How much will I improve?
Give the range for your training status, say people scatter around it mostly from test noise, and name the retest that will replace it.
The question it answers
An athlete asks how much they will improve, by when, what time they can run, whether they have stalled or are "a non-responder", and whether their VO2max or W/kg is good. Short answer: the base can give a group range for their training status and dose, the noise any reading carries, and for road marathons a prediction with a known error; it cannot give one person's gain, a date, or a W/kg norm, and it says so.
What this framework owns and what it leaves
| Owns (decided here) | Leaves to |
|---|---|
| The size of a forecast gain by training status, and how to say its spread | Which lever a block pulls, the block thesis, the consistency gate and the review → SYNTH-42 |
| What counts as a real change: the noise table, the first-retest effect | Which test, run the same way each time → SYNTH-53; which device number to trust → SYNTH-91; VO2max from power → SYNTH-78 |
| "Am I a non-responder?" and the plateau call | Responder labels and athlete types → SYNTH-13; VO2max interval formats → SYNTH-33; the first 12 weeks → SYNTH-75 |
| Race-time prediction from races and its error | Marathon build and pacing → SYNTH-64; long-course hours → SYNTH-57; short time-trial pacing → SYNTH-48; the taper's gain → SYNTH-34 |
| VO2max norms and cycling calibre ladders; W/kg questions | Masters decline → SYNTH-35; women → SYNTH-37; hours and intensity mix → SYNTH-26; block order and length → SYNTH-90 |
| Whether a fitness-fatigue chart forecasts | Load, ramp and deload → SYNTH-29; detraining → SYNTH-36 |
Default (what to prescribe first)
Not from a study; the base's own practice, built on the entries below:
Every forecast has three parts: the range for the athlete's training status and planned dose from a named source, one clause on why individuals scatter around it (mostly measurement noise, not a label), and the test and date that will replace the group number with the athlete's own.
Pick the group figure by training status. Untrained or recreational: VO2max about +5 mL/kg/min against no exercise (REF-1276, A-). Trained: about 2% in a time trial from an interval block (REF-1275, A-), and for VO2max only an added effect (+2.5 mL/kg/min for adding work above the maximal metabolic steady state over training kept below it, REF-989, A-).
Near the ceiling: a few per cent over seasons (REF-1279, REF-1280, transfer). Never extend a beginner's first-block gain to later blocks (REF-529, an inference in that entry).
Read every change against its noise (the noise table below): a change no larger than about the test's typical error is no change, and a first-ever retest flatters by about 1% (REF-1278, B+).
Call a plateau only when a block that was done gives two comparable readings inside the noise. First rule out a block that was not done, illness, under-fuelling and overreaching (SYNTH-42, SYNTH-29); then change one lever — intensity, amount or stimulus — and name it (direction from adults who were untrained or not described as trained: REF-1277, REF-995, REF-996).
A date to a target only from the athlete's own slope across at least two comparable readings, as a window, with the group figures as a ceiling check; say that no study follows amateurs' FTP or W/kg over years.
What counts as a real change — the noise table
"Noise" is one SD of measurement and day-to-day variation (a typical error) unless the cell says otherwise; the difference between two readings carries about 1.4 times that (the square root of two, the base's arithmetic). "Read a change as real when" is practice built on the noise column.
