For a woman who reads the papers
In October 2026 the program was audited against the current research on women's strength, bone, balance, mobility and muscle growth, and rebuilt from the audit. This page is the long version: six chapters, 222 cited sources, and under every finding the rule it became in the program. Where the evidence is thin, indirect, from men or from animals, it says so.
Every citation here was read for the rebuild and is linked to its source. Population labels follow each reference: PRE is healthy premenopausal women, the group EVLVE is written for; YOUNG, POST, MEN, MIXED and ANIMAL are what they say. A claim that rests on animal or male-only data is marked in the text.
Chapter 1
This chapter covers the shape of an EVLVE session and of the training "week": what the longevity evidence actually asks of a woman in her thirties and forties, why the program counts four sessions in a rotation rather than a calendar week, why each session opens with impact and a balance task rather than an "activation" drill, what the ramp-up sets and stretches are for, and why a shorter session loses sets, not exercises. The conclusion: most of what longevity asks for lives inside the lifting itself, so the minutes that are not hard sets go only to what lifting cannot deliver.
The strongest reason to lift for longevity is a set of large observational cohorts, not a training study, and the honest version of the claim keeps that label. Pooling 16 cohort studies, muscle-strengthening activity was associated with a 10–17% lower risk of all-cause mortality, cardiovascular disease, cancer and diabetes, with the largest reduction at roughly 30–60 minutes a week and a J-shaped curve above that whose meaning the authors call unclear 1. In 412,413 US adults with a mean age of 44, muscle strengthening was associated with a hazard ratio for all-cause mortality of 0.81 in women against 0.89 in men, and for cardiovascular mortality 0.70 against 0.89; women's peak benefit appeared at one session a week and held or grew at two or more 2. These are self-reported, one-time measures, and the authors say causation cannot be presumed.
What the cohorts measure best is capacity, not minutes. Across 38 studies, higher handgrip strength was associated with 31% lower all-cause mortality overall and 40% lower in women 3. In 3,889 adults aged 46–75, the lowest category of relative muscle power carried a hazard ratio of 6.90 in women, against 1.71 for low strength, which did not reach significance; the authors argue power predicts mortality better than strength 4.
How EVLVE is built
The program does not claim its third and fourth sessions are a longevity necessity. Two sessions a week sit where the cohort curves flatten; sessions three and four exist for the physique and capacity goals, which are themselves the measured predictors. Because power predicts strongly and is the quality fatigue blunts, every session opens with a short impact block (30 light or 12–20 hard contacts, by rung) on fresh legs, serving bone and power at once. Copy says "associated with", never "proven", and "protect" bone, never "build".
When weekly sets are held equal, how they are split across days changes little. A review of 25 studies found strong evidence that frequency does not meaningfully affect muscle growth once volume is equated 5. The largest dose-response analysis (67 studies, 79% male, mean age 25) found weekly volume drives both growth and strength, and frequency compatible with negligible effects on growth 6. In the one trial in women, 50 untrained women aged about 23 did a four-day upper/lower split or a two-day full-body plan with the same weekly sets: leg press rose 28.3% against 29.2%, muscle mass 1.7% against 1.9%, jump height 12.5% in both. Five of the six dropouts came from the four-day arm, four citing lack of time 7. Across 24 studies in healthy women, no programme variable significantly moderated growth 8. The two-lower, two-upper layout is therefore defensible but not privileged; its cost is adherence, which is why session length matters more than session count.
Calendar weeks matter less than apps imply. Among 89,573 adults with accelerometers, 150 minutes of activity concentrated into one or two days carried the same cardiovascular associations as the same minutes spread evenly 9. A week off mid-programme in 39 trained adults (10 women) left muscle growth identical to continuous training 10. In adults aged 20–35, one-ninth of a prior volume, once a week, preserved strength and fibre size for 32 weeks; adults aged 60–75 lost size on the same doses 11.
How EVLVE is built
A rotation is four completed strength sessions: Lower A (squat and thrust), Upper A (vertical pull and horizontal press), Lower B (hinge), Upper B (overhead press and row); a block is 16 work sessions plus up to four lighter ones. Every tier-1 region (glutes, outer glutes, lats, side shoulders) is trained in two sessions per rotation, and the hamstrings always get one hinge and one curl. Because the cohort evidence is per calendar week, a member who takes 14 days to finish a rotation is on half the weekly dose, enough for maintenance under 45 but not for progression; cardio is counted by calendar week toward 150 minutes, vigorous minutes counting double. Where flexibility ends and under-dosing begins is a judgement call; the trials do not settle it.
The warm-up evidence for lifting is thinner than most apps admit, and almost all of it is in men. A review of 11 studies found no agreement that a general plus specific warm-up beats a specific one alone; performance tended to be better when the warm-up loads were closer to the working load 12. In a crossover of 29 resistance-trained men and women, zero, one or two specific warm-up sets before four sets to failure at a 10-rep-maximum load made negligible difference to repetitions or volume 13. For upper-body work, high-load dynamic warm-ups improved strength and power, short static stretches had no effect, and there are no trials at all linking a warm-up to injury, which the reviewers call a striking knowledge gap 14. "Glute activation" drills, which the old EVLVE warm-up opened with, have not survived testing: in 17 professional rugby players, peak gluteus maximus activity in the loaded pulls that followed an activation warm-up was significantly lower, not higher 15.
Exercise order matters for strength and not for size. Across 11 studies, strength gains favoured whichever exercise was done first (effect size 0.32 for multi-joint lifts placed first), while growth was unaffected by order (0.03) 16. The other constraint comes from concurrent-training research: explosive strength is the one quality cardio in the same session measurably blunts (−0.31 when within 20 minutes) 17, and the deficit was present in women specifically (−0.52) 18.
How EVLVE is built
The skeleton is fixed: prime (impact, then one balance task, then one or two prep drills for the first lift; 7 minutes at the 60-minute setting), ramp-up sets inside the first lift's time, two main lifts, one multi-joint secondary lift then single-joint work with priority regions first, two short unloaded core sets, and a 3.5-minute cool-down. Impact goes first because landing quality falls with fatigue (inference; no trial tests placement), and the lift that should gain the most strength follows. Ramp-up sets scale with her working weight, rounded down to plates she owns: three before a heavy first lift (40% × 5, 60% × 4–5, 80% × 2–3), one skippable set at 60% × 5 before an 8–12 lift, one feeler before the second main lift, nothing after the mains. The percentages are convention; the evidence supports only the direction. Prep drills are chosen by the first lift's prep targets (ankle bend and hip flexion before a squat, hinge before a deadlift, shoulder blade before a press) from a library of 49 drills, 45 seconds each. The activation drill and six timed mobility drills are gone. The optional interval track never precedes a lower session or the impact block. No claim is made that any warm-up prevents injury.
The longevity case for balance rests on observational data in people older than EVLVE's members. Among 1,702 adults aged 51–75, 20.4% could not hold a 10-second one-leg stance, and failing was associated with a hazard ratio for all-cause death of 1.84 over a median seven years 19. The sitting-rising test, which depends on strength and flexibility as well as balance, showed a graded association with mortality in 4,282 adults aged 46–75 (hazard ratio 3.84 for the lowest scores) 20. None of this is an intervention study, and none starts below age 46.
What the trials show is that balance is trainable and specific. In 25 randomised trials in adults aged 16–40, balance training improved steady-state balance with a standardised mean difference of 0.73; the best-performing dose was 11–12 weeks, three sessions a week, 11–15 minutes a session 21. The gains are task-specific: across six trials the trained task improved (SMD 0.79) and untrained tasks did not 22. Strength does not substitute: maximal lower-limb strength and dynamic balance correlate at 0.09 in young adults 23. For injury, strength training itself is the most protective exposure in athletes (relative risk 0.315), ahead of balance training (0.550), while stretching does nothing (0.963) 24.
How EVLVE is built
One balance task per session, in the prime block, 2 × 30 seconds per side at the 60-minute setting (one set at 45; a second, moving task at 75). The ladder has seven stages: tandem stance; single-leg stance; single-leg with head turns; single-leg eyes closed (2 × 20 s); single-leg reach (2 × 5 rounds); eyes closed on a folded towel; and hop-and-stick, with a no-impact towel reach as the alternative. In session one she tries stage 2 then stage 4 and starts at the first she cannot hold cleanly (amendment E3). The task never rotates for variety; it changes only when she holds full time on both sides without a touch-down in two consecutive sessions and accepts the next stage; repeated touch-downs offer the stage below. Times are fixed per stage because no validated intensity measure for balance exists. Single-leg lifts are not credited as balance work. The dose is stated honestly: about eight minutes of task time per rotation, against the 33–45 minutes a week that performed best in trials, so copy says "practice", not "protection". The entry rule assumes most women this age pass the eyes-open stand quickly; the norms behind that could not be verified in the audit's evidence pass, so it is labelled an assumption.
Full-range lifting is itself a mobility stimulus. Across 55 studies in 2,756 people with a mean age of 24, resistance training improved range of motion against control (effect size 0.73), was no different from stretching (0.08), and adding stretching to lifting added nothing (0.00); bodyweight-only work did not reach significance 25. An earlier pooling of 11 trials, seven in women or mostly women, found the same 26. In a six-week trial in young adults, two minutes a week of static hamstring stretching and two minutes a week of full-range Romanian deadlifts both beat control on a reach test, with no significant difference between them 27. Where a specific range is limited, static stretching works at small doses: across 189 studies, long-term gains plateaued at about 10 minutes a week per muscle group, and frequency, intensity, sex and age were not moderators 28. A 2025 Delphi consensus names full-range strength training as an alternative route to flexibility and does not recommend stretching for injury prevention, recovery or posture 29.
The cool-down cannot be justified with recovery language. An active cool-down is largely ineffective for next-day performance, soreness, stiffness or psychological recovery 30. Post-exercise stretching changed next-day soreness by −1.04 points on a 100-point scale in a Cochrane review of 12 trials 31, and a pooling of 11 trials found no effect on strength recovery or soreness at 24, 48 or 72 hours 32. Slow breathing has small, low-quality evidence for reduced self-reported stress in general adults (g −0.35), none of it specific to the end of a workout 33.
How EVLVE is built
A stretch earns its place only as a dose for a range the day's lifts do not load at length. Lower days default to hip flexors, calves, and one of inner thigh (added by amendment B6) or hamstrings; upper days to chest, lats, and one of upper-back rotation or neck. Three stretches of 60 seconds at the 60-minute setting, two at 45, three longer at 75, from a library of 60 stretches and 22 active-range drills. She can pin one stretch to every session, which is how a region reaches the 10-minute weekly plateau; the default gives a region 2–3 minutes per rotation, enough to produce change, and the copy says so. With the hypermobility flag, every passive end-range stretch becomes an active-range drill; this is expert opinion, not a tested rule. One minute of slow breathing is available, off by default, with no recovery claim. The old cool-down's six stretches (6.5–9 minutes) are gone, as is any copy crediting stretching with recovery or injury prevention.
