The Banister fitness-fatigue model behind training-load charts can fit performance data, but its individual parameters are very uncertain, so app charts are rough summaries rather than predictions
Interpretation
The model (Banister 1975, described by Kontro 2026) fitted a season of performance in 9 elite swimmers (R2 = 0.79) but key time constants had wide confidence intervals (38 days, 17-59; 19 days, 6-32) and correlated parameters (Hellard 2006). In 6 untrained people a variable-fatigue version fitted better and implied an inverted-U dose-response (Busso 2003). A model-designed exponential taper beat a step taper in triathletes' cycle ergometry but not significantly in a 5 km run (Banister 1999). Load inputs are themselves limited: single-number scores ignore training type (Kontro 2026) and session variability (Passfield 2017). Graded limited: few, small studies in swimmers, untrained people and small triathlete groups; no study tests app-default charts in age-group triathletes.
Practical implication
Use the chart to see training trends, plan gradual build-ups and check that fatigue falls in a taper, alongside subjective ratings, which tracked load better than objective measures in a review of 56 studies (Saw 2016). Don't treat the fitness line as measured fitness or chase a steeper line; fixed app defaults may not fit you, and transfer to three-sport triathlon training is inference.
Sources
6 sources, independently verified.
Methods/theory paper. Traces the impulse-response model to Banister and colleagues (1975), reduced to fitness and fatigue components that decay exponentially with their own time constants; notes popular platforms compute fitness (chronic training load) as an exponentially weighted moving average of daily load. Argues single-metric load inputs wrongly assume adaptation is independent of training type, and proposes a three-dimensional model based on critical power, W' and maximal power. Full text read via Europe PMC.
The Banister model related training load to performance in every swimmer (R2 = 0.79 ± 0.13) and estimation seemed stable, but 95% confidence intervals of key parameters were wide (e.g. 38 days, 17-59; 19 days, 6-32; 19 days, 7-35; 43 days, 25-61) and some parameters were highly correlated, making their interpretation worthless.
Training improved 5-min all-out power by 30 ± 7%. A nonlinear model letting fatigue vary with training load fitted better than earlier models (standard error 6.47 ± 0.71 W vs 9.20-10.31 W) and implied an inverted-U relationship between daily training amount and performance. Very small, untrained sample.
Modelling then field testing in triathletes: an exponential taper improved cycle ergometry performance more than a step-reduction taper, and a fast exponential taper (time constant 4 days) more than a slow one (8 days); 5 km run results went the same way but were not significant.
Mobile power meters provide valid, reliable field measurements and are useful for monitoring training and race demands, but simple summary methods cannot reflect a session's variable power distribution or its metabolic stress, and current methods of quantifying training load are inadequate.
Subjective and objective measures generally did not correlate; subjective measures reflected acute and chronic training load with greater sensitivity and consistency. Subjective well-being was impaired by acute load increases and chronic training and improved with an acute decrease in load. Mixed sports and levels.