| Outcome | Its noise in the base | Read a change as real when | Support and level |
|---|---|---|---|
| Laboratory performance test (incremental peak power, a constant-power test of 1 min to 3 h, lactate-threshold power) | Typical error as a CV: peak power ~0.9%, constant-power tests 0.9-2.0%, lactate-threshold power ~1.5%; 1.3 times larger in non-athletes, 1.4 times in female than male non-athletes, up to 1.6 times for ~1-s and ~1-h tests against 1-min tests | It repeats, or clearly exceeds the band: with a CV of 1.5%, a 2% move on one retest is about one noise SD of the difference, compatible with no change (the entry's arithmetic) | REF-1278 B+ (101 reliability studies of healthy adults, abstract, 2001) |
| The first retest of a series | The second trial was 1.2% better than the first and later trials only 0.2% apart; the first pair's CV was 1.3 times the later pairs' | One familiarisation test before the baseline that counts; the first ~1% of a first-ever retest is learning, not training | REF-1278 B+ |
| Output at a fixed heart rate (heart-rate-clamped submaximal stages) | Typical error 1.3-4.4% in the reliable stages | A move inside about that band is no change (SYNTH-42); read with RPE (SYNTH-41); untested across ordinary sessions and for running pace | REF-752 B (17 well-trained male cyclists, one standardised test) |
| Field best power within a season (training and race files) | SEM < 5% (ICC > 0.90) pooling training and racing; SEM < 8% within one context | A best that moves less than the SEM is noise; compare bests from the same context | REF-479 B (18 professionals) |
| Season-best power from one season to the next | CV 3.2-5.9% for season-best power from 10 s to 30 min (lowest beyond 1 min); critical power 3.3% (95% CI 3.0-3.7%); the authors' expected variation about 6% for efforts of 1 min or less and 3% for longer ones, with changes beyond 12% and 8% infrequent — by the entry's arithmetic a two-CV threshold is crossed by about one season-to-season change in six | For a professional, a season-best critical power within about 3% of last season's is normal variation; an amateur's season bests depend on how often they went all-out, so their band is probably wider (inference) | REF-1279 B (61 male professionals, median 4 seasons, abstract) |
| Laboratory VO2max change after one block | Most of the observed spread of change between people is measurement error; baseline VO2max reproduced (ICC 0.98) while the individual change did not (ICC 0.19) | One block is one noisy point; a pattern over 2-3 blocks on the same benchmark (SYNTH-42, practice) | REF-1274 A- (24 controlled trials, abstract); REF-498 B+ (42 untrained middle-aged adults, two identical blocks) |
| VO2max estimated from a watch or from power | 95% limits of agreement (about two SDs, not one) of about ±10 mL/kg/min for a watch estimate from runs; about ±12 for the best power-based equation in men | Not a measure of change: a drop or rise of a few points sits inside the estimate's error (SYNTH-91, SYNTH-78) | REF-1253 A- (14 validation studies); REF-1017 B+ (580 competitive cyclists) |
| The athlete's own repeats | Where the same effort has been repeated under the same conditions, the athlete's own typical error | It replaces the group figures above | Not from a study; the base's own practice |
Decision table
"(practice)" in a cell marks a rule that is not from a study; the base's own practice.
| Question | What the assistant may say (range, inputs, caveat) | What it must not say | Support and level | How strong |
|---|---|---|---|---|
| "How much will my VO2max rise?" | Untrained or recreational: about +5 mL/kg/min over a programme against no exercise (continuous training +4.9, 95% confidence limits ±1.4; intervals +5.5, ±1.2; typical study means 3.6-6.8), about 12% from a start near 41 (the entry's arithmetic), more in the less fit and over longer programmes, probably somewhat less given publication bias. Trained: the base has no gain over no exercise; adding work above the maximal metabolic steady state gave 2.5 mL/kg/min more than training kept below it. Inputs: training status, a measured baseline, planned intervals and weeks. Caveat: individual results scatter around the mean, mostly from noise; the retest decides | A personal number as a promise; the untrained mean for a trained athlete; +2.5 as a trained athlete's gain over doing nothing | REF-1276 A- (28 controlled trials, 723 adults aged 18-45, no funding); REF-989 A- (14 short studies, trained athletes); REF-749 A (inactive adults: interval training reached 80% of its gain at ~11 min a session, flat by ~23); REF-465 A (lower initial fitness, larger percentage gains); REF-1274 A- | Moderate for untrained adults; for trained athletes only an added effect, weak as a forecast |