The evidence on long-horizon design is permissive rather than prescriptive. Planned exercise variation did not harm growth and raised intrinsic motivation (effect size 1.28) while a fixed programme lowered it (−0.75), in 21 trained men; constant random rotation showed hints of blunted gains 34. A scheduled week off neither helped nor harmed growth 10. Supervised programmes produced larger lower-body strength gains in women than unsupervised ones 8, the gap an app has to fill with load prescription and effort calibration. Across 66 studies, autonomous motivation and a growing sense of competence were the consistent predictors of continued exercise; appearance motives were mixed 35.
How EVLVE is built
There are three levels. A level is the highest movement-skill tier the resolver may pick by default (one: goblet squat, leg press, trap-bar deadlift; two: barbell squat and deadlift, overhead press; three: unsupported single-leg Romanian deadlift, weighted pull-up). It is not an equipment gate and does not change set counts, rest, or the bone and balance ladders. Starting level comes from stated experience, and she can always choose lower. When 60% or more of her 12 main and secondary slots have progressed over a block, the next level is offered slot by slot; the 60% line is convention.
Three phases. Foundation is sessions 1–16, once: the first rotation runs two sets per slot (amendment B4), barbell and trap-bar mains sit at 8–12 for the squat and 5–8 for the hinge with three reps in reserve for sessions 1–8, and a calibration set runs only on dumbbell, band, machine and bodyweight lifts, never on a bar (amendment A1). Build is 16 work sessions repeating for life; for members with a bar the squat moves to 5–8 and the hinge to 4–6 as a standing heavy zone, because a block gives each lift only four exposures. The lighter rotation is four sessions at half sets, same weights, with impact and balance unchanged because bone work is a standing dose; it is an offer, pre-selected only when the 16 work sessions took 42 days or fewer. Accessories marked "alternate" switch between two candidates at each block start; she can pin any exercise. Nothing is a "week" because the evidence attaches to volume and effort, not the calendar. The 42-day line is a judgement call; the trials do not settle it.
The time-efficiency literature is clear about what to cut first. A review for time-limited trainees recommends a minimum of four weekly sets per muscle group at 6–15 reps, weekly volume over frequency, a warm-up restricted to exercise-specific sets, and stretching only if flexibility is itself a goal; it notes traditional programmes with warm-up and stretching "often exceed an hour" 36. Because neither order 16 nor frequency at equal volume 5 changes growth, what a shorter session can afford to lose is sets on non-priority regions, not the exposures of the priority lifts.
How EVLVE is built
The program is 47 slots across four sessions, and the budget setting is her volume dial: 58 working sets per rotation at 45 minutes, 74 at 60, 97 at 75. The estimate she sees is arithmetic run on the exercises she will actually do (a two-leg set at 4–8 reps counts 40 seconds, a single-leg set 90, set-up at a barbell main lift 120, a ramp-up set 55, rest from the rest table); the set times are stated inferences, to be corrected from logged durations. In the full-gym Build reference the four sessions estimate at 55.8, 51.0, 55.2 and 49.2 minutes, with a design target of 57 so real sessions have reserve.
When 60 becomes 45, work leaves in a fixed order: the prime shortens to four minutes and the cool-down to two stretches, but impact and balance are never removed; then one core slot; then the triceps and biceps slots; then one set each from the curl and rear-shoulder slots; and only last, one set each from the tier-1 glute, outer-glute and side-shoulder slots, which still sit at least 1.25 times their floor. Going to 75 reverses the order and adds calves, a two-handed carry (amendment B9), a second moving balance task and a third stretch. No main lift ever loses rest to fit a budget; only accessory rests flex, down to 45 seconds, as the last resort. The shorter session is a design, not a trim: it keeps every movement pattern, every ramp-up set and the pillars lifting cannot supply, and gives back sets where the volume curve is flattest.
Chapter 2
This chapter covers loading bone in healthy women aged 28 to 45: the rules bone cells follow, the jumping trials, the heavy-lifting trials that moved the spine, and which equipment delivers either stimulus. It explains the seven-rung impact ladder that opens every EVLVE session, the heavy squat and deadlift for bar owners, and the pain and leaking check that gates progression. The conclusion: brief, hard, rest-inserted impact on most days protects the hip, heavy axial lifting is the only route to the spine, and the honest promise at this age is to hold what she has, not to build.
Most of what is known about the rules of bone loading comes from animal models. Loading must be dynamic: in the avian ulna, static loading produced 13% bone loss, about the same as disuse, while dynamic loading for 100 seconds a day over eight weeks produced a 24% gain 1. Magnitude matters, with the response linearly related to peak strain and no "lazy zone" 1. Rate matters more: in sheep, strain rate explained more of the variation in new bone than peak magnitude did, and in the rat ulna a five-fold increase in strain rate increased bone formation 1. A fast, hard landing is a better signal than a slow squeeze.
The response saturates quickly: 36 to 1,800 cycles per episode produced no additional response beyond 36 1, and bone cells desensitise within perhaps 50 to 100 repetitions 2. Rest restores sensitivity: a 10-second rest between cycles increased bone formation by 72% despite 100-fold fewer cycles 1, and four bouts separated by three hours produced gains of 11.7% against 6.9% for the same cycles in one bout 2. The ACSM position stand calls the human relevance of these interactions "not known" 3.
Bone responds to novelty and only at the site strained 1. In premenopausal women, high-impact leg exercise raised femoral neck density but did nothing for the distal radius 4, and adding upper-body resistance to jumps changed spine density (+1.4% against −0.9%) but not hip density 5. In women aged 21 to 40 loading the forearm, a combined measure of magnitude, rate and bout count was the best predictor of radius gain but explained only about 12 to 15% of the variance 6.
How EVLVE is built
Contacts are few and fixed by rung, about 50 when each is light and about 20 when each is hard, and are never progressed; what progresses is how hard each contact is. Light contacts run in sets of 10 with 30 seconds between sets; hard contacts are single efforts about 8 seconds apart, an inference from the animal rest data, since no rest interval for bone has been studied in women. Because sensitivity returns within hours in animal data, the day card on any non-session day offers the same block at her current rung as an optional extra. Placing impact first in every session is an inference: landing force and quality fall with fatigue, and no trial has tested placement.
The jumping trials agree on two things: the hip responds, the spine does not. A meta-analysis of six randomised trials of brief high-impact exercise in premenopausal women found a standardised effect of 0.64 (95% CI 0.38 to 0.90) at the femoral neck but 0.04 (−0.23 to 0.31) at the lumbar spine 7. A second found the same pattern 8, and a third found that only impact combined with high-magnitude resistance moved the spine as well as the hip 9. Pooling all exercise types across seven trials and 521 women, the femoral neck benefit was about 1.1%, made up of +0.4% in exercisers and −0.7% in controls 10.
The individual trials show the dose. Fifty vertical jumps six days a week, at 8.5 cm and about three times body weight, raised femoral density 2.8% in five months in premenopausal women; the same programme did nothing in postmenopausal women 11. In women with a mean age of 34, 50 multidirectional one-leg hops at 2.5 to 2.8 times body weight produced a femoral neck change of 0.0% at two days a week, +0.9% at four (interval crossing zero) and +1.8% at seven; frequency, not the hop, was what mattered 12. In women aged 20, 10 maximal jumps three days a week for six months raised both femoral neck (about +2.6%) and lumbar spine (about +2.4%), the one jumping trial to move the spine, in women ten years younger than the youngest EVLVE member 13.
The caveat is the most recent analysis. A 2026 meta-analysis of 22 randomised trials in adults aged 18 to 45 found no significant overall effect of exercise on density at the spine (SMD 0.15, −0.09 to 0.38) or femoral neck (0.06, −0.11 to 0.22). Under 30, both sites improved; over 30, neither did. High impact alone showed no consistent effect, and only two of the 22 studies were at low risk of bias 14. This is the strongest reason not to promise a woman over 30 that jumping will raise her density. What the older trials show more consistently is prevented loss 10.
How EVLVE is built
From rung 2 upward, four sessions in a week with about 50 light or 20 hard contacts each matches the typical trial protocol of about 50 jumps a day on four days a week 15, at about 3 minutes a session. A member training two or three days a week is below the frequency at which moderate hopping worked, which is why the off-day option exists and why higher rungs matter. The stated expectation is hip bone held, possibly 1 to 2% ahead of doing nothing over 6 to 12 months. The block is a standing dose, unchanged in a lighter rotation, because the structural benefits of 18 months of impact training were no longer significant 3.5 years after it stopped 16 and a premenopausal trial found that detraining reversed the gains 17.
The evidence for loading the spine leans on older and younger populations. The clearest trial is in postmenopausal women with low bone mass: deadlift, back squat and overhead press for five sets of five at more than 80 to 85% of one-repetition maximum, plus drop landings, twice a week for 30 minutes over eight months, raised lumbar spine density 2.9% against −1.2% in controls and femoral neck 0.3% against −1.9% 18. The control group, lifting 10 to 15 repetitions below 60% of maximum with hand weights of at most 3 kg, lost bone at both sites 18. In another postmenopausal trial, maximal-strength squats with a fast start to each lift raised spine and femoral neck bone mineral content by 2.9% and 4.9% in 12 weeks, in groups of eight 19.
In women aged 18 to 30 with below-average bone mass, 10 months of twice-weekly training showed impact beating resistance at the distal radius and tibia, and resistance beating impact at the proximal femur (femoral neck cortical density +3.7% against −4.1%) 20. No premenopausal trial compares heavy with moderate loading for bone. The 80 to 85% threshold comes from postmenopausal trials and from the Australian position statement, which prescribes about 80 to 85% of maximum on squats, deadlifts, split squats, back extensions and rows at least twice weekly, plus impact above four times body weight 21.
How EVLVE is built
For a member whose first lower-body lift resolves to a barbell or trap bar, the squat is trained at 5 to 8 repetitions and the deadlift at 4 to 6 in every Build block, a zone that reliably reaches about 80 to 88% of maximum. It is a standing zone, not an alternating block, because strength and bone both fade when the stimulus stops. The hip thrust and Romanian deadlift stay at 8 to 12 as the growth lifts. Foundation (sessions 1 to 16) keeps bar lifts lighter while technique settles, as the postmenopausal trial's familiarisation month did 18. The exact rep zones are a judgement call inside the trial range. Each upper session also contains one hand-supported element, since the wrist responds only to load through the hand 6; no bone claim is made for it, because no trial of push-ups on wrist density exists.
