Peripheral scope (exercise-programming) — admitted on the same evidence bar as any exposure, kept low in the ranking because attention is an anti-signal and this is a heavily-discussed, mostly small-between-option domain. It earns a page only because the evidence is gold (a Bayesian network meta-analysis, 178 strength / 119 hypertrophy RCTs) and it settles one genuinely decision-relevant thing: the resistance-training dose is not one dial — strength and hypertrophy are driven by different variables, and the gap between any-RT and no-RT dwarfs the gap between prescriptions.
Evidence-tier note (confidence: low). Single gold NMA (Currier 2023). The surrogates (1RM
strength, muscle size) are RCT-grade; the health-outcome transmission is not established by this source
(see the surrogate boundary below), and the primary trials are unblindable (moderate–high risk of bias).
Held low on total web support pending an independent line — and Currier is not independent of the
staged ACSM 2026 stand or of Morton’s protein RCT-MA (shared Phillips/McMaster team), so a second
same-lineage source would not raise it. (inferred from Currier et al., 2023)
The big rock: any prescription beats none — the between-prescription differences are second-order
Currier compared 12 prescriptions (load H ≥80% 1RM / L <80%; sets M multiset / S single; frequency ≥3 / 2 / 1 per week) against non-exercise control (CTRL). Every estimated node had a positive point estimate vs CTRL — all 12 for strength (SMD 0.75–1.60 vs CTRL) and the 10 with hypertrophy data (SMD 0.10–0.66; HS1/LS1 are N.D., no data, for hypertrophy in Table 2); the 95% CrI excluded zero for all but the sparse nodes HS1/LS1 (strength) and HM1/HS2/HS3 (hypertrophy) — see the minimal-dose section. But once you are training, the choice of prescription barely separates: «The 95% CrI contained zero for a striking 91% (101/111) of all between-RTx comparisons.» So the decision that carries the effect is train vs not-train, not which protocol — which is why Currier’s own conclusion is that «adults should engage in RT, even if they cannot meet existing recommendations», not that they hit an optimal scheme. This is the Layer-1 ranking made concrete: the first dollar (start RT) buys almost everything; optimizing the prescription is the long tail.
The decomposition (the value): strength is load-driven, hypertrophy is volume-driven
The one place prescription does matter splits by outcome — and the two do not track together, so the resistance-training dose is a terminological conflation of two different curves:
| Outcome (surrogate) | What drives it | Top-ranked RTx | Effect vs CTRL (SMD, 95% CrI) | Between-RTx separation |
|---|---|---|---|---|
| Strength (1RM) | load (≥80% 1RM), multiset | HM3 (heavy, multiset, ≥3×/wk) | «1.60 (1.38 to 1.82)» | 9 of 10 non-zero comparisons were HM2/HM3 vs a lower-load RTx |
| Hypertrophy (muscle size) | volume (sets), load ~irrelevant | HM2 (heavy, multiset, 2×/wk) | «0.66 (0.47 to 0.85)» | only 1 of 45 comparisons excluded zero |
- Strength: «higher-load, multiset programmes caused the largest strength gains» — the only variable that reliably separated prescriptions. Robust under sensitivity analysis and threshold analysis (HM3->HM2 the only revision).
- Hypertrophy: «All RT prescriptions may comparably promote muscle hypertrophy, and the influence of load was less apparent» — sets/volume, not load, rank the top prescriptions. Training to failure did not explain the hypertrophic response in these (mostly untrained) participants (network meta-regression for ‘failure’ didn’t improve fit); Currier flags failure «may… be increasingly important for trained individuals».
Why this is a real distinction, not a fake tension (parameter-table check): the two SMD columns are different outcomes measured on different instruments (1RM force vs cross-sectional area / lean mass), so load matters for one and not the other is not a contradiction to reconcile — it is two curves. Filing it as a tension would compare non-commensurable quantities.
