Figures look learned because they're visual — the layout is familiar, the arrows make sense while you look, and the whole thing feels understood. But that familiarity lives in the printed figure, not in your head. Cover it, and the "understanding" was really just the ability to read a diagram that's sitting in front of you — recognition of the artwork, not retrieval of the mechanism.

Studying a figure means producing it. Cover the original, redraw the boxes and arrows from memory, then narrate each arrow in one sentence — the causal step it represents, not the line it draws. Then break one arrow on purpose: delete a step, and say what fails downstream. If you can do that, you know the figure as a mechanism. If you can't, you know it as a picture.

Why staring at the figure fails

The published figure is a polished object — clean boxes, labeled arrows, a layout designed to be readable. Staring at it trains reading, which is the exam's lowest bar. The exam doesn't ask "does this figure look familiar" — it asks "what happens if this step fails" or "draw the pathway" or "which arrow represents the rate-limiting step." Those are production questions, and staring at the artwork doesn't train production.

The redraw is the retrieval — pulling the structure out of memory instead of recognizing it on the page. The arrow narration is the self-explanation — the mechanism behind each step, not the picture of it. The break-one-arrow test is the transfer — the concept survives a perturbation, which is what the exam's "what if" questions actually probe. Redrawing is generation; explaining arrows is self-explanation of the mechanism, not the artwork — and neither happens while the figure is uncovered.

The figure workflow

Cover the figure completely. Not squint at it, not glance and look away — cover it. The redraw has to come from memory, which means the printed version can't be there filling in what you don't have. A partial cover gives you partial retrieval, which gives you partial learning.

Redraw the skeleton first. Boxes and arrows before labels — the structure is the mechanism, the labels are decoration on top. If you can place the components and connect them correctly, the figure is learned at the level that matters. Label-filling is the easy part once the skeleton is right.

Narrate every arrow in one sentence. "A causes B" isn't enough — "A phosphorylates B, which activates it" is. The arrow is a causal step; the sentence is the mechanism. If you can't narrate an arrow, that's a gap the redraw didn't catch — the box was placed but the step wasn't understood.

Break one arrow on purpose. Delete the interneuron, block the enzyme, cut the feedback loop — and say what fails downstream. The break tests whether the figure is a mechanism in your head (predictable consequences) or a layout (memorized connections). If the break produces a coherent downstream story, the figure is learned.

Misses become re-tests. A box you forgot, an arrow you couldn't narrate, a break you couldn't predict — those are next session's stems, dated 1/3/7 days out. The figure isn't done when it's drawn once; it's done when the redraws keep working across the spacing.

A 40-minute session on the clock

The weak answer is a symptom guess — "slower" is the kind of vague prediction that sounds right without understanding the mechanism. The strong answer traces the break: the interneuron is the spinal shortcut, so losing it kills the rapid withdrawal while leaving slower paths intact. That's the difference between recognizing the figure and knowing it — one sees a labeled part, the other sees a functional role.

Why the redraw beats the look

A figure encodes two things: the layout (what it looks like) and the mechanism (what it means). Looking trains the layout — the visual pattern you recognize on sight. The exam tests the mechanism — the causal structure you have to produce, explain, or perturb. The two feel identical while the figure is visible, because the layout cues the mechanism. Cover it, and the difference is exposed: the layout was memorized, the mechanism wasn't built.

The redraw is the test that separates them. If you can produce the structure unlabeled and narrate the arrows as causal steps, the mechanism is there. If you can't, the "learning" was just the layout — and the exam's "what happens if" question will find the gap the figure's familiarity was hiding.