| "How much will my FTP, time trial or 10 km improve this block?" | About 2% in a time trial on average after an interval block for trained people, up to about 6% at lower training status (before-after changes, mostly young men; the abstract does not say whether the figure is time or power). In one matched trial, three interval groups of highly trained cyclists and triathletes gained 4.4-5.8% in a 40-km time trial in 4 weeks against −0.9 to +1.1% in controls — a good case, not the expectation. If the 2% is power, it sits near one retest's noise (inference; the unit is not given), so the first retest may not settle it. For FTP itself no entry gives a gain by training status; carry the time-trial figure as an estimate (inference) | "Your FTP will rise 10% in 6 weeks"; that more intervals will add more (the authors: dose beyond a minimum may not add); a first block's gain as the next block's | REF-1275 A- (29 studies, 491 adults, abstract); REF-999 B (38 highly trained cyclists and triathletes, matched, not randomised); REF-1278 B+; REF-989 A- (no shown time-trial effect, underpowered) | Moderate for the direction; the size for trained amateurs is transfer |
| "How much more over the years?" / "When will I reach 4 W/kg or a sub-X?" | Gains shrink near the ceiling: world top-100 track-and-field athletes peaked at 25-27 (marathoners and male throwers about 28-29), and the running, hurdling and jumping groups gained less over the five years before the peak than the throwers' 7.0%; professional cyclists' season-best critical power varies about 3% a season. A date only from the athlete's own slope across at least two comparable readings, as a window, with these figures as a ceiling check (practice); say there is no study of amateurs over years | A date from a group number; a first-block slope extended over years; a figure for runners' own 5-year gain (the abstract's bracket is ambiguous); elite curves as an amateur's timetable | REF-1280 B+ (14,937 athletes, abstract); REF-1279 B; REF-988 B+ (109 Tier 4-5 athletes: submaximal variables and peak test speed or power improved in most reports, VO2max inconsistently); REF-1003 B (delta efficiency 23.61 → 26.97% over five seasons with no significant VO2max rise, 12 world-class cyclists); REF-529 A | Weak; elite data only; the date rule is practice |
| "How many hours a week for a 250-W FTP, a sub-3:30 marathon or a sub-5 70.3?" | No entry maps hours to an FTP or a triathlon time. In running, more training goes with faster marathons between people: cohorts that ran more distance, more runs and more hours each finished faster (R² 0.38-0.81 across the measures); 50 rather than 30 miles a week went with a 25-32-min faster marathon between runners; the fastest marathoners ran more than three times the volume of slower ones, the extra almost entirely easy running. In one Ironman build, more total and more easy training went with a faster race. The guides' hour bands are practice (SYNTH-26, SYNTH-64, SYNTH-57) | "X hours buys Y watts or Z minutes"; a between-runner association as one athlete's forecast | REF-1004 A- (85 studies, between-cohort associations); REF-1282 B+ (self-reported); REF-197 B+ (cross-sectional training logs); REF-784 B (9 triathletes, correlations) | Association only; nothing for cycling or triathlon times |
| "Am I a non-responder?" / "My partner improves faster" | One block cannot show it: in controlled trials most of the spread in VO2max change was measurement error, with no strong evidence of true between-person differences in trainability within one programme, and people repeating an identical block barely reproduced their own change. HERITAGE's range, from little or nothing to more than 1.0 L/min, is observed change in a single-arm cohort. Raising the amount or the intensity removed apparent non-response in untrained adults | "You are a low responder"; "VO2max is not trainable" (that is not what the meta-analysis says); "your genes cap you"; a large SD of change with no control arm as proof of responders | REF-1274 A- (abstract); REF-498 B+ (ICC 0.19); REF-429 B+ (single-arm, no control arm); REF-1230 A- (four controlled studies: exercisers' SD of change 1.9 against controls' 1.3 mL/kg/min); REF-995 B+, REF-996 B+, REF-1277 B+ (untrained, sedentary or not described as trained adults); SYNTH-13 | Moderate in untrained adults; unmeasured in trained athletes |
