  When the schedule leaves no room

  If a muscle can't reach its minimum on the days available, its target for the week comes down and the table says why. The usual case is hamstrings when their only other day falls before a class. See it in the sample week.

• Enforced in code

  Too few

  A main lift counts if it's a distinct movement with at least 2 working sets. Fewer main lifts or ab exercises than your minimum fails the check.

• Enforced in code

  Too many

  More than your minimum plus two main lifts, and code removes extras from the end: never one of the first two, and never one that today's muscle minimums depend on.

• Enforced in code

  Fitting your time

  Code estimates every session: 8 minutes of warm-up, then 35 seconds per working set plus the rest prescribed after it. More than 10% over your time and it trims: sets off accessory movements first (never below 2), then whole accessory movements, and only then sets off the two lead lifts. Never below your counts or today's minimums, and the session says what was trimmed. With no time limit, nothing is.

• Enforced in code

  Rest is never what gives

  Trimming takes sets, never rest. Rest written on paper that you can't actually take is how a session quietly loses its quality.

• Rule for the AI

  The shape of a session

  2 to 4 working sets per movement. The first two movements are the day's lead lifts, straight sets at full rest. After that, movements that don't compete can be paired as supersets.

• Enforced in code

  Supersets

  A superset's 0-second rest stands only if its partner trains a different main muscle, the pair isn't two ab exercises, and the first movement has no more sets than the second. Otherwise code gives the rest back.

1. 1

   Draft

   AI

   Writes the session from your logs, this week's table, your goals, schedule, equipment, injuries and any photos, with the evidence reference attached.

2. 2

   Tidy

   Code

   Drops muscle tags that aren't on the list, undoes supersets that don't qualify, and removes main lifts over your maximum.

3. 3

   Check

   Code

   Counts lifts and ab exercises, checks every muscle against today's minimum and the weekly ceiling, and applies the day-before-a-class rules.

4. 4

   Audit

   AI

   A second pass re-reads the draft against an 11-point checklist, is told exactly what the code check failed, and corrects it. Its note says what it changed. On by default as Deep review; you can switch it off.

5. 5

   Check again

   Code

   The audited session is tidied and checked the same way.

6. 6

   Repair

   AI

   If anything still fails, one more pass gets the list of failures and is told to fix only those. Its version is kept only if it fails on fewer points.

7. 7

   Verdict

   Code

   Trimmed to your time, checked once more, and printed at the top of the session: “Meets your requirements”, or “Short of your requirements” with what is short.

• Calculated in code

  Finding stalled lifts

  Code reads your logs for any movement whose best set, weight and reps, was identical in its last two sessions, and hands the AI that list as a stagnation watch. Sets logged with one tap from the prescription are marked assumed and ignored here.

• Rule for the AI

  Double progression

  Hit the top of the rep range on every set and the load goes up, back to the bottom of the range. Hit the middle and the load holds while the reps climb. Every repeated movement must be harder than last time.

• Rule for the AI

  Stalled: jump or swap

  Every lift on the stagnation watch gets a concrete jump in load or reps, or is swapped for another variation for the same muscle, and its note says which.

• Rule for the AI

  Missed twice, swapped

  Miss a rep target on the same exercise twice in a row and it is swapped for a machine equivalent or a different variation.

• Rule for the AI

  Small steps

  Upper-body isolation may only move 2.5 to 5 lb at a time. Forcing bigger jumps breaks form and stalls the lift.

• Rule for the AI

  Assumed sets prove nothing

  A set filled in with one tap counts as done, but never as proof you can go heavier. If every set last time was assumed, the prescription repeats and the note asks for your real numbers.

• Rule for the AI

  No invented first loads

  The first time you do a movement, there is no number to progress from, so none is made up. The target reads “Find it: a weight you could lift for 3 more reps”. The same rule applies when you swap an exercise.

• Rule for the AI

  Rotation

  Variations rotate every 4 to 6 weeks, but never away from a lift that is still progressing.

• Rule for the AI

  Reps in reserve

  Most working sets end 1 to 3 reps short of failure. The last set of a machine or isolation movement may go to technical failure. Free-weight compounds never go to outright failure.

• Rule for the AI

  Rep ranges

  5 to 10 for the first compound while you're fresh, 8 to 15 for most of the work, 12 to 20 for isolation and finishers.

• Rule for the AI

  Rest between sets

  2 to 3 minutes after heavy compounds, 60 to 90 seconds on isolation.

