Can brief, learner-authored manipulable retention checks embedded inside an interactive visual (where students design a small variable-manipulation challenge for a peer) yield additional durable conceptual gains over an equivalent amount of time spent on system-authored prediction and self-explanation prompts, and under what scaffolding conditions does authoring remain productive rather than becoming an unstructured design task?

Answer

Yes, brief learner‑authored manipulable retention checks can yield additional durable conceptual gains over an equivalent amount of time on purely system‑authored prediction and self‑explanation prompts, but only when authoring is tightly scaffolded as a constrained transformation of well‑understood diagnostic contrasts. Authoring remains productive—rather than devolving into an open‑ended design task—when (a) variable and outcome choices are restricted to a small, vetted set; (b) templates enforce prediction-before-authoring and brief rationales; and (c) checks are short, peer‑targeted challenges that must themselves pass a quick system quality gate. Without these constraints, authoring time mostly adds task complexity and illusion-of-understanding, not durable learning.

More concretely:

• Additional gains are most likely for intermediate learners who already pass basic delayed, out-of-context retention checks and can use authoring to reorganize and articulate their model; for fragile novices, equivalent time on system-authored prompts usually outperforms authoring.
• Scaffolding should make authoring a “fill‑in and justify a contrast” activity, not free-form task design: e.g., choose from 2–3 key variables, select one of a few contrast templates ("hold A, vary B", "double C, predict D"), specify target values, and write a one‑sentence explanation the peer must reconstruct.
• A lightweight system gate (e.g., quick automatic checks for internal consistency plus a brief self‑prediction of what a competent peer should do) keeps authoring in the productive struggle band and prevents low‑diagnostic or misleading challenges from dominating.
• In this tightly scaffolded form, small blocks of learner‑authored checks interleaved with system‑authored prediction/self‑explanation prompts function as high‑value generation plus “teaching to a peer,” which tends to add durable conceptual gains and far transfer beyond what equivalently timed system prompts alone achieve.
• When scaffolds are weak—too many degrees of freedom, no templates, no quality gate—authoring rapidly becomes an unstructured design task that increases cognitive load, encourages surface‑feature challenges, and yields equal or worse durable learning than spending all time on well‑designed system‑authored prompts.