Myth correction — soreness (DOMS) does not track growth, and neither does “muscle confusion” [2026-09-25, belief-harvest WS-026]
“No pain, no gain” — the belief that delayed-onset muscle soreness (DOMS) marks an effective session — runs against the decomposition above. Hypertrophy is volume-driven, not damage-driven: Currier found training to failure did not explain the hypertrophic response in these (mostly untrained) participants, and load was largely irrelevant to muscle size (§decomposition) — so the muscle damage that produces soreness is not the growth signal. DOMS is a muscle-damage / unfamiliarity signal that attenuates as the muscle adapts to a movement (the repeated-bout effect) — so soreness fades with training precisely while gains continue, and its absence is not evidence of a wasted session. Relatedly, “muscle confusion” — constantly varying exercises to keep muscles guessing — has no separate lever here: the drivers that separated prescriptions were load (strength) and volume (size), not novelty; variety earns its place through adherence and joint-health, not a distinct growth mechanism. Chase volume for size and load for strength; do not use soreness or novelty as the dose signal.
Minimal effective dose — a floor, not a located knee
- «There was a 95% probability that RT with at least two sets or two sessions per week increased strength … and training with at least two sets and two sessions per week resulted in hypertrophy.»
- Lower-CrI floor across prescriptions: «at least a moderate (SMD>0.47)» strength and «small (SMD>0.16)» mass increase — but not for every prescription, as the next sentence says: «Such certainty is not possible for all prescriptions, though, because the 95% CrI crossed zero for two RTx for strength (HS1 and LS1) and three RTx for hypertrophy (HM1, HS2 and HS3), meaning these prescriptions might increase, not change or decrease muscle strength and size.» Currier attributes this to sparse nodes rather than ineffectiveness: «These strength (HS1 and LS1) and hypertrophy (HM1, HS2 and HS3) nodes included <60 participants and contrib- uted little direct evidence (figure 2).» So the floor holds for the well-populated prescriptions; for once-weekly single-set (HS1/LS1) and a few hypertrophy nodes the network estimate is too imprecise to show it (within-study, strength rose vs control/baseline in those trials). (corrected 2026-10-06: universal lower-CrI floor -> floor excludes the five zero-crossing nodes; Currier chunk 02) (Currier et al., 2023)
The curve’s shape is under-determined here, and that is a G-gap, not a plateau. Currier coded load / sets / frequency categorically (H/L, M/S, 1/2/3), not continuously, so a true knee within load or volume cannot be located from this analysis — Currier says so and calls for continuous, model-based dose-response NMA. So the honest reading is: a large step from zero, then a broad flat region across prescriptions (hypertrophy) or a modest load-gradient (strength) — with the minimum effective dose a region (~2 sets, ~2×/week) rather than a point. (inferred from Currier et al., 2023) — the categorical-coding limit and its shape-under-determined consequence are Currier’s stated limitation read against the wiki’s dose-response vocabulary (a threshold quoted from categorical data marks the edge of the evidence, not a feature of the curve). This is another instance of The Underivable Optimum — a broad flat region (its Route 1) plus categorical/measurement under-determination (its Route 3), the same under-identification the protein ~1.62 g/kg knee carries: hold the RT dose numbers loosely too, and read the ~2 sets / ~2x per week as a floor, not an optimum.
Once-weekly training — what the frequency arm does and does not show (the popular belief once-weekly RT matches twice-weekly for strength, esp. in older adults). Currier’s strength summary is a disjunction — «at least two sets or two sessions per week» — so once-weekly multiset training is inside it, while the hypertrophy summary is a conjunction («at least two sets and two sessions»). In the network, once-weekly multiset strength beat control with CrIs excluding zero (HM1 1.54, 0.81 to 2.30; LM1 1.07, 0.47 to 1.67 — Table 2); once-weekly single-set (HS1, LS1) crossed zero on sparse nodes. Between prescriptions, the only strength comparisons that separated (9/66) were HM2/HM3 versus a lower-load RTx — none is a once-vs-twice-weekly contrast at matched load and sets — so the network detects no frequency difference, on imprecise estimates: not a demonstrated equivalence. (Currier et al., 2023) Neither summary is a necessity claim (it says what was shown to work, not that less fails) and neither speaks to function or quality of life; and the older-adult once-weekly case is not separately estimated here (age showed no modifying effect in meta-regression, on sparse data). Remaining G-gap: an older-adult once-weekly-vs-twice-weekly strength/function head-to-head SR is not held. (corrected 2026-10-06: once-weekly is not covered by the held estimate and is arguably run against by it -> once-weekly multiset is covered for strength; once-weekly single-set is imprecise; Currier chunk 02) (inferred from Currier et al., 2023)
