unassisted-evidence-checkpoint

An unassisted evidence checkpoint measures phantom attainment by having learners attempt a new problem without AI help, capturing pre- and post-attempt confidence to compare scaffolded versus unassisted performance.

Phantom attainment occurs when AI-assisted work inflates apparent mastery. Prevent it by scheduling an unassisted evidence checkpoint that blocks hints and feedback, comparing scaffolded practice performance against independent capability results.

Capture confidence calibration by collecting pre- and post-attempt confidence scores directly before and after the learner completes the unassisted practice attempt, tagging the session metadata as unassisted for accurate learning assessment.

Use an unassisted evidence checkpoint after scaffolded practice when preparing for exam-readiness calibration or transfer checks. It is needed for domains where AI-assisted work can inflate apparent mastery and evidence-gating requirements demand independent verification.

Evaluate transfer checks by scoring the unassisted attempt as passed, failed, or skipped. Guide learner reflection on the results to identify exactly what to practice to close the assistance gap when the learner depends on support.