Student backtracking before setting up the correct induction decreased across exercises in both deployments, indicating a learning curve¶

Claim · All claims
Evidence · 1 study · design r3 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Backtracking before setting up the correct induction fell from 94% to 50% of students across exercises in the first deployment and from 33% to 0% in the second. → Matthew Keenan 2026

Evidence¶

Matthew Keenan 2026¶

Matthew Keenan, Nishant Kheterpal, Jean-Baptiste Jeannin, Cyrus Omar. (2026). Hazel Prover: A Classroom Proof Assistant for Learning Structural Induction. https://arxiv.org/abs/2608.23309 doi:10.48550/arxiv.2608.23309

q2 · i? · design · r3

Fine-grained analysis of replayed usage logs (Table 1(a)) from 18 analyzed first-deployment submissions and 14 second-deployment submissions, splitting proofs into milestones. The printed percentages are descriptive; no effect size is reported.

"In the first deployment these went from 94% of students backtracking the first exercise to 50% of students backtracking in the last exercise. In the second deployment, it went from 33% to 0%."

Discussion¶

Related Claims¶

• Students backtracked far more before applying the inductive hypothesis than after, suggesting difficulty finding where to place it — related
• After the design iteration, no student deleted an evaluation step, in contrast to frequent step deletion in the first deployment — related
• Requiring students to write out evaluation steps manually improved transfer to pen-and-paper exam performance between the two deployments — related