Answer-revealing tutor turns are followed by less independent student work on every tutor base, a deterministic judge-invariant result¶

Claim · All claims
Evidence · 1 study · causal r2 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? Leakage predicted less independent student work on every base, with pooled coefficients from −0.386 to −0.592, all p<10−11. → Shuyi Fan 2026

Evidence¶

Shuyi Fan 2026¶

Shuyi Fan, Boyuan Deng, Mengyu Xu, Jiale Liu, Hongyang Zhang, Qiaoxin Yang, Chongyang Gao. (2026). Rethinking LLM-Judged Helpfulness as a Pedagogy Signal: A Pre-Registered Audit Across Tutor Models. arXiv. https://arxiv.org/abs/2607.28128 doi:10.48550/arxiv.2607.28128

q3 · i? · causal · r2

Deterministic process analysis across the three tutor bases (358, 379, and 442 answer-phase tutor turns), using string-match leakage detection and a reasoning detector; a replicate-clustered sensitivity analysis yielded comparable estimates on all three bases (all p≤.006).

"Leakage predicted less independent student work on every base (βfrom −0.386to−0.592, allp<10 −11; Table 2b)."

Discussion¶

Related Claims¶

• On the primary base, a turn that leaks the answer predicts a higher Opus helpfulness rating and less next-turn independent student work — a narrower finding that bears on this claim
• Single-turn tutors disclose answers pervasively while challenge scores are near zero across every model and subject — related
• The leakage manipulation fires: ConvTutor leaks far more than PedTutor on the primary base, while probe accuracy saturates and cannot distinguish conditions — related