| "My retest is up (or down) 2% — is it real?" | Set it against the test's noise: laboratory tests carry a typical error of about 0.9-2.0% of power, so a 2% move on one retest is about one noise SD; a first-ever retest runs about 1.2% better from practice; field tests are probably noisier (inference). Same protocol, conditions and device (SYNTH-53, SYNTH-91); real when it repeats or clearly exceeds the band (practice) | A verdict on a change inside the band; a first-ever test against the second read as training; a comparison across devices or protocols | REF-1278 B+ (2001, abstract); REF-752 B; REF-479 B | Moderate for laboratory tests |
| "Have I plateaued, and what do I change?" | A plateau is two comparable readings inside the noise after a block that was done (practice). First check that it was done, then illness, sleep and life, under-fuelling and overreaching (SYNTH-42, SYNTH-29). Then change ONE lever and name it — intensity at the same amount, amount at the same intensity, or the stimulus. In sedentary adults, higher intensity at equal energy left 0% non-responders against 17.6%, double the amount at low intensity cut 38.5% to 17.6%, and low-intensity non-response stopped falling after 8-16 weeks; a fixed load stopped raising VO2max after about 3 weeks in probably untrained people | A plateau from one flat reading or from a block that was not done; "you have hit your genetic ceiling"; the untrained timetable as a trained athlete's | REF-1277 B+ (121 completers, abstract, non-response undefined there); REF-995 B+; REF-996 B+; REF-1218 B (n = 9, uncontrolled); REF-997 B+ (response to one mode did not predict another); REF-1278 B+ | Direction moderate in untrained adults, transfer to trained; the plateau rule is practice |
| "My season-best power hardly changed" | For a professional, a season-best critical power within about 3% of last season's is normal; bests of a minute or less vary about 6%. An amateur's bests depend on how often they went all-out at that duration, so compare like with like: a race or a planned maximal effort (SYNTH-42, practice) | A 2-3% season-best change as progress or decline; a season of maximal racing against a season without | REF-1279 B (61 professionals, abstract); REF-479 B | Moderate for professionals; transfer to amateurs |
| "How fast will it come?" | In probably untrained people most of a fixed load's VO2max gain came within about 3 weeks (half-time about 10-11 days) and rose again only when the work rose; interval blocks of 4 weeks or less gave smaller VO2max gains (g 0.16-0.25) than longer ones (0.55-0.79), length confounded with training status; in highly trained cyclists a 40-km gain was measurable after 4 weeks of added intervals. Read at the block's review date (3-6 weeks, practice, SYNTH-42), not weekly | A weekly rate of gain; the 3-week plateau as a trained athlete's timetable (SYNTH-90) | REF-1218 B; REF-1243 A- (subgroup); REF-1277 B+; REF-999 B | Thin; no time-course entry in trained athletes |
| "What marathon can I run off my half or 10 km?" | From a flat half run about six weeks earlier: marathon minutes ≈ 2.28 × half minutes − 6 (men) or − 12 (women), off by 5.7% on average and by 4.1-4.8%, about 10-11 min, near a 4-h finish (the entry's arithmetic), poorer beyond 4:30; Daniels' VDOT was closer under 3 h. Riegel's formula is fair up to the half but put the marathon 10 min or more too fast for about half of recreational runners; adding weekly mileage cut that to about a quarter — from a 1:40 half, Riegel gives 3:30 and the mileage model about 3:44 on 30 miles a week, 3:35 on 50 (the entry's arithmetic). Critical speed from 16 weeks of training logs predicted finish time with an average error of 7.67%. Quote a window and pace from the slower estimate (SYNTH-64, practice) | A plain Riegel marathon as the prediction; a time to the minute; the same window for a hilly or hot race | REF-1283 B+ (7,663 Valencia runners, official times, one flat race pair, no funding); REF-1282 B+ (2,303 self-reported runners, 156 in the validation comparison); REF-1245 B+ | Moderate; flat road marathons only |
| "Predict my gran fondo, road race, time trial, 70.3 or Ironman" | The base holds no validated prediction for cycling events or triathlon; the nearest is a 16-km road time trial predicted from laboratory critical power and W′ in 12 competitive men (27.5 vs 26.7 min, r = 0.89, group level only). Plan the effort from power, heart rate and fuel (SYNTH-48, SYNTH-57, SYNTH-10) and use the athlete's own past results on the same course as the anchor (practice) | A finish time from a calculator; a running formula applied to a ride | REF-040 B (n = 12); otherwise a gap | Gap |