Every exercise in the library carries a bone credit of impact, heavy axial, hand load or none, and most of the library is none, because only two routes have trial support: impact, and axial lifting at about 80% of maximum or more. Light loads are not a bone stimulus. The postmenopausal control arm lifting at most 3 kg lost bone at both sites 18, and six months of low-load, very-high-repetition barbell classes in women aged 56 to 75 did nothing at the hip 22. Modelling in postmenopausal women found that isolated hip-machine exercises at 40 to 80% of maximum strained the femoral neck no more than walking: machine abduction at 80% produced 1,480 microstrain, walking at 4 km/h 1,474, and single-leg hopping 9,517 to 10,373 23. The abduction machine therefore earns no bone credit; nor does the hip thrust, which appears in no bone trial found and does not load the spine axially, nor the leg press, which leaves the spine unloaded. Bands load slowly and peak at end range, the opposite of what the rate rule asks for. This conclusion rests on postmenopausal trials and modelling; no bone trial of dumbbell-only or band-only training in premenopausal women exists.
Impact, by contrast, needs no equipment. In 21 premenopausal women with a mean age of 43, single-leg hop landings reached 4.02 to 4.93 times body weight at loading rates of 237 to 319 body weights per second, above the thresholds of 3 body weights and 43 body weights per second 24. In the same women, star and stride jumps with a rebound reached 3.90 to 5.38 times body weight, the second landing 12 to 48% higher than the first 25. In early postmenopausal women measured at the sacrum, a countermovement jump produced 10.7 g, a drop from a 20 cm box 9.6 g, heel drops 7.3 g, and a single-leg stamp only 3.5 g, judged insufficient 26. Two things lower the stimulus: a 10 cm soft mat cut single-leg landing peak force by 19% and the initial impact peak by 34% in ten young women 27, and a weighted vest of 10% of body mass reduced landing height and peak force 28.
How EVLVE is built
The bone tag attaches only to impact rows, to the barbell and trap-bar squat and deadlift, and to hand-supported upper-body work. A member with no bar sees "Your jumps are your bone work. The weights you have build muscle; bone needs impact or very heavy lifting", with no purchase suggested. Impact is done on a firm floor or thin mat, never holding weights, and never after cardio. Rungs 6 and 7 need a step of 8 inches or more; without one she gets rung 5 with the direction changed each block.
The ladder follows the staged, symptom-gated progression used clinically for athletes returning to impact: heel drops, then jumps, then hops, then landings from a small step, advancing only after two symptom-free sessions 21. EVLVE's rungs, with contacts at the 60-minute setting (30 light or 12 hard at 45 minutes), are: 1, heel drops, 50 (7 to 8.6 g in the postmenopausal lab sample, with the lowest quadriceps demand measured 26); 2, low two-foot hops, 50 (about 3 body weights, the dose that raised femoral density 2.8% 11); 3, two-foot jumps as high as is comfortable, each landing held, 20 13; 4, jumps in new directions including a star jump with a rebound, 20 (3.9 to 5.4 body weights 25); 5, single-leg hops on the spot, forward and sideways, 20 per leg (4.0 to 4.9 body weights 24); 6, step-off landings on two feet from 8 inches, 20 (9.6 g from 20 cm 26, the step height the Finnish trial reached at six months 29); and 7, step off and rebound, a single-leg landing from 6 to 8 inches, or a 12-inch step-off, 20, which is convention after the clinical ladder, supported by drop jumps producing higher peak force than plain landings in 15 women 30.
Everyone starts at rung 1 from session one, at every level; gating the height of the rung, rather than hiding impact behind two program levels as the old program did, is the proportionate control. Every step up is offered, never automatic, one rung at a time: rung 1 to 2 after the first rotation of four sessions and at least 14 days, every later rung at a block boundary and at least 42 days after the last change, matching the roughly two-monthly progression of the Finnish trial 29.
The check is one tap after the block: "Any pain or leaking with those? No, A little, Yes". "Yes" once, or "A little" twice running, drops the next session one rung without asking, and she can step down herself at any time. By a later amendment, pain reported on any rung drops her to the low-impact track for that session, suspends step-up offers for at least two rotations, and surfaces the pain rule and the refer-out list; leaking does the same and shows the pelvic-floor signpost.The pain rule for impact is "These should not add to any pain you came in with." The refer-out list includes leaking that limits training, with a pelvic health physiotherapist named as the route, following the UK consensus that urinary incontinence may need to be addressed before impact training 31. No trial of pelvic-floor outcomes with osteogenic jumping was found, and the two-symptom-free-sessions gate is convention. The low-impact track, for the low-impact preference and as the starting point for the pelvic floor, knee and outer-hip flags, keeps heel drops at 50 contacts with a return to rung 2 by offer. If she swaps heel drops away, the fill is fast step-ups or brisk marching, marked as no bone credit: a firm stamp measured only 3.5 g 26, and stair climbing raises hip loading above walking without reaching an impact threshold 32.
Hip bone mass peaks early and spine bone mass late: in Canadian longitudinal data, women's total hip peak occurred at ages 16 to 19 and the lumbar spine peak at 33 to 40 33. A woman aged 35 to 40 loses about 0.5% of density a year on average 29. Against that, the measured benefit of exercise in premenopausal trials is in the low single digits and is substantially prevented loss: pooled controls lost 0.7% at the femoral neck while exercisers gained 0.4% 10, and in the twice-daily jumping trial the jump groups held steady over 16 weeks while controls lost density 34. Formation markers rise within four months, before density change is measurable 14, so 6 to 12 months is the shortest honest horizon.
Safety has been good but thinly reported: of seven premenopausal randomised trials, only one reported on adverse events, and it reported none that were serious 10; an eight-month maximal-jump programme had three minor musculoskeletal complaints 35. The authors of the postmenopausal heavy-lifting trial do not recommend unsupervised training of that kind for people with low bone mass 18, one reason the app wraps heavy work in a familiarisation phase, ramp-up sets and conservative load steps.
How EVLVE is built
The app may say "Protect your bones", "Bone responds to impact, small jumps, done often", and "At this age the aim is to keep what you have and carry the habit forward". It may not say "Build bone" or "Increase bone density", may not promise anything for the spine or wrist from jumping, and may not attach a bone claim to any band or light-dumbbell exercise, the hip thrust, the leg press, the abduction machine, walking, cycling or swimming. Heel drops are "the best low-impact option we know of", never "as good as jumping". Making bone work a standing habit rather than a phase rests on the arithmetic that each 1% preserved is roughly two years of loss at 0.5% a year; that framing is an inference, not a trial result.
Chapter 3
This chapter covers the second of EVLVE's two goals: an hourglass shape, meaning fuller glutes and hips, a waist that reads narrower, and visible muscle through the shoulders and back. It reviews what the trials say about how muscle grows, which lower-body patterns favour the glutes, what the waist can and cannot be trained to do, and how a woman's skeleton and the equipment she owns change the prescription. The conclusion: shape is built from hard sets through a full range, counted by region and kept inside a budget, around a waist that is trained for strength and never loaded.
The dose that grows a muscle is the number of hard sets it does each week. In a meta-regression of 67 studies and 2,058 participants, 79% of them men, growth rose with weekly sets with diminishing returns, and the best-fitting model counted an indirect set, such as a row for the biceps, at half a direct set 1. An earlier meta-analysis found that fewer than five weekly sets still produced about 5.4% growth, just more slowly 2. In trained young men, 12–20 weekly sets per muscle was the recommended standard 3. A preprint found no further benefit beyond about 11 counted sets per muscle per session 4.
How heavy the weight is matters less than people expect. A network meta-analysis of 28 studies in which every set went to failure found no difference in growth between loads above 15RM, 9–15RM and 8RM or fewer; heavier loads did build more one-repetition strength 5. The floor is roughly 30% of maximum, below which growth falls to about half 6. Effort matters: growth improves as sets end closer to failure, while strength gains are similar across a wide range of reps in reserve 7. Resting more than 60 seconds gave a small growth benefit, with nothing further beyond about 90 seconds 8.
Women respond the same way. Across 29 studies with 1,278 male and 1,537 female data points, absolute growth slightly favoured men but relative growth did not differ (0.69%, with an interval spanning zero); the authors conclude that females have a similar potential for hypertrophy 9. That allows male-dominated dose data to inform a women's program, though it remains an extrapolation for the glutes, where no dose-response study exists.
How EVLVE is built
The 60-minute rotation holds 74 working sets across four sessions. Main lifts are prescribed with two reps in reserve ("Leave 2"), accessories with one, and band or light-dumbbell work finishes its last set "to the last good rep". Rest is looked up by the kind of lift: 150 seconds on two-leg lower-body lifts at eight reps or fewer, 120 at 8–12, 90 on upper-body compounds, 60 on isolation work. No rest number is tested in women, and EVLVE never shortens rest because the member is a woman.
Direct evidence on glute growth is thin: a 2025 meta-analysis found 12 trials and about 318 people (144 women), with a pooled standardised effect of 0.71 for gluteus maximus size 10. The best single trial is Plotkin and colleagues: 34 untrained adults, 68% women, trained either the hip thrust or the squat for nine weeks at 8–12 reps to failure. On MRI the two lifts grew the glutes equally in every region, while squats grew the quadriceps by 3.6 cm² more and the adductors by 2.5 cm² more. Surface EMG was higher in the hip thrust at every gluteal site, yet EMG did not predict who grew most 11. "Feeling it" is not a measure of growth.
Range matters. In untrained men, full squats grew the gluteus maximus by about 6.7% in ten weeks against 2.2% for half squats 10. A 2026 narrative review of 13 range-of-motion trials concludes that range itself is not the key factor; what matters is meaningful load while the muscle is long, which is necessary but not sufficient 12. No published trial compares a long-length hip extension such as the Romanian deadlift against the hip thrust for glute size, and Plotkin's result cuts against a strong version of the "stretch" claim.
Adding a thrust pattern to a squat-and-hinge base helps. In 33 untrained young women, ten weeks of leg press plus stiff-leg deadlift grew gluteus maximus thickness by 6.0%; adding a barbell hip thrust raised that to 9.3% (P = 0.016). Volume was not equated, so this shows that more glute sets grow more glute, not that the hip thrust is special 13. Knee-dominant machine work is not glute-neutral: 15 weeks of leg press, extension and curl grew gluteus maximus volume by 12.9–16.6% in men 10. Hamstrings are a separate job: squats and hip thrusts barely grew them 11, and a nine-week trial in 36 untrained people found the stiff-leg deadlift grew the hamstrings by 7.0% with a semimembranosus bias while a knee-flexion exercise grew the semitendinosus, so the authors suggest both 14.
