HCD.10:11 - SoTA-Echoing

The working question is: How should a practitioner select variation, spacing, interleaving, retrieval, support, and challenge changes across practice episodes for one target action and evidence need?

The selected best-known line is task-, similarity-, support-, and horizon-sensitive rather than recipe-based. Carpenter, Pan, and Butler’s 2022 review (COG-01) supports spaced encounters and effortful retrieval as credible candidates for durable access under source-specific conditions while selecting no universal gap, repetition count, threshold, or curriculum. Brunmair and Richter’s 2019 meta-analysis (COG-03) shows that interleaving depends materially on between- and within-category similarity, complexity, and material type; some source families showed no benefit or favored blocking. HCD.10 adapts these contributions into the separate timing, retrieval, and task-selection steps in §§4.2–4.5.

The serious ordinary alternative is a bundled desirable-difficulty recipe: space all topics on one calendar, require closed-book retrieval, interleave them, remove help, and raise challenge together. It is easy to standardize, but a failure cannot distinguish timing, access, task selection, support, load, or capability. HCD.10 retains each candidate move while rejecting their automatic combination and any universal difficulty ladder.

Czyż, Wójcik, and Solarská’s 2024 motor-learning meta-analysis (COG-04) reports a heterogeneous random-versus-blocked transfer result whose applied-setting estimate was not significant; laboratory adult evidence does not establish a general professional schedule. Paas and van Merriënboer’s 2020 review (COG-05) supports changing worked examples and guidance with prior knowledge and task complexity, not fixed fading or “less load means more learning.” These limits produce the stable-practice entry and conditional progression rules in §§4.5–4.7.

Tullis, Goldstone, and Hanson’s 2015 laboratory motor experiment (COG-11) makes assisted practice and later unassisted performance different observations and offers a bounded support-scheduling cue, not a universal fading rule. Bastani and colleagues’ 2025 school-mathematics trial (AI-04) supplies counterexample evidence that higher AI-assisted practice performance can coexist with no improvement or worse later unassisted performance and that guardrail design matters; it establishes no adult-workplace effect or general AI ban. These qualifications motivate the separate support state and the narrow no-AI observation in §§4.4–4.7 and 5.2.

The selected line costs more than copying a schedule, but it preserves the target discrimination, legitimate tool use, interpretable failure, safe backoff, and later evidence boundary. Reopen it when the target action, prior attempts, task similarity, intended horizon, interval evidence, source or support configuration, risk, E.23.CAE differential, or serious alternative changes, or when representative use shows that fewer distinctions preserve the same decision value.