Why Learners Can Follow the Tutorial but Cannot Recover Alone

Reproduction is not independent performance

A learner can mirror visible actions while the presenter supplies every cue, option, and verification. The diagnostic question is what happens when one surface detail changes or the expected state fails to appear. Capture a parallel task and mark the first point where the learner cannot select the next action, explain the reason, or verify the result. That divergence is more useful than a general completion rate.

How People Learn II emphasizes context, prior knowledge, feedback, and transfer. It supports testing use beyond the rehearsed sequence, but it does not prove why one learner struggled. Preserve task conditions, assistance, and accessibility before choosing a cause.

Evidence: National Academies of Sciences, Engineering, and Medicine

Branch 1: starting context was hidden

Expected trace: the tutorial works only with the presenter’s version, permissions, sample data, or preconfigured state. Learners fail before the first substantive decision. Compare environments and list unstated prerequisites. A version label alone is not enough if account role, prior setup, or regional interface changes behavior.

Repair by documenting a reproducible starting-state check and routing missing prerequisites. Avoid making learners repeat the whole tutorial when one condition differs. CMU’s worked-example approach is relevant because an example must make the problem state and reasoning visible, not only the solution steps.

Evidence: Carnegie Mellon University Eberly Center; National Academies of Sciences, Engineering, and Medicine

Branch 2 and 3: imitation or missing recovery model

Imitation failure appears when learners can act with the same labels but cannot recognize the principle under different wording or data. Ask for a prediction or explanation before the next click. Missing recovery appears when learners proceed under ideal conditions but stop at an error, wait state, or accidental action. Present a safe constructed fault and ask what observation should come first.

Do not add a large troubleshooting appendix before diagnosing which model is missing. For imitation, expose cues and rejected alternatives. For recovery, teach symptom, competing cause, non-destructive check, correction, and verification. CMU’s active-learning strategies support eliciting reasoning; the method should reveal a decision, not merely create more screen interaction.

• Imitation trace: success depends on identical labels or examples.
• Recovery trace: first unexpected state ends the learner’s decision path.

Evidence: Carnegie Mellon University Eberly Center

Branch 4: prompts or polish conceal the gap

Narration, highlights, zooms, cursor trails, and editing can direct attention so effectively that learners never practice finding the cue. Remove one support in a parallel attempt and record whether the learner knows where to look. The diagnosis is not that production quality is bad; it is that support needs deliberate fading and a transfer check.

Likewise, a polished sequence may delete waiting, confirmation, and uncertainty that occur in real work. Compare the edited timeline with the actual process. If omitted states change the decision, restore or explain them. Do not use raw realism indiscriminately; keep irrelevant delay compressed while preserving consequential cues.

Evidence: Carnegie Mellon University Eberly Center; National Academies of Sciences, Engineering, and Medicine

Branch 5: the medium creates an access barrier

W3C’s media accessibility guidance provides checks for captions, transcripts, description of essential visuals, and accessible players. A learner may miss information conveyed only through cursor motion, color, tiny text, sound, or rapid timing. Test alternative modes and keyboard or assistive access where relevant. If the barrier is removed and performance changes, credit accessible design rather than a learner deficit.

CAST’s UDL Guidelines support multiple representations and ways to act or express understanding. Keep the task objective stable while changing an irrelevant medium constraint. Synchronize alternatives; an obsolete transcript can generate a different instruction path and appear as a reasoning failure.

Evidence: CAST; W3C Web Accessibility Initiative

Use the first-divergence evidence sheet

Record task context, expected cue, learner observation, action, support used, result, competing cause, and next separating observation. Compare one successful and one unsuccessful path. Choose one repair and retest on a changed example. Do not remove required accommodations or expose learners to unsafe errors for diagnostic purity.

This method can reveal instructional gaps but cannot establish durable learning or causal impact from a small sample. Escalate tasks with privacy, financial, safety, health, or regulated consequences to appropriate expertise. Keep a correction owner and recheck by 2027-02-10 or when the interface, process, audience, or media asset changes.

Evidence: CAST; W3C Web Accessibility Initiative; National Academies of Sciences, Engineering, and Medicine