How EVLVE is built
Each lower session carries the three patterns the trials collectively cover. Lower A: a squat (3 sets, 5–8 with a bar in the Build phase, 8–12 otherwise), a hip thrust (4 sets, 8–12) and a single-leg lunge pattern (2 sets, 8–12 per side). Lower B: a heavy hinge (3 sets, 4–6 with a bar or trap bar), a Romanian deadlift (3 sets, 8–12), the lunge pattern (3 sets) and a hip-extension accessory (2 sets, 12–20). Hamstrings always get one hinge and one curl per rotation; no budget or substitution may remove either. Rep caps are 15 on two-leg lifts and 20 on single-leg lifts, inferred from leg-press data showing about 19 reps at 70% of maximum 15. Tempo is convention.
The upper-outer "shelf" is the gluteus medius and minimus, and evidence for growing them in healthy women is nearly absent. In Plotkin's trial the combined medius and minimus area did not differ between squat and hip thrust training 11. The only volumetric trial is in 27 adults with hip osteoarthritis, where a targeted program grew the minimus but not clearly the medius 16. The rest is EMG. In 16 untrained women, band abduction produced higher medius activity than a machine at matched effort 17, and moving a band from the knees to the ankles to the feet raised stance-leg medius activity by 43% and a further 19% in a mixed-sex sample 18.
What cannot be trained is bone. Hip dips come from the shape of the pelvis and femur; a hospital health article states they cannot be got rid of, though added muscle may fill the divots a bit 19.
How EVLVE is built
Because two-leg squats and thrusts are not a sufficient abductor stimulus, both lower sessions carry a direct abduction slot of 4 sets at 12–20 reps with one rep in reserve, giving 8 direct outer-glute sets per rotation (6 at 45 minutes, 10 at 75); single-leg lunge patterns credit the outer glutes at half, so the counted figure is about 10.5, with a floor of 6 and a ceiling of 14. Band walks are not counted sets. There are no dose data for this muscle; keeping direct work near 8 is a judgement call. A band is progressed by moving it toward the feet before buying a heavier one (placement EMG, otherwise inference). The copy says "Hip width is bone. Training adds muscle above and behind it," and never promises to fix dips or widen hips.
Training the abdominal muscles does not remove the fat over them. In 24 sedentary adults, six weeks of abdominal exercise five days a week changed neither abdominal fat nor waist circumference, though curl-up endurance rose 20. A meta-analysis of 13 studies and 1,158 participants of both sexes found the effect of local muscle training on local fat to be a Hedges' g of −0.03 (95% CI −0.10 to 0.05), which is nothing 21.
The opposite worry, that loaded or banded core work or heavy compound lifting thickens the waist, is equally untested: no trial with a waist-girth or oblique-size outcome was found in either direction. What does change the look is the ratio. Across three studies, female figures with a waist-to-hip ratio of about 0.7 were rated most attractive, and the ratio in published samples held at 0.68–0.72 across decades 22. Adding hip circumference (glutes) and shoulder and back width (side deltoids, lats) lowers those ratios with no change to the waist; that is the only mechanism the evidence supports. Protein supplementation across 49 trials added 0.30 kg of fat-free mass to training, with no further gain above about 1.6 g per kg per day 23; fat loss needs a deficit and muscle gain needs at least maintenance, so the two halves of this goal pull against each other.
How EVLVE is built
Core is trained for strength and control and never loaded: no band, no added weight, by locked rule. Each session ends with two short sets (anti-extension, anti-rotation or anti-lateral) as holds of 20–45 seconds or 6–12 reps, with a floor of 4 and a ceiling of 10 sets per rotation; progression is by time, then a harder lever. Both Pallof presses and the suitcase carry left the program under this rule. The copy says "Core work here is for strength and control" and "The waist reads narrower as glutes, back and shoulders grow around it." The unloaded rule is a precaution, stated as one; the trials do not settle it. On food the app gives guardrails only and never computes a deficit.
"Toned" is muscle you can see: growth plus enough leanness to show it. Relative growth is the same in women as men for the upper body too (0.60%, spanning zero) 9, and the rate is slow enough that accidental bulk is not a realistic risk. In trained lifters, a third of them women, 10 hard weekly sets of lateral raises for eight weeks grew the side deltoid by 3.3–4.6% with a cable and 3.4–3.9% with a dumbbell, with no difference between implements 24. Non-priority muscles can be held cheaply: after 16 weeks of training, cutting volume to a third or a ninth preserved muscle size for 32 weeks in 20–35-year-olds 25.
Which upper muscles matter for the frame is a design judgement; no study tests it. Posture claims should be softer still: exercise improved posture angles only in people with upper-crossed syndrome 26, and nothing shows training changing resting posture in healthy women of this age.
Grip often ends the pull-up before the lats do. Women's grip strength averages about 31 kg at ages 30–39 against 47 kg in men 27. In 11 trained men a 2-inch bar cut deadlift maximum by 28% and a 3-inch bar by 55%, with smaller hands losing more 28. In 10 trained women, lifting straps allowed more reps at 80% of deadlift maximum and preserved grip strength 29.
How EVLVE is built
Each upper session carries a vertical pull (3 sets; pull-up 5–8, pulldown 8–12, band 12–25), a press, a row (3 sets, 8–12), a lateral raise (4 sets, 10–20) and a rear fly (2–3 sets, 12–20). Per rotation that gives 12 counted lat sets and about 9.5 side-deltoid sets, both tier-1 regions; chest, front deltoids and arms sit at 2–3 direct sets, held rather than grown, on the one-third maintenance finding. Light isolation lifts use a 10–20 rep range because below 25 lb every step on a standard rack is a 20–67% jump; a 10–20 range absorbs a 25% jump (inference from the reps–load curve). Pull-up copy never gives a date: "A first pull-up usually takes months. Your rows are doing the building meanwhile." Under a grip consideration, rows carry the cue "shorter sets, straps if you have them."
The anatomical differences that matter are smaller than the folklore. In 2,072 Norwegian 19-year-olds, women had 53° of hip internal rotation against 39° in men, and shallow sockets were about twice as common 30, so the typical woman has more bony room for deep hip flexion than the typical man; depth limits "to protect the hips" are not supported for women as a group. The static Q-angle is a few degrees larger in women on average, but the ranges overlap almost completely (1–23° in men, 2–21° in women), the measurement is unreliable and it does not predict injury 31. The one kinematic study of a mini-band around the knees during loaded squats found it increased knee valgus 32. The claim that women have relatively longer femurs is not supported: in the 1988 US Army survey, sitting height was 52.3% of stature in women and 52.1% in men 33.
Equipment fit is where differences bite. In 103 untrained college women the bench-press maximum was 28.7 ± 6.7 kg, so a 20 kg bar is about 70% of the average maximum and more than the weakest lifter could press 34. For growth the implement does not matter: free weights and machines gave the same hypertrophy across 13 studies at every training level 35. For keeping a set short it matters. The leg press allows about 19 reps at 70% of maximum and 13 at 80% 15, and in 13 trained women of 58 kg the ten-rep hip thrust averaged 87 kg 36, so a pair of 20–25 lb dumbbells runs out quickly on two-leg hip work. The fix is single-leg work: a meta-analysis found no hypertrophy difference between unilateral and bilateral training, on two small trials in women 37, and in 13 sedentary women a bodyweight squat ladder from two legs to one grew the glutes as much as barbell squats over six weeks 38.
How EVLVE is built
Stance and depth are never set by sex; the copy may not mention knee angle or thigh-bone length, and band-around-the-knee squats are not defaulted onto loaded work. The 20 kg bar is a milestone for pressing, not an entry point; squats, hinges and thrusts can start with the bar. A load-adequacy test runs on every slot: a weight is `light` when her heaviest load is under half the entry load, or when every set hits the rep cap in two consecutive exposures (the half line is inference; logged reps correct it within two sessions). A `light` two-leg lift whose family has no harder rung may switch families: goblet squat to the split-squat family, two-leg hip thrust to single-leg hip thrust, two-leg Romanian deadlift to the single-leg hinge. Each family is a ladder of real exercises (the hip thrust family runs bridge, single-leg bridge, shoulders raised, B-stance, single-leg, paused, foot raised), and a harder rung is offered after two consecutive exposures at the rung's top.
The fractional-set model from the dose meta-regression is the accounting rule for the whole program 1: a set counts 1 for each region the exercise works fully and ½ for each it assists. A deep squat credits glutes and quads fully, because squats and hip thrusts grew the glutes equally; the Romanian deadlift credits glutes at half, an inference, since no trial has measured it alone. The glute line sums to 18.5 counted sets at 60 minutes, 14.5 at 45 and 21 at 75, with at most 9.5 in a session. The floor is 10 and the ceiling 22; the ranges of 6–12 for a novice, 12–18 intermediate and 16–22 advanced are inferred from mostly-male dose data plus the two glute trials, where 6–18 weekly sets grew glutes in untrained people 11 13. Set counts do not change with level; her dial is the session length.
The old program held about 26 counted glute sets, where one more detectable step of growth was costing about 11 more sets on the dose curve 1; the rebuild gave back about 7 and bought 30–60 seconds more rest on every heavy leg set, four impact blocks and four balance tasks.
How EVLVE is built
The budget is enforced at run time. After every resolution, rotation or swap, the engine recomputes counted sets per region from the exercises actually resolved (a hip thrust swapped in for a squat no longer credits quads); if a tier-1 region falls under its floor and a loadable option exists, sets move to it from the lowest-tier slot above its own floor, and if none exists the app states the gap once. Trimming to 45 minutes removes every tier-3 and tier-4 direct slot before any tier-1 set leaves. Which lifts are "direct" for the glutes is itself a judgement; the trials settle the squat and hip thrust, and the rest is inference.
Chapter 4
This chapter covers what the research says about how a woman recovers between sets and between sessions, how hard each set should be, whether the menstrual cycle or the pill should change a program, how much training is enough, what cardio belongs outside the sessions, and the food guardrails an app can state without diagnosing anyone. The conclusion is plain: the evidence gives no reason to program women differently from men on rest, frequency, volume, effort or cycle phase, so EVLVE uses the general dose and spends its design effort where women's lower absolute loads make a real difference.