| "Is my VO2max good for my age?" | A laboratory VO2max can be ranked against the FRIEND reference for its decade, sex and test mode — cycle values run lower than treadmill ones, men higher at every age, athletes high by construction; the base holds the registry's description, not its percentile tables, so quote a rank only when the athlete's laboratory report gives one and name the reference (practice). A VO2max estimated from power or a watch gets its error first (about ±10 mL/kg/min); a percentile from an estimate can move several bands | A percentile or a "good/excellent" band from memory; a treadmill norm for a cycle test; the registry's 4.0-4.3 mL/kg/min fall per decade as the athlete's own loss | REF-1284 B+ (22,379 US laboratory tests, abstract); REF-1253 A-; REF-1017 B+; SYNTH-78; SYNTH-35 | Moderate for the reference; its tables are not in the base |
| "Where does 4 W/kg sit?" / "What level am I?" | The base has no peer-reviewed amateur W/kg norm. The two cycling ladders (five levels from untrained to professional; a separate female ladder) are built from research groups' laboratory VO2max and graded-test peak power, their numeric bands are not in the base, and an FTP or 20-min power is not that number (time to exhaustion at 95% of 20-min power was a median 35 min in recreational riders). Place the athlete by tier from training and results (REF-529) and compare the 4 W/kg with their own history and their goal's demands (practice) | A level, percentile or cut-off for a W/kg; a woman on the male ladder; professional power profiles as an amateur's target | REF-1285 B+; REF-1286 B+ (both abstracts, bands not read); REF-747 B; REF-529 A | Gap for norms |
| "Do women, or older athletes, gain less?" | Given the same interval protocol, women and men gained alike (VO2max g = 0.57 each, 19 trials); with sex-matched endurance training, untrained men gained 1.95 mL/kg/min (0.76 to 3.15) more in 8 small studies. Give both sexes the same relative forecast (practice) and compare in mL/kg/min or per cent — absolute gains in L/min are smaller in smaller people. Age: sedentary groups over 45 improved with intervals too (subgroup judged not robust by its authors); former elite runners who kept training hard lost about 6% of VO2max a decade against about 15% after stopping; a fall per decade across a population is not a person's rate (SYNTH-35) | A lower forecast for a woman; a loss per decade quoted at an individual; "too old to improve" | REF-1243 A-; REF-1281 A- (abstract); REF-595 B; REF-1138 B+; REF-1284 B+ | Moderate that sex is a small modifier; trained women and masters athletes untested |
| "Can my fitness, fatigue or form chart (CTL, ATL, TSB) predict my race?" | It records how load changed (SYNTH-29). In two small elite datasets a fitted model's fatigue term fitted past performance better but did not predict later performance better (cross-validated RMSE +0.001, p = 0.57; −8.8 × 10⁻⁵, p = 0.41), and fitness and fatigue could not be separated; judge form from a repeated benchmark, field bests and how the athlete feels (practice) | "Your fitness is X, so you will ride Y"; a race-day form forecast from the form number; a fixed-constant chart called validated | REF-1287 B (7 and 14 elite short-track skaters, sprint tests, coach-scored load); REF-108 B (taper simulations on a fitted model: fit, not prediction); REF-1227 B+ (load ratio in runners) | Moderate that the model is no validated forecast; untested in endurance athletes |
Key Findings — what the evidence shows
Most of the spread between people in one block is measurement noise. A meta-analysis restricted to 24 aerobic-training studies with a non-exercising control found that most of the variation in observed VO2max change scores was measurement error, that a single study is too small to estimate the true individual-response SD, and no strong evidence that it is above zero; the authors conclude there is not strong evidence for between-person differences in VO2max trainability across single interventions (REF-1274, A-, abstract; populations not described, mostly untrained adults by inference).
When 42 untrained adults repeated an identical 8-week block, baseline VO2max reproduced (ICC 0.98) and the individual change did not (ICC 0.19; 15-min power 0.20) (REF-498, B+). HERITAGE's range of change in 481 sedentary adults is single-arm, with no control arm (REF-429, B+).