• Rule for the AI

  Order and selection

  The hardest compound goes first, isolation later. Movements that load a muscle in its stretched position are preferred, as are machines and cables for stable, joint-friendly work. Redundant pairs are cut.

• Calculated in code

  Conservative mode switches on

  The health check at intake switches on conservative mode when any answer flags a health condition, pregnancy, or a birth in the past six months.

• Rule for the AI

  Conservative mode rules

  Every set stops at least 3 reps short and nothing goes to failure. No heavy low-rep work: main lifts stay at 8 reps or more. Machines and supported positions are favoured, reps go up before load, and the session suggests checking in with your doctor before pushing hard. In pregnancy, also no long spells flat on the back, no heavy bracing or breath-holding, nothing with a fall or impact risk, and no heavy loaded ab work.

• Calculated in code

  Your split

  Code picks the split from how many days you lift, in the order of your week, as in the table. From two days up, each muscle is trained about twice a week.

• Enforced in code

  Lower days stay lower

  Lower days carry leg work plus only delts, traps and forearms. Chest, lats, upper back, biceps or triceps as an exercise's main muscle on a lower day fails the check. The day before a class is the exception: the upper body fills it.

• Rule for the AI

  Upper days stay upper

  Upper, push and pull days carry no lower-body work, and each day does its own job rather than another day's.

• Rule for the AI

  Covering the gaps

  Hamstrings get both a hip-extension and a knee-flexion movement across the week, and arms get some direct work. A weak-points day goes to the two or three muscles furthest behind.

• Calculated in code

  Which week it is

  Code counts weeks from the Monday of your first session: weeks 1 to 3, then a lighter week 4, then the next block. The session header shows it.

• Rule for the AI

  Weeks 1 to 3

  Push progressive overload against your recent numbers.

• Rule for the AI

  Week 4

  Keeps the movements and lowers the stress, as set out above.

• Enforced in code

  Still checked

  A lighter week is checked against your counts and muscle minimums like any other.

• Calculated in code

  Trends know the difference

  Your trends page shows lighter weeks but leaves them out of every comparison, so a planned easy week never reads as a decline.

• Calculated in code

  The hinge day moves

  If your hamstring-and-hinge day would land the day before a class, it trades places with a lower or full-body day that doesn't. Same sessions, different order.

• Enforced in code

  No hamstrings, calves or hinges

  The day before a class, no exercise may have hamstrings or calves as its main muscle, and no hinge may appear: back extensions, glute-ham raises, Nordics, reverse hypers, pull-throughs, Romanian deadlifts, good mornings or deadlifts.

• Enforced in code

  Light legs

  That day quads get at most 9 effective sets, glutes 4 and adductors 3, with a set and a half of slack because those arrive in halves from compound lifts. Hamstrings and calves are held to zero, with half a set of slack.

• Rule for the AI

  How the legs are trained

  Legs first while fresh: quads only, 2 to 3 machine movements you've done before, about 8 to 9 sets at 2 to 3 reps in reserve, nothing to failure, no slow eccentrics, no jumping. The rest of the session is upper-body work at moderate effort.

• Calculated in code

  A class is not lifting

  Classes never count toward your weekly sets, and a class day gets guidance for the class rather than a lifting session.

See a session the engine wrote for the day before a class

• Rule for the AI

  Hard constraints

  Whatever you write at intake under injuries, or movements you must avoid, is given to the AI as absolute: it is told never to prescribe it under any implement, variation or alternate name, and to replace it with a movement for the same muscles. A banned squat or hinge becomes its machine alternative.

• Rule for the AI

  Swaps too

  Every exercise swap is written under the same constraints.

• Calculated in code

  Protein

  2.2 g per kg of bodyweight, rounded to 5 g. That sits comfortably above the roughly 1.6 g per kg where research puts the plateau for muscle gained from training.

• Calculated in code

  Fat, carbohydrate, fibre

  Fat 0.8 g per kg, as a floor. Carbohydrate takes whatever calories remain, because it fuels training. Fibre is 14 g per 1,000 kcal.

• Calculated in code

  Gaining muscle

  Aim for 0.25% of bodyweight a week and never more than 0.5 lb, with the surplus held between 5% and 10% over maintenance. Faster than that on a trained lifter is mostly fat.

• Calculated in code

  Strength, recomposition, general health

  Strength: 5% over maintenance. Recomposition and general health: maintenance.

