Can a prompt replace Fitbod?

Build a workout generator whose rules are visible and arguable.

Data model: an exercise has a name, an equipment requirement, a primary muscle group and a list of secondary ones, with a contribution weight for each. A set is exercise, reps, weight, RPE if you record it, and a timestamp. That is all the state there is.

Generation runs on a scoring pass you can read in one screen. For each muscle group compute the volume of the last seven days, weighting secondary contributions at their listed fraction. Compute hours since it was last worked. Score each group by how far its weekly volume sits below the target you configured, gated by a minimum recovery window per group — larger groups get longer. Then pick exercises covering the top-scoring groups, filtered by the equipment available today.

Every generated session must be explainable. Print, beside each exercise, why it was chosen: which group it serves, that group's current weekly volume against target, and hours since it was last trained. An opaque generator is indistinguishable from a random one, and after two weeks you stop trusting it.

Progression is a separate, simple rule: if the last session at a given exercise hit every set at the top of its rep range, raise the load by the smallest increment your gym's plates allow; if it missed the bottom of the range twice running, lower it. Put the increments in config, per exercise, because a 2.5 kg jump on a lateral raise is not the same decision as on a squat.

Add a per-muscle weekly volume chart, since the whole model rests on that number and you should be able to see it.

Do not build form checking from video, and do not add a deload algorithm without a way to override it — a program that insists you rest on the one week you feel good is a program you delete.