Against these, in four small controlled studies the exercisers' SD of change exceeded the controls' (1.9 vs 1.3 mL/kg/min, Bayes factor 12.4), which those authors read as true differences (REF-1230, A-; previously sedentary adults).
Mean gains are large in the untrained and small in the trained. Against no-exercise controls, healthy untrained adults aged 18-45 (start 40.8 mL/kg/min) gained 4.9 (±1.4) mL/kg/min with continuous endurance training and 5.5 (±1.2) with intervals, intervals adding 1.2 (±0.9) over continuous training; gains were larger in the less fit and over longer programmes, and Egger's test indicated publication bias in all three analyses (REF-1276, A-, author manuscript without its tables).
Pooled before-after time-trial gains from interval training were about 2% in trained people and up to about 6% at lower training status, with a dose-response for HIIT sessions, weeks and total work but none for sprint intervals (REF-1275, A-, abstract; 400 of 491 men).
In trained athletes adding work above the maximal metabolic steady state gave 2.5 mL/kg/min more VO2peak than training kept below it, with no shown effect on peak power or time trial (REF-989, A-). Lower initial fitness meant larger percentage gains in VO2max, mitochondria and capillaries across 5,973 participants (REF-465, A).
A test's noise may be the size of a trained athlete's expected gain. Typical errors of 0.9-2.0% for laboratory power tests, larger in non-athletes, and a 1.2% gain from the first to the second trial (REF-1278, B+, 2001, abstract) sit next to a ~2% expected time-trial gain (REF-1275) if both are in the same unit: its abstract does not say whether the 2% is time or power, and in cycling a time change is smaller than the power change behind it (inference in that entry); output at a fixed heart rate carries 1.3-4.4% (REF-752, B); professionals' field bests repeat within an SEM under 5% in a season (REF-479, B) and their season-best critical power varies by a CV of 3.3% between seasons (REF-1279, B, abstract).
Gains shrink near the ceiling and over years. World top-100 track-and-field athletes peaked at 25-27 (marathoners and male throwers about 28-29); throwers improved 7.0% over the five years before the peak and the running, hurdling and jumping groups less — whether their own improvement was 3.4-5.2% or about 1.8-3.6% cannot be read from the abstract; the top 10 improved 1.0-1.8% more than those ranked 11-100 except in throws, and women more than men except in sprints (REF-1280, B+, abstract).
In longitudinal reports of Tier 4-5 athletes submaximal variables kept improving while VO2max did not consistently (REF-988, B+); world-class cyclists' delta efficiency rose over five seasons with VO2max flat (REF-1003, B).
A fixed load stops paying; more intensity or more amount restarts it — shown in untrained adults. VO2max rose for about 3 weeks at constant work rates, then held until the work rose (REF-1218, B, n = 9).
In 121 sedentary, abdominally obese adults, non-responders after 24 weeks were 38.5% (low amount, low intensity), 17.6% (high amount, low intensity) and 0% (high amount at 75% of VO2peak); at low intensity the count stopped falling after 8-16 weeks (REF-1277, B+, completers only, abstract) — a pattern a larger mean gain alone would produce, with no control-arm spread to calibrate it.
Energy-matched higher intensity took responders from 16 of 31 to 13 of 15 (REF-995, B+), and two extra sessions a week removed non-response in maximal power in an uncontrolled second period (REF-996, B+; only non-responders were retrained, so regression to the mean is not excluded — inference); response to endurance training did not predict response to sprint intervals (REF-997, B+).
Sex: no difference with identical intervals, men ahead with sex-matched endurance training (the protocol or the analysis may explain the gap); age: continued hard training slowed the loss but did not stop it. With identical interval protocols women and men gained alike (VO2max g = 0.57 each; threshold power g = 0.38 each) (REF-1243, A-); with sex-matched endurance training untrained men gained 191 mL/min (99 to 283) and 1.95 mL/kg/min (0.76 to 3.15) more, I² = 0% (REF-1281, A-, abstract).