The most repeatable sex difference in the lifting literature is that women lose fewer reps across a run of submaximal sets. In 22 men and women doing three sets of up to ten bench press reps at 75% of maximum, women completed 26.9 of a possible 30 reps with one minute of rest against 21.1 for men; with three minutes, 29.7 against 25.8 1. In 21 trained men and 21 trained women matched for relative bench strength, doing sets of five at 75% with 90 seconds of rest until they failed, women completed 58.3 total reps against 29.6; a single set to failure did not differ (9.0 versus 8.6 reps) 2. A scoping review of 34 studies in trained people still found "minor-to-no sex differences in neuromuscular fatigue" overall, the male disadvantage showing mainly in free-weight lifts at moderate loads with short rests 3. Part of the mechanism is fibre type and blood flow: women are "usually less fatigable" in sustained isometric work, though the gap shrinks at higher intensities 4; in intermittent isometric knee extension ten women lasted 62.4 minutes above their critical intensity against 30.4 for ten men 5.
It is tempting to turn this into "women need less rest". The notes behind this page could not support that, and the studies that manipulated rest within women pointed the other way. In the only chronic trial, 18 active women trained for five weeks at 15–20 reps a set; 80 seconds of rest improved their three-rep leg press by 45.9% against 19.6% with 20 seconds 6. In 18 young women doing leg press sets to exhaustion at a 10–12 rep load, four minutes of rest produced more total volume than two 7; in 14 resistance-trained women, one minute of rest produced less total volume on chest press and leg press than three 8. So the sex difference is real for submaximal sets stopped short of failure and mostly disappears for hard sets on the big lower-body lifts. No training study has randomised women to different rest lengths and measured muscle growth.
Between sessions the picture is simpler. The best-controlled recent comparison found no sex effect on strength, bar velocity or soreness over 24–72 hours after a bench session taken to failure, even though the women had done about twice the reps 2. Smaller studies split both ways: after squats, 8 women lost more jump height at 24 hours than 11 men (20% versus 10%) 9; in untrained people, recovery was slightly slower in women in one study 10 and no different at 24–72 hours in another 11. A Bayesian meta-analysis of 29 studies with 1,537 female data points found relative muscle growth did not differ between the sexes (0.69%, interval spanning zero) and concluded that programs "can follow similar structures" for both 12.
How EVLVE is built
The general evidence on rest is mostly in men. A Bayesian meta-analysis of nine studies (six male-only, one female-only) found a small hypertrophy benefit for resting longer than about 60 seconds and nothing detectable beyond about 90; the arm effect was 0.13 and the thigh 0.17, both with wide intervals 14. In 21 trained men, three minutes of rest produced more squat and bench strength and more thigh thickness than one minute over eight weeks 15. A review of 23 studies (413 men, 78 women) concluded that trained people need more than two minutes to maximise strength, while 60–120 seconds appears sufficient for the untrained 16. The 2026 ACSM position stand, an overview of 137 reviews, found that rest shorter or longer than a minute did not affect strength and that there were "insufficient data" on hypertrophy 13. Rest, then, is less a lever on the outcome than on how much work a set delivers and how long the session takes.
The women-specific data, all from single sessions, say where time can be saved: upper-body sets and short sets. Women resting one minute on the bench kept about as many reps as men resting three 1, and in 36 physically active women 60 seconds fully restored knee-extensor torque between short maximal sets while 15 and 30 seconds did not 17. For hard lower-body sets, women still paid a volume cost at two minutes compared with four 7. Rest should therefore be keyed to the lift's mechanics and how close to failure the set ends, not to a number pinned to each exercise.
How EVLVE is built
Rest is looked up from the resolved exercise's mechanic and the rep zone in force. No exercise carries its own rest, so a substitute never brings a stray number with it. In seconds:
If she falls under the bottom of her range on set two or later, that exercise's rest grows by 30 seconds at its next exposure (at most +60) until two exposures back in range. After the last set there is no timer. The table is a time budget as much as an efficacy choice: a full-gym lower session with these rests already estimates at about 56 minutes, which is why accessory rests are not longer.
Training until the next rep cannot be completed is not better for growth than stopping a rep or two short. Across 15 studies (11 male-only, one female-only, three mixed), sets taken to any definition of failure outgrew non-failure sets by a trivial 0.19 that did not survive sensitivity analysis; with failure defined strictly the effect was 0.12 and not significant 18. In the one direct within-person trial in trained lifters (12 men, 6 women), one leg trained to failure and the other to a perceived one to two reps in reserve: quadriceps thickness grew 0.181 cm versus 0.182 cm, with more fatigue in the failure leg 19. The ACSM stand reached the same conclusion from three reviews of 1,371 people 13. Closeness still matters: a meta-regression found hypertrophy increased as sets ended nearer to failure while strength showed a negligible relationship, though its authors called the analysis exploratory 20.
How well can a woman judge her reps in reserve? About as well as a man, which is to say roughly one rep off. A meta-analysis of 12 studies and 414 participants found people under-predicted their remaining reps by about 0.95 on average, with accuracy better close to failure and worse in sets longer than 12 reps 21. In 34 young trained adults, accuracy was higher at one rep in reserve than at three and higher at 75% of maximum than at 50%, with no sex effect except in one exercise 22. The direction of the error matters: people think they are closer to failure than they are, so a novice who logs "two in reserve" probably had three or four left.
How EVLVE is built
The claim that training should follow cycle phase rests on two old studies: one inferred ovulation from body temperature in about 20 women; the other had 59 women train unsupervised, assumed 28-day cycles, mixed in contraceptive users, and reported a lean-mass change of 15.1 to 15.3 kg, inside the scanner's measurement error. An umbrella review called the evidence for phase differences in strength, performance or hypertrophy "scant low-quality and largely inconsistent" and concluded that phase-based prescription "is not an evidence-based approach". It also noted why calendar tracking cannot rescue the idea: cycle length varied by more than seven days within women in one sample of 130, and a temperature rise matched the hormone surge in only 22% of cycles in another 23.
Since then the evidence has become direct. In 12 young women whose phases were verified by blood and urine, the rise in muscle protein synthesis after resistance exercise was the same in the late follicular and mid-luteal phases (1.52 versus 1.46 %/day) 24. In the first rigorous randomised trial, 24 women with confirmed ovulatory cycles trained across three cycles with legs assigned to continuous training, high-volume follicular with low-volume luteal, or the reverse. All training beat control for thigh lean mass, cross-sectional area and one-rep max, and the three conditions did not differ; the authors wrote that phase-based adjustments "could be based on individual preference but are not necessary" 25. The honest caveat comes from a meta-analysis of 22 studies and 433 women: relative to the first days of the period, isometric strength was higher in the late follicular phase (effect 0.60) and isokinetic strength higher at ovulation (0.39), but 15 of the 22 studies were rated low or very low quality 26. That is a reason to allow an easier day, not to re-plan the month.
Oral contraceptives do not measurably change what training does: in a meta-analysis of 325 participants, differences between pill users and non-users were 0.01 for hypertrophy, 0.10 for strength and −0.04 for power, all spanning zero 27. Other hormonal methods have not been tested; that is a gap, not a reassurance.
How EVLVE is built
The dose-response data are mostly male. The largest analysis, 67 studies and 2,058 participants (79% male, mean age 25, about ten weeks), found muscle growth keeps rising with weekly sets at a slowing rate: about four "fractional" weekly sets is the smallest dose with detectable growth, and each further detectable step costs roughly six more sets between 5 and 10, 8.5 between 11 and 18, and nearly 11 between 19 and 29. Counting a set in full for the prime mover and as a half for assisting muscles fit the outcomes better than any other count. Strength flattened after about four weekly sets and responded to frequency instead, rising from a 12.7% gain at once a week to 17.3% at twice 28. Per session, growth could not be told apart beyond about 11 fractional sets for one muscle 29. The ACSM stand's hypertrophy enhancer is at least ten weekly sets per muscle group 13; the time-efficiency review sets a floor of four weekly sets and notes that weekly volume matters more than how it is split across days 30.
Three findings matter for a program meant to run for years. Adding sets over a block is not required: 30 recreationally trained women who trained their quadriceps twice a week for 12 weeks at a constant 22 weekly sets grew as much as those ramping to 26 or 38 31. The claim that more than ten weekly sets is counter-productive in trained women traces to a paper that data sleuths formally asked to have retracted in 2020 for "atypical, improbable" data patterns 32; EVLVE rests nothing on it. Breaks are survivable: in 55 untrained adults (45% women, mean age 32), a ten-week break cost about 5% of leg press strength and about 10% of quadriceps size, five weeks of retraining restored everything, and there was no sex difference 33. And a dose, once built, can be held on less: in adults aged 20–35, one session a week at the same load held muscle size for 32 weeks at a third of the building volume 34. The notes found no trial showing a planned deload improves outcomes; the ACSM stand is silent on deloads 13.
How EVLVE is built
US guidance asks adults for 150 minutes of moderate activity a week, or 75 vigorous, or an equivalent mix, plus muscle strengthening on at least two days 35. The notes corrected two popular claims. First, that "zone 2" is a uniquely valuable intensity: a 2025 review, read through a digest, concluded that at equal volume fitness gains are larger above zone 2 and that in already-active people zone 2 alone is generally not enough to raise maximal oxygen uptake 36. Second, that one hard interval session a week raises fitness: in the one trial that isolated a once-weekly dose, 26 recreationally active adults (14 women) did four rounds of four minutes hard once, twice or three times a week for six weeks. Once a week left maximal oxygen uptake unchanged (53.0 to 53.6), twice raised it (53.5 to 57.6), and a third session added no clear benefit; a small exploratory study in 23-year-olds 37.
How EVLVE is built
A program that pairs four lifting sessions a week with a shape goal has to say something about food, carefully. The 2023 IOC consensus defines relative energy deficiency in sport (REDs) as impaired physiological or psychological functioning caused by prolonged or severe low energy availability. Reproductive impairment is among the best-documented consequences, missed periods are both a sign and a symptom, and some effects reverse when eating improves while bone mineral density loss may not. The thresholds often quoted (about 45 kcal per kilogram of fat-free mass per day as adequate, 30 or below as low) came from sedentary women, and the statement says the 30 figure "was intended as a guide, rather than a diagnostic end-point". In athletes with menstrual dysfunction, a year of transdermal oestradiol raised bone density scores while combined oral contraceptives did worse, which is why a withdrawal bleed on the pill is not evidence that a woman is eating enough 38. The prevalence of low energy availability among recreational women lifters is not known. The practical point is that bone and muscle goals pull against aggressive dieting, and an app that cannot measure energy availability can still say when to see someone.
How EVLVE is built
Chapter 5
This chapter covers how a strength program should decide, session by session, when a woman adds reps, when she adds weight, and what happens when she cannot. It reviews the trials on reps versus load, effort ratings, rates of progress in women, deloads and time off, then states what an audit of EVLVE's previous progression engine found and the rules that replaced it. The conclusion: reps and load are interchangeable levers for muscle, load must eventually rise for strength, and the engine's job is to move her up her own equipment in steps her reps can absorb, never taking weight away without a reason she can see.
