Much current work assumes that the main benefit of interactive visual explanations comes from within-visual manipulation of variables; if we instead treat the visual as a stable external memory and require learners to do most manipulation as offloaded mental simulation (e.g., predicting unseen configurations or intermediate states that are only later revealed in the visual), are there concept types or learner profiles for which this “visual-as-check, mind-as-simulator” design yields stronger durable learning and far transfer than traditional designs that emphasize continuous on-screen manipulation?

Answer

Yes, for some concepts and learners a “visual-as-check, mind-as-simulator” design will likely beat continuous manipulation on durable learning and far transfer, but not universally.

Most promising cases:

• Concept types

  • Medium abstraction, stable rules, moderate visualizability (functions, kinematics, basic probability, simple dynamical systems).
  • Topics where intermediate states and multi-step causal chains matter (e.g., composition of transformations, multi-step pipelines).
  • Structures with a small set of deep invariants (conservation, monotonicity, symmetries) that can be mentally projected across cases.

• Learner profiles

  • Intermediates with basic schema who risk illusion-of-understanding from playful manipulation; forcing mental simulation + delayed reveal adds generation and retrieval.
  • Higher prior knowledge learners who can mentally run the system and benefit from errorful prediction before visual confirmation.
  • Regulated but lower-WM learners, if visuals remain simple and serve as stable reference frames, reducing search while keeping prediction effortful.

Less suited / risky cases:

• Very fragile novices or high anxiety: heavy mental simulation with sparse feedback can feel opaque; they often need direct, stepwise manipulation first.
• Highly opaque or symbol-heavy domains where visuals are weak cues; mental simulation here risks guessing rather than structured reasoning.
• High-dimensional interfaces: mental simulation of many coupled variables quickly overloads learners.

Mechanism sketch:

• Shifts effort from outcome-matching via sliders to predict–check–revise loops.
• Increases desirable difficulties: spaced retrieval of mappings, forced articulation of intermediate states, and tighter alignment between mental model and later visual feedback.
• Reduces some illusion-of-understanding from rich manipulation traces that are actually shallow.

Net: treat this design as a strong candidate for intermediate learners on medium-complex, rule-governed concepts where internal simulation is feasible, but not as a blanket replacement for well-constrained interactive manipulation for novices or very complex visuals.