The retention corollary — the floor holds through a layoff. The MED above is an acquisition floor; there is also a retention side. In older adults, functional-capacity gains from a completed resistance or multicomponent program persist at a medium effect over never-trained controls through a 1-2 month training cessation (ES = 0.88 [0.47-1.29]), and the residual does not depend on modality or intensity — but its decay time-course and any reduced maintenance dose are unestimated. So a forced break does not reset the prescription clock; see Detraining and Residual Effects in Older Adults for the retention half of the dose question and the maintenance-floor gap. (inferred from Buendía‐Romero et al., 2025)
Effect modifiers — mostly absent (route-b is quiet here)
Network meta-regression found no obvious modifying effect on relative RTx effects from age, training status, proportion female, duration, volitional fatigue, relative weekly volume load, measurement tool / region, or publication year — data-sparse nodes reduced precision. So there is little positive evidence that the relative ranking of prescriptions changes by stratum: personalization of the protocol rests on preference and constraint (Route e), not on demonstrated effect modification (Route b). Baseline (untrained) status still governs absolute gain — the big step is largest for the untrained. (Currier et al., 2023)
The load gauge — how many reps sit at a given %1RM (operationalizing the 80% cut)
The strength finding above is stated in %1RM, but almost nobody outside a lab knows their 1RM; the field proxy is the rep count. Nuzzo meta-regressed 952 repetitions-to-failure tests (7289 people, 452 groups, 269 studies, 1961-2023), admitting only studies where «the 1RM was tested rather than estimated» and the reps test was done fresh, and modelled both the mean reps and the between-person SD of reps at each %1RM (natural cubic spline on log means; linear on log SDs). Analysed set: 425 groups, 898 tests, 6970 people, traditional concentric-eccentric reps only. (Nuzzo et al., 2023)
Main-model table, mean reps to failure [95% CI] and between-person SD [95% CI] — fitted to all groups (bench press 42% and leg press 14% of groups); the authors recommend it for exercises other than those two, while noting «minimal data» for commonly prescribed lifts such as overhead press and pulldown. Values from the Fig. 2 table (p. 308):
| %1RM | Mean reps [95% CI] | Between-person SD [95% CI] |
|---|---|---|
| 90 | 4.94 [4.35-5.61] | 1.90 [1.69-2.14] |
| 85 | 7.15 [6.69-7.65] | 2.19 [1.97-2.42] |
| 80 | 9.75 [9.32-10.20] | 2.51 [2.29-2.75] |
| 75 | 12.37 [11.87-12.88] | 2.88 [2.65-3.13] |
| 70 | 14.80 [14.23-15.40] | 3.31 [3.06-3.58] |
| 65 | 17.11 [16.36-17.90] | 3.80 [3.51-4.12] |
| 60 | 19.53 [18.52-20.59] | 4.37 [4.01-4.76] |
(Nuzzo et al., 2023) Tabulated range 15-95% 1RM, the range of the data; precision is tight from the heavy end down to 65%: «The precision of estimates for both means and SDs are tight up to 65% 1RM range to ~ 1 repeti- tion.» Below that, data thin and intervals widen. (Nuzzo et al., 2023)
- The between-person spread is the new information, and it grows as the load falls. «For example, at 80% 1RM, the estimate for the SD about the point estimate is 2.51 repetitions, whereas at 60% 1RM the estimate is 4.36 repetitions.» The authors do not claim the SD is all true person-to-person difference — «Thus, although large SDs could be due to between-individual heterogeneity in repetitions completed, a mathematical phenomenon, or heteroskedas- tic measurement errors, this information is still practically useful because it illustrates the amount of variance that can be expected.» (Nuzzo et al., 2023)