The clearest trials compare two ways of making a set harder: keep the load and add reps, or keep the reps and add load. In 43 resistance-trained adults (27 men, 16 women) training eight weeks with every set to failure, the group that held 8 to 12 reps and raised load gained about as much muscle as the group that kept its starting load and added reps (rectus femoris thickness +10.1 mm against +12.3 mm), while squat one-rep maximum rose 21.8 kg against 19.3 kg, slightly favouring load 1. In 39 untrained adults (20 men, 19 women) training one leg each way for ten weeks, strength and size rose the same on both legs: 1RM 52.9 to 69.1 kg against 51.7 to 66.8 kg 2. Both trials took every set to failure, and the reps group in the first "appeared to have greater difficulty approaching true failure" 1.
The wider literature agrees. The 2026 ACSM position stand, an umbrella review of 137 systematic reviews covering more than 30,000 adults, found no effect of load on muscle growth between 30% and 100% of maximum when effort is matched, but a dose-response for strength favouring loads of at least 80% of maximum; progression is not required for benefit but is required for continued gains 3. The floor is about 30% of maximum; at 20% growth was about half, and few studies included women 4. In 25 untrained men, 30% loads matched heavy loads only when sets went to failure (quadriceps +7.7% with failure, +2.6% without), whereas at 80% failure added nothing 5.
Adding sets has weaker support. In the one trial in trained women (30 women, 12 weeks), a constant 22 weekly sets grew the vastus lateralis as much as adding 2 or 4 sets every two weeks, read second-hand because the paper could not be retrieved 6. A meta-regression of 67 studies (21% female) found diminishing returns to weekly sets, sharper for strength 7. Tempo has tolerance rather than progression value: repetition durations of 0.5 to 8 seconds grew muscle similarly, with reps over 10 seconds inferior 8. Rest interval had no effect on strength 3, so density is not an overload variable.
How EVLVE is built
Every loaded lift runs rep-then-load progression on the equipment she owns: the rep target follows her worst set (worst plus one, up to the move-up number), and when every shown set reaches that number all sets move to the next weight and the target restarts at the bottom. Rep zones are wide where equipment steps are large: `build` 8 to 12 for compounds, `mid` 10 to 15, `wide` 10 to 20, `high` 12 to 20, `band_high` 12 to 25. Caps limit the most reps ever asked for: 15 on two-leg lower-body lifts and upper-body compounds, 20 on single-leg and light single-joint work, 25 with a band. Sets are never added as progression; the old engine's promise of earned sets was removed. The caps are inferences from the reps-to-load curve and the discomfort of long sets, not trial results.
After each exercise EVLVE asks whether it was Easy, Right or Hard. The evidence limits what such an answer can mean. In a meta-analysis of 12 studies and 414 participants, people under-predicted the reps they had left by 0.95 on average, believing they were closer to failure than they were; the error barely grew with set length up to 12 reps (0.06 per rep) but steeply beyond it (0.47 per rep), and training status made no difference 9. In 27 men and 31 women, accuracy was unaffected by sex or experience and improved closer to failure 10. Well-trained lifters calling their reserve on 70% squats were off by 5.15 reps 11. A woman can tell near failure from not near failure; she cannot count reps in reserve beyond three or four, or in sets over 12.
Choosing loads by effort works at least as well as fixed percentages: in 21 trained men an RPE-guided group gained 17.1 kg on the squat against 13.9 kg, not significant, and trained heavier as a result 12; a meta-analysis of six studies, five in men, found a 2.07 kg edge (95% CI −0.32 to 4.46) and called the approaches equivalent 13. The larger risk runs the other way: left to choose, people pick about 53% of maximum regardless of sex or experience 14, 25 or more reps from failure. No study validates a three-option rating in anyone. Failure is not required: it had no effect on strength and no consistent effect on growth, and the ACSM suggests stopping two to three reps short 3; growth improves modestly, strength not at all, as sets end closer to failure 15.
How EVLVE is built
Reps achieved are the primary signal; the answer modifies them. On a normal target, Easy below the top of the range does nothing (amendment A9), Right keeps building reps, and Hard with a jump of 10% or more repeats the move-up number once before moving. "Failed" was retired; missing the range is read from the reps. Because under-loading is the dominant error, an estimated starting weight is pre-filled at the middle of its range (E10), and an Easy answer with reps at or above the middle moves the load once per exposure, by at most 10% or one rung within 15% (A3), stopping the moment a set is rated Right or Hard. On sessions 1 to 8, after a 12-week layoff, and the first time she does a new loaded lift, the last set reads "as many good reps as you can, stop when your form changes"; beating the top of the range by 4 or more moves the load at once to the rung predicted to land mid-range, capped at +20%. That calibration set is never used on a barbell or trap-bar lift (A1). Effort shown is "leave 2" on main lifts, "leave 1" on accessories, and the last set "to the last good rep" only on bands and her two lightest dumbbell pairs. Nothing above "leave 3" is ever displayed. The mapping of Easy, Right and Hard to reps in reserve is an inference; the trials do not settle it.
Untrained women gain strength at least as fast as men in relative terms. A pooled analysis of 63 comparisons and 3,332 mostly untrained subjects, not peer reviewed, found gains of 37.4% in women against 29.4% in men, with the largest gap in the upper body (47.5% against 34.9%) and equal muscle growth (12.2% against 13.2%) 16. A women-only meta-analysis of 24 studies found large strength effects (upper body g = 1.70, lower body g = 1.40) and moderate hypertrophy (g = 0.52), with supervised training beating unsupervised 17. Strength rises at roughly 1.6 to 2.0% a week early on and about half that in trained people 18, and gains slowed after six weeks even in untrained men 19.
The problem for women is step size. Between 6 and 15 reps one rep is worth roughly 2 to 2.5% of maximum, so an 8 to 12 range absorbs only about a 10% jump 20. Below 25 lb per hand, where most upper-body and isolation work sits, every step on a standard rack is a 20 to 67% jump. A meta-regression of 269 studies found sex, age and experience do not change the reps-to-percentage relationship but the exercise does (about 13 reps on leg press against 8 on bench press at 80%) 21, with wide individual spread (squat at 70%: 6 to 28 reps) 20. A preprint of 303,494 app-logged sets, 20% from women and without measured maxima, found light dumbbell lifts follow a different curve and every classical equation loses accuracy above about 10 reps 22. The published jump conventions (2 to 5% upper body, 5 to 10% lower body) cite no studies 23.
How EVLVE is built
There are no female-specific slower rules; the evidence does not support them. The engine models her ladder instead. Every jump is computed from two real weights she owns, and the move-up number M is the reps at the old weight that predict the bottom of the range at the new one: M = 30 × ((1 + jump) × (1 + bottom ÷ 30) − 1), rounded up and never less than the top of the range. If M is at or under the zone's cap the jump is absorbable. In the `build` zone, 30 to 35 lb (+17%) gives M = 15; 100 to 105 lb (+5%) gives the top of the range, 12; 20 to 25 lb (+25%) gives M = 18, above the cap, and goes through the bridge. The largest absorbable jumps are 11% in `heavy`, 14% in `strength`, 18% in `build`, 20% in `mid`, 25% in `wide`. The divisor of 30 is an inference and the cautious bound; her logged reps after each jump correct it.
In October 2026 the previous engine was read line by line and run against model members. The headline findings, stated plainly: a member who logged exactly the pre-filled numbers, which the audit notes about 90% of production logs resemble, could lose weight over time. The verified reproduction on a back squat seeded at 50 lb went 50 → 45 → 40 → 35 → 30 lb across 81 sessions with no increase, because the stall detector counted sessions in which the engine itself had frozen the target, so a woman returning from a lighter week was read as stalled and backed off 10%. The deload flag was evaluated one session late, discarding earned progress once every block. Sessions were judged against the normal set count rather than the sets shown, so a lift trimmed for time never moved. Timed holds spiralled from 45 seconds to 5, because the lowered target was pre-filled and fell short again every exposure. A step-in to a heavier dumbbell had no exit at jumps of 60 to 100%. The working weight could be an average nobody lifted. The bar was not modelled, so an empty-bar beginner could be asked for 40 lb on a 45 lb bar. Balance holds and the farmer's carry never progressed; "Hard" never stopped an increase; a too-light estimate took weeks to correct. The evidence notes had already flagged that an automatic 10% back-off after three flat sessions is barbell convention with no trial support.
How EVLVE is built
Eight reading rules now sit under every category. A session is judged against what she was shown, with the resolved plan saved alongside it (R1). The working weight is the weight used on most sets, ties to the lighter, never an average (R2). Earned progress is always written; the engine never predicts whether a future session will be lighter (R4). Lighter-rotation sessions, sessions the engine held, partial sessions and the first rotation after a layoff never count toward a stall, as exposures or as days (R5, A4). Ramp-up sets are never judged (R7). A new exercise starts at its own estimate; nothing carries across exercises except, for the same lift in a new rep zone, a conversion that applies two-thirds of any predicted increase, caps it at 10%, rounds down and marks the result an estimate (R8). Table 10.12 of the design maps every audited defect to the rule that removes it.
No trial has tested double progression as a package, the width of a range, or the trigger rule; it is a convention assembled from two levers that each have support 24. The protocol closest to it inside a trial trained bench press at 3 sets of 6 and, when 3 sets of 8 was reached, added 4.54 kg and restarted at 6; 1RM rose 6.5% in four weeks 25. The reps-to-load curve adds the constraint: after a +20% jump from 12 reps she should expect about 5 reps, after +25% about 3 to 4, after +33% about 1 to 2 20. Ways to absorb a jump a range cannot follow from that curve, none with head-to-head evidence: finer increments, a lower floor for the first sessions, a hybrid session with one set at the new weight, or a harder version at the old weight. A slower lowering is the best-supported interim step: 3-second eccentrics at 50% of maximum matched 1-second reps at 80% over 12 weeks, in small, mostly untrained samples the reviewers call inconclusive 26.
How EVLVE is built
Her ladder is hers. Nothing is a rung unless she registered it or it is the standard gym rack; a logged weight never becomes a rung; barbell ladders start at her confirmed bar weight and step by twice her smallest plate. The bar, the minimum pin and her lightest pair are the floor; nothing is ever asked for below the bar.
