Solution: Ten Years of Experience, or One Year Repeated Ten Times?

MCQ answer: (1) — B has been engaging in deliberate practice, A has not

Equal volume with divergent outcomes is the classic signature Ericsson’s research was built to explain: raw years of experience and total repetition volume correlate only weakly with expertise once you look closely, and the reliable predictor is whether that time included deliberate practice’s specific ingredients — a challenging goal, focused attention, immediate corrective feedback, and repeated refinement targeting weaknesses — versus routine, comfortable-difficulty repetition. Option (0)’s “natural talent” explanation is exactly the folk intuition this research pushes back against — while raw aptitude plays some role, large, well-documented performance gaps between equally-experienced practitioners are far better explained by how that experience was structured than by assumed innate differences. Option (2) (pure volume) is explicitly what the puzzle rules out by stipulating equal reading volume. Option (3) dodges the question the research was specifically designed to answer with a real, replicated mechanism.

Part 2 — Why routine practice actually plateaus, not just slows

Lesson 18’s power law describes improvement when every trial still provides some learning signal. Routine, comfortable-difficulty repetition breaks that assumption: once a skill component is executed successfully and automatically (a familiar chunk, recognized and applied without conscious effort — Lesson 19’s chunking, running smoothly), further repetitions of that same, already-mastered component provide essentially no new corrective signal to learn from — there’s no error to notice, no feedback pointing at a gap, nothing for the underlying representation to update against. The trial still happens, but it’s closer to executing a stored production rule (Lesson 3’s condition→action firing) than to building or refining one.

Genuine improvement requires trials where the current skill genuinely gets tested against something at or beyond its current edge, with feedback specific enough to reveal exactly what went wrong. Comfortable, routine repetition systematically avoids exactly that zone — which is why it produces an apparent hard plateau rather than Lesson 18’s ever-slower-but-still-positive curve: it isn’t that the rate of learning per trial is decaying toward zero along a smooth power-law curve, it’s that most routine trials are contributing close to zero learning signal individually, because nothing about them is challenging the current skill level or supplying corrective information.

Part 3 — A deliberate-practice regimen: code review skill

Deliberate version: pick a genuinely difficult category of bug you currently miss sometimes (e.g., subtle race conditions) — that’s the challenging goal, just past current ability. Review real pull requests from that category with full, undistracted attention, writing down a specific prediction of every issue you spot before checking against a more senior reviewer’s actual findings or the bug’s eventual real-world outcome — that’s the immediate, specific feedback. Where your prediction missed something the senior reviewer caught, deliberately study why you missed it (what pattern were you not recognizing?) before moving to the next PR — that’s repetition with targeted refinement, focused specifically on the gap just identified, not just moving on to the next task.

Routine contrast: reviewing PRs as they come in, at a comfortable pace, approving or requesting changes based on familiar checks, with no systematic comparison against a stronger reviewer’s judgment and no deliberate focus on a specific known weak spot — genuinely useful for keeping the codebase healthy, but structurally unlikely to push the reviewer’s own skill past wherever it currently plateaus, for exactly the reason given in Part 2.

Part 4 — Deliberate-practice analogue, and the plateau failure mode, in agent training

A close analogue: reinforcement learning or fine-tuning regimes that specifically target an agent’s known failure modes with tasks calibrated just past its current success rate, paired with clear, specific reward/correction signal tied to exactly what went wrong — structurally close to deliberate practice’s four ingredients (challenging difficulty, focused signal, immediate feedback, targeted refinement). Contrast that with simply running an agent on a large volume of tasks it already handles comfortably, with vague or absent feedback about why any given output was better or worse — closer to routine repetition.

Expected plateau failure mode: an agent trained mostly on high-volume-but-routine experience (comfortable tasks, weak or generic feedback) would be expected to perform reliably on familiar-shaped tasks while showing little continued improvement on its genuine edge cases or known weak spots over time — exactly matching Radiologist A’s flat 10-year accuracy curve despite continuing high case volume: lots of executions of already-mastered patterns, very little targeted correction of the specific gaps that would actually move performance forward.

The pattern

Rule: raw hours or repetitions predict expertise only weakly — what predicts it is whether practice included deliberate practice’s specific structural ingredients. And even genuine expertise, however it was built, transfers far less to other domains than intuition suggests — consistent with Lesson 19’s finding that expertise is chunking of specific meaningful patterns, not a general cognitive upgrade that should carry over anywhere.

Where this goes: Stage 4 closes with one more piece — spacing and interleaving, the scheduling principles (not just how you practice, but when and in what order) that this whole puzzle path’s own design is built on.