- Exercise is the one moderator that clearly shows up — leg press vs bench press. «For example, at 80% and 70% 1RM, the estimated number of repetitions in the leg press were 13.1 [95% CI 9.8–17.5] and 19.0 [95% CI 14.2–25.5], respectively, whereas for the bench» press the figures were 8.8 [95% CI 7.7–10.1] and 14.1 [95% CI 12.4–16.1]; separate tables are given for those two lifts, the main table for everything else. The source states the resolved contrast as a load range — fewer mean bench reps «up to ~ 50% 1RM» (and lower bench SDs up to ~35%) — without saying from which end that range runs; at 80% the two exercises’ 95% CIs overlap (bench 8.82 [7.72-10.08], leg press 13.05 [9.78-17.41], Figs 3-4 tables). (Nuzzo et al., 2023)
- Sex, age and training status: no resolved moderation (a null on imprecise contrasts, not a demonstrated equivalence). «The impact of most of the moderators was uncertain based on the precision of estimates for the contrasts.» Almost all contrast-ratio intervals included 1, and the authors conclude «Analysis of moderators suggested little influences of sex, age, or training status on the REPS ~ %1RM relationship, thus the general main model REPS ~ %1RM table can be applied to all individuals and to all exercises other than the bench press and leg press.» (Nuzzo et al., 2023)
- Who the table describes. Groups were 66% male, 97% healthy, 92% under 59 (median group age 23), 60% resistance-trained; «Most data from the REPS ~ %1RM relationship have been collected on healthy individuals who are aged 20–40 years.» Rep duration (1.4-6.0 s, reported by 46% of studies) tended to lower reps when slower, but almost all contrast intervals included 1. Eccentric-only testing was ~1% of the data, so no eccentric table exists. (Nuzzo et al., 2023)
- Design caveats, stated by the authors. «Our search strategy did not follow standard guidelines for meta-analy- ses.» (a mixed personal-knowledge + keyword + snowball search; data and code are public on OSF). Conflict of interest: «Pinto were pre- viously employed at Vitruvian, a company that designs and sells re- sistance exercise equipment.» (Nuzzo and Pinto). (Nuzzo et al., 2023)
- Not addressed: safety or feasibility of 1RM testing in untrained people. The paper is a descriptive load-reps model; it reports no injury or adverse-event data for 1RM or reps-to-failure testing [searched: safe/risk/novice/familiar = 0 across chunks 01-02; injur = 1 hit, a reference title only]. The rep gauge replaces a 1RM test with a reps-to-failure test — still a maximal-effort set — so it relocates the safety question for an untrained adult rather than answering it. (inferred from Nuzzo et al., 2023)
Parameter table — the coding rule vs the measured relationship. Currier’s NMA converted rep maxima to %1RM by %1RM = 100 − 2.5 × RM (quoted under Limits); the textbook table Nuzzo set out to update also puts 8 reps at 80% (its Table 1: 90% = 4, 80% = 8, 75% = 10, 70% = 11, 65% = 15 reps).
| Parameter | Currier coding rule | Nuzzo main model | Same quantity? |
|---|---|---|---|
| What it maps | a reported RM (reps to failure at a fixed load) -> %1RM | %1RM -> mean reps to failure (tested 1RM) | Yes, inverted direction — both are reps-to-failure at a load relative to a measured 1RM; Nuzzo’s is a population mean curve, so inverting it gives the load at which the mean rep count equals that RM (approximately the average person), not any individual’s |
| Reps at 90% | 4 (rule) | 4.94 [4.35-5.61] | yes |
| Reps at 80% | 8 (rule: 8RM = 80%) | 9.75 [9.32-10.20] | yes |
| Reps at 75% | 10 | 12.37 [11.87-12.88] | yes |
| Reps at 70% | 12 | 14.80 [14.23-15.40] | yes |
| Between-person spread | none (a deterministic rule) | SD 2.51 reps at 80%, 3.31 at 70% | rule has no counterpart |
| Exercise-specific? | no | yes — leg press higher, bench lower (separate tables) | — |
(Currier et al., 2023) (the rule) (Nuzzo et al., 2023) (the measured values)
What the comparison shows (the wiki’s reading, type-F). The 2.5%-per-rep rule is close at the heavy end and drifts at lighter loads, always in one direction: at a given %1RM the average person does more reps than the rule says (80%: ~9.75 vs 8; 70%: ~14.8 vs 12), equivalently a given rep max sits at a heavier %1RM than the rule assigns. Interpolating the main-model table, an 8-rep max sits near ~83% 1RM, a 10-rep max near ~80%, 12 reps near ~76% and 15 reps near ~70% (rule: 80 / 75 / 70 / 62.5%). So for the strength cut the rule is conservative: a person who trains at their 8RM is on average a few points above 80%, not at it. But the between-person SD (~2.5 reps at 80%, roughly ~5 %1RM points given the local slope of ~0.5 rep per %1RM) means a rep count places the average person, not this person: for roughly two-thirds of people (if the SD is mostly true between-person spread rather than measurement error), an 8RM lies between the high-70s and high-80s %1RM. On point estimates the leg press runs higher — its 80% load averages ~13 reps, putting an 8-rep leg-press max near ~90% 1RM — but its interval at 80% spans ~10-17 reps (and 90% leg press = 8.69 [3.71-20.39]), so a leg-press 8RM could sit anywhere from ~80% to >90%. (Currier et al., 2023; inferred from Nuzzo et al., 2023)