• Calculated in code

  Losing fat

  0.5% of bodyweight a week, a rate chosen to keep muscle. The deficit is capped at 25% of maintenance, and the target never goes below the higher of your BMR and 1,200 kcal (female equation) or 1,500 kcal (male).

• Calculated in code

  Reading the scale

  Bodyweight moves 2 to 4 lb a day on water and food alone, so no decision rests on a single weigh-in. The trend is the average of the last 7 days of weigh-ins against the 7 days before, and needs at least 3 weigh-ins in each.

• Calculated in code

  Calorie suggestions

  None until at least 8 weigh-ins and a trend. Losing fat: if the trend isn't falling, it suggests 150 kcal less (never below the floor); if you're losing more than 1% of bodyweight a week, 150 kcal more. Any other goal: if the trend is off target by 0.25 lb a week or more, it suggests closing the gap at 3,500 kcal per lb, rounded to 50 kcal and capped at 400 either way. A suggestion only: your targets never change on their own.

• Calculated in code

  Screening switches

  Answer yes to the intake question about eating and weight and there is no calorie deficit, no scale-driven suggestion or weight prompt, no weight numbers on the dashboard, and no photo analysis. Pregnant or within six months of giving birth: maintenance instead of a deficit, and no scale-driven suggestions, because the scale is supposed to move.

• Calculated in code

  What it reads

  Photos are optional. When you add them, up to your six most recent go to the AI with each session request, oldest to newest.

• Rule for the AI

  Weakest first

  The report opens with the single weakest thing it can see, stated plainly, not softened with praise.

• Rule for the AI

  What hasn't changed

  If a region looks the same as in your oldest photo, it says so, and for how long.

• Rule for the AI

  No flattery

  No “great work” or “keep it up”, and no improvement claimed unless something visibly differs between photos. If body fat rather than muscle is what limits the look you want, it says that too.

• Rule for the AI

  Honest about the camera

  Lighting, pump, angle, posture and time of day change how a body looks more than months of training. When photos differ too much to compare, it says so instead of inventing a change. No body-fat percentages or measurements, which a photo can't give.

• Rule for the AI

  Ranked, then trained

  Three to five lagging areas, worst first. Each session targets the top one that fits the day, and that exercise's note names the finding. If none fit, the strategy names the day that will.

• Rule for the AI

  Audited without the photos

  The audit strips flattery, unsupported improvements and body-fat figures. It never sees the photos, so it is told not to rewrite the findings otherwise.

• Enforced in code

  Off when it should be

  If your intake answers say food or weight is a sensitive subject, photos are never sent to the AI and any photo commentary is blanked.

1. Schoenfeld BJ, Ogborn D, Krieger JW (2016). Effects of resistance training frequency on measures of muscle hypertrophy: a systematic review and meta-analysis. Sports Medicine 46(11): 1689–1697. doi:10.1007/s40279-016-0543-8

   Used for: Training each muscle about twice a week.

2. Pelland JC, Remmert JF, Robinson ZP, Hinson SR, Zourdos MC (2026). The resistance training dose response: meta-regressions exploring the effects of weekly volume and frequency on muscle hypertrophy and strength gains. Sports Medicine 56(2): 481–505. Published online December 2025; first posted as a SportRxiv preprint in 2024. doi:10.1007/s40279-025-02344-w

   Used for: The cap of about 11 effective sets per muscle per session.

3. Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO (1990). A new predictive equation for resting energy expenditure in healthy individuals. American Journal of Clinical Nutrition 51(2): 241–247. doi:10.1093/ajcn/51.2.241

   Used for: Resting energy (BMR), the base of every calorie target.

Code comments also name Schoenfeld 2019, on keeping each session's dose for a muscle productive, and Schoenfeld again beside the per-session cap. They don't say which papers, so rather than guess, no reference is given for them here.

• It's an AI, and it can be wrong

  It can misjudge a load, tag a muscle wrongly, misread a photo or miss a constraint. The rules above make that less likely. They don't make it impossible.

• Code checks arithmetic, not judgement

  It can count sets against your minimums. It can't tell whether a load suits you today, or whether an exercise is right for an injury described in words.

• The audit isn't always there

  You can switch it off, and it's skipped when time runs short. The session always says when it wasn't audited.

• Not medical advice

  Stop if anything hurts. The health check makes coaching more conservative; it doesn't make it supervised. If you have a condition, talk to your doctor.

• No professional reviews your program

  Chris is not a certified trainer or a dietitian, and nobody checks sessions by hand. Every session is written and checked by software, and the AI is told never to claim to be a person or to hold a qualification.