Methodology
The phase.fitness generator does not improvise. Before it writes a phase it reads a set of training-science documents, and the numbers in them become bands the program has to respect. Here are all of them, published as the generator sees them, with their references.
How the knowledge base constrains the generator
Before writing a phase, phase.fitness selects the documents that match the goal, the sport and the experience level, and passes their principles to the model as context. The weekly set ranges, rep ranges, intensities and RPE targets in those documents are also extracted into prescription bands that the finished plan is checked against — so the numbers are a constraint, not a suggestion the model may or may not have read. The references are stripped from the prompt, because the model is not allowed to cite studies, and published here instead.
Progressive overload, fat loss, general fitness, later life, pain, warm-ups: the rules that apply to everyone.
How many sets, how often, and when to back off.
Linear and block periodization, and RPE-based autoregulation.
When to substitute an exercise and how to vary it without losing the thread.
One document per discipline: what the sport demands, the phases, the in-season rule, the injuries that end seasons.
Before writing a phase, the phase.fitness generator selects the documents relevant to the goal, the sport and the experience level, and receives their principles as context. Weekly set ranges, rep ranges, intensity bands and RPE targets are extracted into prescription bands that the plan is checked against. The generator is told not to cite studies; the references stay in these pages.
They are working documents written for the phase.fitness generator, not journal articles. Thirty-one of the thirty-four list their sources in a References section; several of those have been verified against the published papers, and each document shows its last update date. Where a document has no references yet, the page says so.
Because a training app that says it is science-based should show the science. Publishing the documents phase.fitness actually reads is the only honest version of that claim: what the AI is constrained by is what you can read and check.