- Does the drift bias Currier’s own result? Direction, not size. For the average person the rule codes an RM load a few points too light, so misclassification runs one way — an arm truly at or just above 80% (a 9-10RM) coded L, never a truly light arm coded H. That dilutes the H-L contrast toward the null: if anything Currier’s load effect is understated. How many arms sat near the boundary is not reported in the held chunks. (inferred from Currier et al., 2023; Nuzzo et al., 2023)
Equipment modality (free-weights vs machines): specific for the test, equivalent for the outcome
The other much-debated RT dial — barbell/dumbbell vs pin-loaded machine — resolves the same way the prescription dials do: it is second-order once you are training. Haugen pooled 13 studies that directly compared the two modalities (n=1016, adults 18-60, free of chronic disease, >=6 weeks; non-athletes; TESTEX quality fair-to-good, none excellent). This is an independent line (a Norwegian group, no lineage overlap with Currier/McMaster) on a different sub-question, so it extends the big-rock finding from protocol to equipment.
The apparent modality advantage is test-specific, not a real difference in adaptation. Trained with free-weights, you gain more free-weight-tested strength (SMD -0.210, 95% CI -0.391 to -0.029, p=0.023); trained on machines, you tend to gain more machine-tested strength (SMD 0.291, 95% CI -0.017 to 0.600, p=0.064). Haugen reads this as the specificity (SAID) principle: «The principle of specificity applies, which states that you should choose the exercise you want to be stronger in.» You get better at the movement you actually train — a testing artifact of the transfer, not a superior modality. (Haugen et al., 2023)
On the outcome itself, no modality difference survives. Comparing each group in the mode it trained, or on a neutral test, the between-modality effect is null for dynamic strength (SMD 0.084, 95% CI -0.106 to 0.273, p=0.387), isometric/neutral strength (SMD -0.079, 95% CI -0.432 to 0.273, p=0.660), countermovement jump (SMD -0.209, 95% CI -0.597 to 0.179, p=0.290), and hypertrophy (SMD -0.055, 95% CI -0.397 to 0.287, p=0.751). Both modalities produced large within-group gains (strength SMD ~0.92 vs ~0.97; hypertrophy ~0.25 vs ~0.21). Haugen’s conclusion: «strength changes are specific to the training modality, and the choice between free-weights and machines are down to individual preferences and goals.» (Haugen et al., 2023)
- One partial exception (small n): a direct-strength sub-analysis favored machines for upper-body strength, with no difference lower-body — the hypertrophy arm rests on only 5-6 studies, so read this as an unsettled wrinkle, not a machine advantage. For hypertrophy Haugen defers to preference: «When the goal is to maximise muscle hypertrophy individual preferences should dictate the choice, but we speculate that a combination could yield the best benefit» (regional muscle growth may differ by modality even when total growth matches). (Haugen et al., 2023)
- Injury risk does not break the tie either. «Summed up, it is uncertain if there are different injury risks between free-weight and machine-based strength training.» The higher free-weight injury counts are mostly weights dropped on people (cross-sectional ED data), not a movement-execution hazard, and no longitudinal trial establishes causation; ACSM’s view that «machines may be safer to use than free-weights based on skill requirements» is a skill-requirement argument, not an outcome finding. (Haugen et al., 2023)
Decision: choose equipment on preference, goal-specificity, and access — not on an expected strength or hypertrophy advantage, because none exists on the outcome. The one place specificity does bind is Route-e (constraint), not Route-b (effect modification): a competitor tested in a named lift (powerlifter, weightlifter) must train that lift; a recreational trainee optimizing size or general strength is free to pick either or mix. The surrogate boundary below applies unchanged — these are 1RM / muscle-size / jump gains, not a health outcome.
