This is a deliberately simple, population-agnostic dose-response and forgetting model, not a prediction for any child. Let K be the fraction of a target skill that is secure (0 to 1). Support raises K with diminishing returns; without practice K fades exponentially (Ebbinghaus-style).
session of h hours: K -> 1 - (1 - K) e^(-a h) a = gain rate (1/hour)
gap of d days: K -> K e^(-f d) f = forgetting rate (1/day)
ODE behind it: dK/dt = a u(t) (1 - K) - f K
plateau (1 session/week, right after a session):
K* = (1 - e^(-a h)) / (1 - e^(-a h) e^(-7 f))What it shows. The same weekly minutes split into more, shorter sessions give a higher average level, because the gain per minute is largest when K is low and forgetting acts on the gaps. In the calendar the session days show up as bright cells followed by a fade; the timeline shows the saw-tooth of gain and forgetting. Simplifications, stated honestly. The three profiles are illustrative parameter pairs invented for the demonstration, not measurements and not diagnostic categories; real children differ widely and an actual individual education plan rests on professional assessment, the child's strengths, interests and family input. Forgetting inside a session is neglected (the RK4 check measures the error this causes). Motivation, difficulty, sleep and the quality of teaching are all outside the model.
Checks. (1) Closed-form session/gap map vs RK4 of the full ODE; (2) weekly plateau closed form vs 300 weeks of iteration; (3) with no forgetting the gain depends only on total hours; (4) new in this 2D view: the mean of the seven weekday-row means of the calendar equals the overall mean, and the numerical forgetting half-life equals ln2/f. The wood backdrop is a Blender render of the 3D page's desk.