For a jump her range cannot absorb, the sequence is: reach the cap on every shown set; a bridge stage, once, at the same weight with the bridge tempo (+2 seconds lowering or a 2-second pause at the long position, whichever the lift lacks), rebuilding reps from bottom to top over at most four exposures; then the step-in, where the first set moves to the heavier weight and the others keep the rep target they had. A heavier set counts at 5 reps; each session every heavier set counts, one more set moves up. Before any step-in the engine predicts her reps at the heavier weight from her best recent set and steps in only if the prediction is at least 5 (A2); otherwise she holds at the cap and is offered the family's harder version. A step-in missed twice is cancelled (E12): she returns to the lighter weight at the rep target she had, no back-off recorded, and may try again after eight exposures. The base weight is never rewritten during a step-in. At the top of her equipment: reps to the cap, the bridge once, the harder version, then an honest hold with a line saying so. One point is open by ruling (E13): a member who only logs the pre-fill will hold at the cap on dumbbell steps of about a third (15 to 20 lb) and be offered the harder version instead; whether to add a last-set rep test there waits for launch data.
Short of the range: one short exposure holds; a second goes down a rung with the rep target at the top of the range; at her lightest weight, two short exposures re-rank the slot to an easier version at once (A8), and no back-off is announced. The bridge, the count of 5 and the two-miss cancel are conventions; the trials do not settle them.
Bands build strength comparably to weights: a meta-analysis of eight studies found no difference for upper-limb (SMD −0.01) or lower-limb strength (0.09) across diverse populations 27, and band training improved hypertrophy against no training 3, though conventional equipment is preferred for heavy lower-body lifts 28. There is no retrievable evidence on stepping between bands, and tension rises with stretch, so load is highest where the muscle is shortest, the opposite of the stretched position the range-of-motion evidence says to emphasise 29. For bodyweight ladders there is one randomised trial: a ten-level push-up ladder with the rule "3 sets of 8 in two consecutive sessions, then restart the next level at 3 sets of 6" raised bench 1RM 3.4% in four weeks, not significantly different from bench training, in 23 participants, all men in the thesis record 25. Push-ups and pull-ups have good evidence; lower-limb bodyweight research is sparse 28. For holds the retrievable guidance is clinician opinion: 10 to 30 seconds in several sets, building to one or two minutes at most, after which the position should get harder rather than longer 30.
How EVLVE is built
Bands are logged as the band by kind and level plus a set-up rung, never as pounds. The order is reps to the top, then the next set-up rung on the same band (mini loop: knees, ankles, feet; anchored: a step further; standing: a wider stance; then doubled), then the next band she owns by bridge first and step-in (B12), a heavier set counting at 8 reps, two misses cancelling. Upper-body band work runs 12 to 25, lower-body 12 to 20. Nothing is prescribed on a band she does not own; no core exercise uses a band.
Bodyweight families are ordered rungs, each a real exercise with its own range. Every shown set at the rung's top in two consecutive exposures, the trial's own rule, triggers an offer of the next rung she can perform, with "the same rung, loaded" ahead of the next mechanical rung when she owns an adequate load; declining brings the offer back after eight sessions. Under the bottom twice, the rung below is applied at once (E11), and every family has a rung below its usual entry. Assisted pull-ups keep a 5 to 8 range and a three-miss rule; a bar alone is not enough for the slot.
Timed core holds gain 5 seconds when every set reaches her target, to a cap of 45 seconds front and 40 per side (anti-rotation 30); at the cap twice, the next rung is offered. Short twice, the target becomes what she held, rounded down to 5 seconds, so a short spell cannot repeat itself; the floor is 10 seconds. Holds are never loaded. The two-handed carry runs 30 to 45 seconds in 5-second steps, then the next rung if the jump is 20% or less, otherwise one set steps in. The hold cap and the carry rule are conventions.
Short flat spells are expected: beginners add weight roughly weekly for a few months, then plan a small personal best per 4-week block, then per 8, then per 12 31; stalling for a couple of weeks is normal and should not be forced 24. Deloads have little trial support either way: a week off at week five neither helped nor harmed growth in 39 trained adults (10 women) but slightly blunted strength (squat +13.0 kg against +16.4 kg) 32; a reduced-volume deload was neutral in 19 untrained men 33; a Delphi panel of 21 coaches agreed only that volume is reduced and that benefit is unresolved 34. Periodized phases add a small strength edge (1RM ES 0.31) and nothing for growth, and nothing at all in untrained people 35. On time off, a review of 20 studies (3 female-only) found no significant strength loss at up to four weeks and strength still about 12% above baseline at 12 weeks while gained muscle was mostly lost, with low evidence quality 36. Early loss on skilled lifts is often rusty motor patterns that return within two or three sessions 37. No trial tests any re-entry rule.
How EVLVE is built
A stall is three exposures spanning at least 21 calendar days at the same weight with no rep gained on the worst set, effort Right or Hard. The response takes one step per stall: repeat with 30 seconds more rest; offer the lighter rotation if she has not had one this block; offer a close variation or the adjacent rep zone; only then lower the load one rung if within 10%, otherwise reset the rep target to the bottom. The lighter rotation is an offer after 16 work sessions, pre-selected only when they took 42 days or fewer, at half the sets with reps pre-filled at her target minus 2 and the same weight; its logs change nothing. Heavy work is a standing zone: after Foundation, bar owners squat at 5 to 8 and deadlift at 4 to 6 every rotation.
Re-entry is keyed on days since she last logged that lift, at the conservative end of each range (A6): up to 14 days unchanged; 15 to 28, rep target at the bottom; 29 to 70, one rung down if within 10% and one set fewer for a rotation; 71 to 84, about 10% off; over 84, 15% off, targets marked as estimates, 2 sets per slot with "leave 3", calibration sets on. For two rotations after a layoff of 28 days or more, Easy moves load as on an estimate, because regain is quicker than first-time gain. The table is convention.
The rules are written as 109 numbered statements, each tied to a branch in a reference implementation and to the vectors that prove it: 287 test vectors, all passing, which the production port must also pass. Alongside them is a simulation placing a model member through 96 sessions under the old engine and the new rules. It is a model, not data; no member's logs were used, and it says what the rules do to a stated member, not what members will do. Under the old engine the pre-fill follower on a barbell Romanian deadlift went 60 to 45 lb and was then asked for 40 lb, below the bar; under the new rules she goes 60 to 80 lb with 4 increases and 0 back-offs, one rung below a member who trains every set to one rep in reserve (60 to 85 lb). With 5 and 10 lb pairs the new rules start her on the single-leg hip thrust and hold honestly at 10 lb × 20 with a line saying the pairs are the limit, where the old engine left her at 9% of capacity. Across every run no load was lost and no lighter rotation changed a target.
Chapter 6
This chapter covers two things a program must get right before anything else: what a woman actually has to train with, and what her body is asking her to work around. It reviews the trials on bands, dumbbells, machines and bodyweight, how far a pull-up progression really goes, what the guidelines support for knees, backs, shoulders, wrists, hips, the pelvic floor and hypermobile joints, and how EVLVE's resolver turns that into a session. The conclusion: for muscle, equipment barely matters and a limitation is a reason to adjust rather than stop, so the program treats her equipment list and her considerations as hard facts and builds everything else around them.
For muscle size, what she owns barely matters. A meta-analysis of 13 studies with 1,016 participants (789 men, 219 women) found no difference in hypertrophy between free-weight and machine training (standardised mean difference −0.055, 95% CI −0.397 to 0.287), and no difference in strength tested on the same task 1. The one head-to-head growth trial for an isolation lift says the same of cables versus dumbbells: in 24 trained adults (8 women) who trained one arm each way for eight weeks, the side of the shoulder grew about 3–5% with either implement, although the cable loaded the stretched position and the dumbbell did not 2. Strength, by contrast, is specific to the implement practised 1.
Bands are a legitimate version of the same thing: in a 10RM-matched comparison of reverse flyes (17 men, 12 women), bands produced prime-mover activity close to dumbbells 3, and ten weeks of bands matched weight machines in 45 postmenopausal women 4. Two caveats. First, these comparisons are short, in beginners or older groups, and measure strength more than size; a review calls bands viable for most work but recommends conventional equipment for heavy lower-body lifts 5. Second, a band's load is unknowable from its label. In a bench test of 141 loop bands, nominally identical bands spread from −37% to +16% around their mean 6; a 48 mm loop rose from 5.5 kg to 26.5 kg between 110% and 190% of resting length, and two "identical" bands differed by 4–23% 7.
Where equipment genuinely matters is bone and heavy strength. A band gives bone nothing and light dumbbells give it very little; only a barbell or trap bar provides the heavy axial loading used in the bone trials, so a home member's bone work is her impact work. Bands also carry an injury risk weights do not: one emergency department saw 11 eye injuries from recoiling bands in six months 8.
How EVLVE is built
Implement is a preference, never a ranking. Bands are logged by kind (mini loop, tube, long loop), level and set-up, never by pounds. The band rep zone is 12–25 upper body and 12–20 lower body; progression runs reps to the top, then the next set-up rung on the same band (a mini loop moves knees → ankles → feet; an anchored band moves a step further away; then the band is doubled), then the next band she owns, by bridge first and then step-in. One safety line shows at band set-up: check for nicks, anchor it, keep it off a line to your face. No core exercise ever uses a band or added load; that is a program rule for the waist, not an evidence finding. The zone and rungs are a judgement call; no trial tests a band progression scheme.
Pressing is fully trainable with a floor and a chair. In 18 men, eight weeks of push-ups adjusted to the load of a 40% one-rep-max bench press grew the chest and triceps and raised bench strength as much as the bench press did 9. The push-up ladder is also a finer progression tool than a light dumbbell set: hand loads run from 41% of body weight with the hands raised 61 cm, through 49% kneeling, 55% at 30 cm and 64% on the floor, to 70–74% with the feet raised, steps of 6–20% 10.
Pulling is the gap. Without something to pull against, the lats, biceps and rear shoulders cannot be loaded progressively; prone "superman" raises are loaded by arm weight alone and have no data behind them. The only evidence-supported bodyweight pull is an inverted row under a bar or straps, and the safety of rowing under household furniture is untested. A bodyweight-only member honestly has push plus trunk; one long band with a door anchor is the cheapest thing that adds a pull. The calisthenics trials are short and almost entirely in men 5.
How EVLVE is built
Every family the program uses for squat, thrust, lunge, abduction, curl, push and core has at least one rung that needs only the assumed basics, so a bodyweight member always receives a full lower-body and core session. The push slot runs wall → chair → floor → feet raised, then band-resisted. Pull slots are not faked: when no candidate resolves they are omitted and she reads "Pulling is not in your plan yet. Lats, biceps and rear shoulders need something to pull against, and nothing you have listed gives that. One long band with a door anchor adds it."