The big-rock pattern, now across two dials [inferred from @currier2023; @haugen2023freeweight]. Currier found 91% of between-prescription comparisons contained zero; Haugen finds the between-equipment comparison null on the outcome. These are two different sub-questions — protocol and equipment — and the same Layer-1 pattern holds across both: once someone is training, the sub-choices — protocol and equipment alike — are second-order to the train-vs-not-train decision. This generalizes the ranking beyond load/sets/frequency; it is not two studies corroborating one finding (they measure different quantities, so it is not a type-E robustness claim), and it does not raise the page’s confidence (Haugen’s own evidence base is thin and «tentative», and the health outcome is untouched).
G-gap — bodyweight / calisthenics is unevidenced head-to-head. Haugen restricted the comparison to
free-weights (barbell/dumbbell) vs fixed-path machines, and explicitly excluded cable, freemotion,
pneumatic, and variable-resistance equipment. No gold SR/MA directly comparing bodyweight /
calisthenics training against loaded RT for strength or hypertrophy is held — a named zero from the
research pass (verified absent at acquisition), not a settled equivalence. A gold head-to-head SR would
close it; until one lands, hold any bodyweight-vs-loaded claim at confidence: low.
(inferred from Haugen et al., 2023)
Sex is not a meaningful effect modifier — one prescription for both (route-b null)
The most-asked stratification of RT — should women train differently? — has a direct answer, and it is a well-bounded null on relative gains. Roberts pooled male-vs-female RELATIVE adaptation to the SAME protocol across 50 studies (ages 18-50, >=5 weeks; supplements/HRT excluded), splitting by outcome. The effect size is male-group ES minus female-group ES, so a negative value favors females; every ES is a within-group, baseline-normalized (relative) change, NOT absolute kg/cm:
| Outcome (relative gain) | k (outcomes / studies) | Pooled ES (male-minus-female) | 95% CI | I2 | Verdict |
|---|---|---|---|---|---|
| Hypertrophy | 12 / 10 | 0.07 | -0.09 to 0.23 | 0 | No sex difference (tight null) |
| Upper-body strength | 19 / 17 | -0.60 | -0.93 to -0.26 | 72.1 | Favors females (moderate) |
| Lower-body strength | 23 / 23 | -0.21 | -0.54 to 0.12 | 74.7 | No sex difference |
| (Roberts et al., 2020) |
- Headline: «males and females adapted to RT with similar effect sizes for hypertrophy and lower-body strength, but females had a larger effect size for relative upper-body strength.» (Roberts et al., 2020) The one non-null runs toward women, not away — so nothing here motivates a lighter/different female prescription; if anything untrained women gain upper-body strength at least as fast relative to baseline.
- The absolute-vs-relative trap, named by the source: «Although it is true that absolute hypertrophy and gains in strength are larger in males after RT, it seems that relative increases in both muscularity and lower-body strength are similar between the sexes, and relative gains in upper-body strength may be larger in females.» (Roberts et al., 2020) Men gain more absolute size/strength (higher baseline mass, more upper-body androgen receptors); the relative response curve is the same. Reading the absolute gap as a different response is the error — it is a different starting point (a Route-a baseline fact, not a Route-b effect modification).
- The upper-body female signal is plausibly an artifact, not biology — the authors flag it: high heterogeneity (I2 ~72%) unreduced by covariates, mostly untrained short trials, and «This could cause a ceiling effect for motor skills that may explain differences in upper-body strength because the studies were conducted in mostly untrained subjects.» (Roberts et al., 2020) Men are often more familiar with upper-body movements (e.g. bench press), leaving women more short-run motor-learning headroom. So even the one non-null may not survive longer training or trained populations — it does not upgrade to a prescription difference.
- Lifting makes women bulky — refuted on BOTH axes. Relative hypertrophy is equal, not greater, in women (ES 0.07; CI -0.09 to 0.23; I2 = 0 — an unusually clean null), and absolute muscle gain is smaller in women. The same training does not build more muscle on a woman than on a man; the testosterone gap that was once invoked to predict blunted female hypertrophy did not produce it. (Roberts et al., 2020)
Convergence with Currier’s covariate null — but a different parameter. Currier’s prescription NMA found «no» modifying effect of proportion female on the relative ranking of PRESCRIPTIONS (a meta-regression covariate); Roberts is a direct male-vs-female contrast of the response itself. These are different quantities (a between-RTx-ranking covariate vs a pooled within-protocol sex-difference ES), so this is two independent designs/teams converging on the same question — sex is not a route-(b) modifier here — not the same measurement re-pooled. (Currier et al., 2023; inferred from Roberts et al., 2020)
Decision: prescribe RT the same for both sexes (load for strength, volume for size — as above). Any sex-tailoring rests on preference/constraint (Route e) or absolute-baseline scaling (Route a), NOT on demonstrated effect modification. The source is explicit that the direct trials do not settle a prescription difference either way: «it is currently difficult to know if exercise prescription should be different between sexes.» (Roberts et al., 2020) The surrogate boundary below applies unchanged — these are 1RM/size gains, not a health outcome.