Former elite runners who kept training hard lost about 6% of VO2max a decade against about 15% after stopping (REF-595, B); in a general population followed from 16 to 63 the yearly loss rose from under 1% to over 2% and the spread between people grew 3- to 25-fold (REF-1138, B+).
Race prediction from races has a known error, for road marathons only. Riegel's formula was calibrated for the 10 km and the half but put the marathon 10 min or more too fast for about half of 156 validation runners and more than 20 min too fast for nearly a quarter; models adding typical weekly mileage cut the mean squared error from 380.7 to 227.6 (one prior race) and 208.3 (two), and the share 10 min or more too fast to about 25% (REF-1282, B+, self-reported survey of 2,303 runners whose marathons were about 40 min faster than New York finishers').
In 7,663 runners of a flat Valencia half and marathon six weeks apart, half time and sex explained 85% of the variance with a mean absolute error of 5.67% (the abstract says 5.9%), against 7.92% for VDOT, which was closer under 3 h (REF-1283, B+, official times).
Critical speed from 16 weeks of training logs predicted finish time with an average error of 7.67% in about 25,000 runners (REF-1245, B+). For cycling only a 16-km time trial from laboratory critical power, at group level (REF-040, B).
Norms describe populations, not athletes, and there is no W/kg norm. The FRIEND registry ranks 22,379 laboratory tests of apparently healthy US adults by decade, sex and mode; its 50th percentile fell 13.5% (4.0 mL/kg/min) a decade on the treadmill and 16.4% (4.3) on the cycle, cross-sectionally (REF-1284, B+, abstract, tables not read).
The cycling ladders class research groups by laboratory VO2max and graded-test peak power, with women on their own ladder, whose trained and well-trained rungs did not differ significantly as labelled (REF-1285, REF-1286, B+, abstracts, bands not read). A 20-min power is not an hour's (REF-747, B).
A fitness-fatigue model fits the past and does not forecast better. In 7 and 14 elite short-track skaters, adding the fatigue term lowered the training-set error (RMSE −0.017, p < 0.001) but not the cross-validated error (+0.001, p = 0.57; −8.8 × 10⁻⁵, p = 0.41); fitness and fatigue parameters substituted for each other (1-4 of 10 Markov chains converged per athlete in one dataset); a fitness-only model with a 30-day half-life fixed in advance had the lowest prediction errors, but no such difference passed the authors' Bonferroni threshold (REF-1287, B, full text).
It does not show how much one trained amateur will gain, whether true individual response exists in trained athletes, how amateurs' FTP or W/kg develops over years, how many hours buy a given FTP or triathlon time, where a W/kg sits among amateurs, or how well any cycling or triathlon finish can be predicted; those rows are practice or gaps (Evidence gaps).
Conflicts and how they are resolved
Practical Application — how the assistant says it
Evidence gaps
Limitations
How this entry was checked and changed
(the direction — forecast from the group figure for the athlete's training status, read every change against its test's noise, never classify a person from one block, carry each prediction's error, treat norms and load charts as descriptions — rests on A- meta-analyses, most read from their abstracts REF-1274 REF-1275 REF-1281 or from a manuscript without its tables REF-1276, A- meta-analyses already in the base REF-989 REF-1243 REF-1230, A meta-analyses in inactive or mixed adults REF-749 REF-465, B+ measurement and observational studies REF-1278 REF-1280 REF-1282 REF-1283 REF-1245 REF-1284 REF-1285 REF-1286 REF-498 REF-429, a B+ trial in sedentary adults REF-1277 and B studies REF-1279 REF-752 REF-479 REF-1218 REF-1287 REF-999. Above C+ because those directions are what the A-/B+ evidence says; below B because the gain evidence is in untrained young adults or, for trained people, uncontrolled before-after changes read from an abstract, the multi-season data are professional or world-class, true individual response in trained athletes is unmeasured, and the default's actionable rules — the plateau call, a date forecast from the athlete's own slope, the same relative forecast for both sexes, judging a lever over 2-3 blocks — are practice)