Most untrained women cannot do a pull-up, and the one trial that tried to change that is sobering: 17 university-age women who could not do one trained three days a week for about three months, raised upper-body strength by 36%, and four could do a pull-up at the end 11. Pulldown strength does not predict it: in 28 elite collegiate swimmers, pull-up count correlated only weakly with pulldown one-rep max (r = 0.34) and negatively with body mass (r = −0.48), and the authors said the two should not be substituted for each other 12.
Band assistance is uneven. From the measured loop-band tensions 6 7, a foot-in-band pull-up for a 65 kg woman gets roughly 12–20% of body weight from a ½-inch band, 30% from a 1⅛-inch, 45% from a 1¾-inch and 65% from a 2½-inch band at the bottom, and much less at the top where the band goes slack. Each band change is a jump of roughly 18–57% in what she lifts. This is arithmetic from bench tests, and no trial compares pull-up progressions in women.
How EVLVE is built
A pull slot is eligible for the pull-up family only when she has a bar and a way to scale it: a long loop band, a chair or step for the feet-assisted version, or three unassisted reps on record. A bar alone is not enough. The ladder runs heaviest loop, foot in → same band, knee in → next lighter loop → … → lowering only (5 s down, 3–5 reps) → unassisted, in a 5–8 range (convention). Less help is earned by three sets of 8 in two consecutive exposures; the first set moves up and counts at 3 reps or more; three misses cancels the step. The card says: "A first pull-up usually takes months. Your rows are doing the building meanwhile." Rows and pulldowns carry the lat volume throughout.
The guidelines for every common limitation point the same way: load management and graded exposure, not avoidance. Leakage of at least moderate severity affects 6.9% of US women aged 20–39 and 17.2% at 40–59, rising with the number of births 13.
Knees: the mechanism is depth. Kneecap stress rises steeply with knee bend 14; in 16 healthy adults (8 women), a side lunge produced 1.5–2 times the stress of a forward lunge between 40° and 100°, and stepping onto a 10 cm platform lowered mid-range stress 15. The guideline makes combined hip- and knee-targeted exercise the treatment and recommends against braces, sleeves and alignment biofeedback 16, so hip thrusts and hinges are the recommended work under a knee flag, not the thing to remove.
Lower back: a supervised deadlift program, started at 10 kg and progressed on technique and pain, matched individualised motor-control exercise in 70 adults (39 women, mean age 42) with mechanical back pain of more than three months; the women reached a mean of 53 kg 17. A trap bar lowers the peak load at the lower back and raises it at the knee, in 19 male powerlifters 18.
Shoulders and wrists: progressive loading is the core treatment for shoulder pain; the usual position rules (presses and pulldowns in front of the body, bench grip at or under 1.5 shoulder-widths) are convention from surveys of mostly male lifters 19. Wrists are two problems: on a bent-back wrist, push-up forces shift toward the back and little-finger side of the joint and a neutral wrist is probably safer 20, whereas carpal-tunnel pressure rises with any grip load regardless of wrist angle 21, so numbness calls for less gripping, not a straighter wrist.
Outer hip, pelvic floor, hypermobility: in the gluteal-tendon trial (204 adults, 82% women, mean age 55, older than EVLVE's members), education to avoid compressive positions plus heavy slow strengthening beat an injection and wait-and-see, with 78% improved at a year versus 57% and 52% 22; the protocol names the piriformis and iliotibial-band stretches as positions to avoid 23. For the pelvic floor, one controlled session of heavy squats and deadlifts at 75–85% of maximum changed nothing measurable in 47 strength-trained young women who had never given birth 24; among 3,934 surveyed women, light lifters reported more prolapse symptoms than heavy lifters 25; a scoping review found parity, not recreational lifting, the consistent factor 26; and jumping, not lifting, is the most consistent trigger of leakage 27. For hypermobility the risk shown in trials is under-loading: twelve weeks of self-guided low-resistance training produced no gains at all in 51 hypermobile women aged 20–40 28, while heavy supervised dumbbell loading beat light band-and-isometric work for hypermobile shoulders in 100 adults, 79% women, mean age 38 29.
How EVLVE is built
Nine considerations plus one preference: my knees, my lower back, a shoulder, my wrists when bent back, numb or tingling hands when I grip, the outside of my hip, a pinch at the front of my hip, leaking or heaviness, very flexible joints, and low impact throughout; any number at once. Each reads position tags, not exercise names: every exercise, drill and stretch carries tags from a list of 25, and a table of 66 rules maps each flag to an action on each tag: 15 avoid, 14 modify, 10 prefer, 6 deprioritise, 21 cue. A modify applies only where the exercise lists that modification; otherwise it acts as avoid.
Knees: depth set by a box; side lunge and kneeling avoided; the hip thrust preferred and never swapped away; impact on the low track (heel drops). Lower back: the hinge kept and started light; the trap bar preferred where owned; supported rows first; no stretch removed. Shoulder: neutral grip, in front of the body, reduced depth; no bar on the back, no passive hangs, nothing behind the neck. Wrists: fists, handles or raised hands; grip: shorter sets and the hip thrust ahead of the heavy hinge. Outer hip: curtsy lunges, pigeon, figure-4 and cross-body glute stretches avoided; slow standing band abduction the reference tool. Pelvic floor: low impact first, symptom-guided load, no exclusions, and a signpost to a pelvic health physiotherapist, because supervised pelvic-floor muscle training is the treatment with evidence 30. Hypermobility: loading targets unchanged; cues ("soft lockout") and active range drills in place of passive stretches. The shoulder, wrist and front-hip rules and most stretch swaps are convention; the rest rest on the trials cited.
Pain during exercise is not, by itself, a reason to stop. A systematic review of seven trials (385 people with chronic musculoskeletal pain) found protocols that allowed some pain gave a small short-term benefit over pain-free protocols (SMD −0.28) 31. The thresholds come from supervised trials: the gluteal-tendon protocol tolerated pain up to 5 out of 10 during heavy strengthening provided it eased afterwards and was no worse that night or the next morning, and accepted no increase at all during movement retraining 23; the hypermobile-shoulder trial accepted the same 5/10 29; and continuing to run and jump under a pain-monitoring model did no harm in 38 people with Achilles tendinopathy 32. All were clinician-supervised, so using the rule unsupervised is an extrapolation.
How EVLVE is built
Two tiers, as the trial had. Strength work: "Some discomfort is fine — up to about 5 out of 10 — if it eases when you stop, is no worse tonight or tomorrow morning, and is not creeping up week to week." Impact, balance and prep drills: "These should not add to any pain you came in with." Every strength card carries "This doesn't feel right", which offers the least change first: lighter, then slower or shorter, then a different exercise, then skip. Pain above 5, sharp pain, or pain that lingers into the next day stops that exercise for the session; a note persisting across two weeks adds "A physiotherapist can look at this properly." A fixed refer-out screen, drawn from patient-facing NHS and NICE guidance, is never answered with a swap: a joint that locks, gives way or cannot bear weight; swelling, change of shape or heat with a fever; pain after an injury or of sudden severe onset; numbness in both legs or the saddle area, or a change in bladder or bowel control; persistent pins and needles in an arm or hand; a vaginal bulge or heaviness, or leaking that limits training; a sudden severe headache during effort; chest pain, fainting or unusual breathlessness; pain that disturbs sleep or lasts beyond two weeks; missed periods while training hard. The app never names a diagnosis or advises on medication, scans or braces; a clinician's advice overrides it.
The audit found a gym template filtered by equipment that still scheduled band exercises for members without bands and gave home members gym-rack weights as targets. The substitution evidence says what a correct replacement must preserve: the primary muscle (a hip thrust matched a squat for glute growth but grew neither the quadriceps nor the adductors, in 34 untrained young adults, 23 of them women 33), enough load to progress, and, when the exercise carries the bone goal, axial load or impact. The implement is free 1, and choosing from a curated list per muscle did no harm in 17 trained men 34. So: equipment selects the program, considerations filter it, she can swap within the same class, and the swap persists.
How EVLVE is built
Every exercise, drill and stretch carries one requirement object: which items must be present, which alternatives suffice (a bench, or a chair, or a box), how many dumbbells, which band kinds and whether an anchor is needed, and any logged capability such as three unassisted pull-ups. One pure function, canPerform, serves the resolver, the swap sheet, the offers, the onboarding preview and the tests. The list has 12 checkboxes plus an explicit "no equipment" answer, 8 follow-up details (her actual dumbbells, band kinds, a door anchor, bench incline, step height, bar weight, stack step) and 5 household basics; a blank list cannot be saved, and bar weight is asked once, never assumed.
For each slot the resolver takes a candidate pool in three tiers (A: same pattern and primary region; B: same region, different pattern; C: bodyweight rungs on assumed basics), drops everything canPerform rejects, then everything carrying an avoid tag for any active flag. Those two filters are never relaxed. Only then come level, her standing swap, ranking by loadability and goal fit, rotation, a collision check so no exercise appears twice, and time fitting. When nothing survives, the slot is omitted and a gap statement attached; the time is not refilled. The swap sheet is computed, never typed, and a swap persists, opens at its own target and is re-checked every session.
The space is large by design: 138,240 equipment lists a member can actually hold × 4 level and experience combinations × 3 session lengths × 7 positions in the program × 61 combinations of considerations = 708,341,760 member configurations. Running every one would take about 1,300 processor-hours, so the full sweep is not run; every three-way combination of the factors is covered instead, in a sweep that completes in under two hours and runs before any change to the library ships. Eighteen invariants are checked on every resolved plan, among them: zero items where canPerform is false and zero with an avoid tag for an active flag, swaps and offers included (I1); every prescribed weight a rung of her own ladder and every band one she owns (I9); at least two real alternatives per strength slot in fourteen reference equipment situations (I10); never a loaded core exercise or a jump holding dumbbells (I15); a gap statement attached if and only if its condition holds (I18).
The gaps are stated, not papered over. There are 18 named gap statements, each shown once when it first applies, with the smallest fix: no pull ("One long band with a door anchor adds it"); no side-shoulder work; glute load capped, so the lift moves to one leg; no loaded hinge; bone from impact only, shown as education with no purchase suggested; bone reduced while jumping is out ("Heel drops stay in, and they are the best low-impact option we know of"); no heavy strength without a bar; pressing omitted while wrists settle; overhead strength not trained for now; joints that do best with real load. The "add next" line shows the first rung she is missing, from a long band with a door anchor up to a bar or a gym; the ordering is inference.
The short version
Impact and heavy loading for the hip and spine. Enough sets for the glutes, counted. Nothing loaded through the waist. Rest by the lift. Effort asked once. Balance practised, not assumed. A weight she owns, never one she does not. And no claim the papers do not support.
The audits, the program blueprint, the engine rules and the test vectors behind this page are available on request.