Limits (Roberts): mostly untrained subjects, short trials, high strength heterogeneity unexplained
by measured covariates, and no formal risk-of-bias scoring (the primary trials cannot blind
exercise, so the authors judged standard quality scales unusable) — a high-tier MA resting on
unblindable primaries, same design ceiling as Currier.
(Roberts et al., 2020)
The surrogate → outcome boundary — the load-bearing honesty
Strength and muscle size are surrogates (Surrogate Outcomes), and Currier is unusually explicit that the health-outcome link is not in this analysis: «We do not know how these RTx affect relevant health outcomes» and «The effects on health outcomes of various RTx remain largely unknown.» (Currier et al., 2023) The transmission differs by surrogate, and the ranking of surrogates matters more than the ranking of prescriptions:
- Strength → moderate transmission. Strength (esp. grip) and muscle-strengthening activity track lower mortality observationally -> Grip Strength and Mortality, Muscle-Strengthening Activity and Mortality (any MSA vs none: all-cause mortality RR 0.85; MSA + aerobic RR 0.60). But that is activity/strength predicting death, not this NMA’s 1RM gains reducing death — no RCT closes it.
- Strength → injury reduction is the one patient-important outcome established at CAUSAL (RCT-MA) grade — see Exercise Interventions and Sports Injury Prevention (Lauersen: strength-training RR 0.315, injuries cut to <1/3; the standout intervention, stretching null). This is the closest the resistance-training case gets to a hard endpoint on interventional rather than observational evidence. Exposure-identity caveat: Lauersen’s strength arms are eccentric / sport-specific injury-prevention protocols in young athletes, NOT the hypertrophy-oriented general RT this page prescribes — so the transfer to a recreational/older gym trainer is a transportability gap, not a settled property of “RT”. (inferred from Lauersen et al., 2013)
- The tissue layer this page is silent on — tendon, which adapts on its OWN schedule. This page’s prescriptions target muscle; Tendon Adaptation to Mechanical Loading shows the connective-tissue parallel and a divergence. Parallel: load magnitude is the driver of tendon stiffening too, and contraction type is irrelevant — the same load is the lever logic. Divergence: tendon adapts slower than muscle, so a dose that builds muscle fast can outrun tendon conditioning — the rate mismatch that motivates gradual load progression, mattering most for the deconditioned/obese novice this page’s dose might otherwise overload. Tendon outcomes there are surrogates, not injury. (the wiki’s cross-page synthesis; the tendon figures and their source are on the linked page)
- Hypertrophy → weakest transmission. Low muscle mass predicts mortality (Low Muscle Mass and Mortality), but that raising size via training lowers mortality is unproven — hypertrophy is largely a surrogate for a surrogate. Do not read the 0.66 hypertrophy SMD as a health effect. (inferred from Currier et al., 2023)
- The closest-to-patient-important signal here is physical function in older adults: LM2/LM3/HM3 improved mobility and gait speed, HM3 improved balance (few studies, ≥55y) — function is on the outcome menu directly, but the evidence is thin. (Currier et al., 2023)
- Why the older-adult stratum needs this dose at all — the mechanism is Anabolic Resistance. Aging blunts the muscle-protein-synthesis response to a given protein dose; resistance training is the non-nutritional lever that partially restores that sensitivity, so the training stimulus and the higher per-meal protein target (Protein Intake for Older Adults: ~1.2 g/kg/day for the active older adult vs ~1.0 sedentary) are complementary, not substitutes — RT raises protein needs, and protein is what the restored response acts on.
Decision relevance
- The big rock is doing any resistance training at all. Prescription choice is a second-order refinement — 91% of between-protocol comparisons were indistinguishable. Someone not currently training should not wait for the right scheme.
