🌱 Lesson-design controls illustrative model
The science

What this is. A teaching toy for lesson designers, built from two well-known ideas. It is an illustrative model with made-up parameters: it does not diagnose, rate or describe any person or group, and the "parameter sets" are just three example settings, not categories of learners.

Forgetting. Retrievability decays as R(t) = exp(−t/S) (Ebbinghaus-style), where S is the memory strength in days. A review at time t succeeds with probability R; success raises strength, S ← S (1 + g q (0.4 + 1 − R)) (reviews help more when a little is forgotten), while a lapse restarts at Sinit.

Cognitive load. Working memory handles only a few chunks at once. Total load is intrinsic + extraneous (1 − 0.7 s), where supports s (plain language, chunking, worked examples, pictures plus words) remove extraneous clutter. Encoding quality is q = 1/(1 + e(load − capacity)/0.8), which is 1/2 exactly when load equals capacity.

Exact expectation. The coloured curves are exact averages over all success/lapse outcome paths of up to 10 reviews; the thin lines are single stochastic learners.

Honest simplifications. One item at a time, no interference between items, no sleep, motivation or context effects, constant parameters, a single made-up logistic for load. Real learning is richer and varies a lot between people; use this only to reason about spacing and clutter, and rely on classroom evidence and specialist advice for real decisions.

Independent checks (run on load):

running…

Schedule comparison

Encoding quality q-
Total load vs capacity-
Initial strength S-
Selected: mean recall-
Selected: recall at end-
Reviews (workload)-
Today (garden)-
-
Drag the Day slider or press Play. Each plant is one learner-item: its height is how well it is still recalled. Educational toy model, no diagnostic meaning.
Made with 🟠 Blender