- If the goal is strength: bias toward heavier loads (>80% 1RM), multiple sets, ~2–3×/week. Load is the one variable that reliably buys more. Without a 1RM test, a load you can lift to failure no more than ~8-9 times sits at or above ~80% for the average person on most exercises studied (Nuzzo’s table; fewer on the bench, more on the leg press) — counted on a fresh first set (later sets fatigue), from data on mostly healthy 20-40-year-olds; a population gauge with a ~2.5-rep between-person SD, not a personal conversion.
- If the goal is size/hypertrophy: chase volume (sets); load is flexible — lighter loads work if the sets are there, and training to failure is not required (untrained). Pairs with the protein lever (~1.6 g/kg/day) on Protein and Resistance Training for Muscle and Strength — the other input to the same adaptation.
- Minimal effective dose: roughly 2 sets, 2×/week captures most of the available strength and size gain; more is a modestly steeper strength curve, not a different category.
- Equipment is a preference choice, not a lever. Free-weights and machines produce equivalent strength and hypertrophy on the outcome (Haugen: all direct-comparison CIs span zero); pick by preference, access, and goal-specificity. Specificity binds only for a competitor tested in a named lift — train that lift — not for general strength or size.
- Adherence and preference win the ties. With prescriptions near-equivalent, the sustainable protocol beats the theoretically-optimal-but-abandoned one — Currier frames the whole result as licensing choice.
- Do not oversell the endpoint. These are surrogate gains; the mortality/function payoff is inferred from separate observational lines, strongest for strength, weakest for pure hypertrophy.
Limits
- Surrogates only — 1RM and muscle size; no mortality/disease endpoint (Currier states this outright).
- “≥80% 1RM” is the trials’ label; reps are the field proxy, by a coding rule only. Currier’s eligible strength outcomes include a direct 1RM test, and loads that trials reported as rep maxima were converted: «RT loads reported as repetition maximum (RM) were converted to a percentage of one-repetition maximum (%1RM) with the equation: %1RM=100−(RM(2.5)).» (Currier et al., 2023). On that rule an 8-rep maximum sits at 80%, the heavy/light cut, and 12-15 reps at about 70-62%. So a person with no measured 1RM can aim at the cut by rep count. The equation is an analyst’s coding convention taken from an unheld reference, not a validated conversion. Measured against Nuzzo’s meta-regression of reps-to-failure at tested %1RM (see The load gauge above), the rule is close at the heavy end and under-counts reps at lighter loads, so an 8RM sits on average near ~83% (conservative for the cut); the between-person SD (~2.5 reps at 80%) and the leg-press/bench difference mean a rep count places the average person, not the individual. Sex, age and training status were not resolved as moderators (imprecise contrasts), and the data are mostly healthy 20-40-year-olds. Still open: whether a maximal test is safe or unsafe for an untrained adult — Nuzzo does not address it, so the fabric holds no data either way. (Currier et al., 2023; inferred from Nuzzo et al., 2023)
- Categorical coding (H/L, M/S, 1/2/3) — cannot locate a continuous knee; periodized programmes, rest intervals, tempo, time-under-tension excluded/under-reported.
- Unblindable primary trials — moderate–high risk of bias (strength 22% high; hypertrophy 18% high); gold design, but the underlying RCTs cannot double-blind exercise.
- Healthy adults only — athletes, comorbidities, frail excluded; older-adult function data sparse. For the person who cannot load heavy (older, comorbid joints/bones, rehab), this dial does not apply; the route-(c)/(e) workaround is Blood Flow Restriction Training, which reaches heavy-load-comparable hypertrophy at light load (though below-heavy-load strength) — and lands the same load-dependence shape from an independent team, corroborating the strength-is-load-driven / hypertrophy-is-not decomposition above.
- Single source, shared lineage — not independent of ACSM 2026 or Morton 2018 (Phillips/McMaster);
confidence: lowuntil an independent line lands. - Equipment facet is thin and tentative (Haugen) — 13 studies, hypertrophy on only 5-6, none rated excellent quality, and the authors call the evidence «tentative»; unblindable primaries, same design ceiling. Its team is disjoint from Currier’s but its evidence base is shallow, so it broadens the big-rock principle without upgrading the page’s confidence.