In the public TutorBench sample, agency-oriented rubric criteria are most concentrated in the active-learning use case (agency-dimension density 0.720, agency-skill density 0.411)¶

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

Subclaims¶

q2 i? Across the three use-case families, the active-learning subset has the highest concentration of agency-oriented criteria, with rubric language rewarding guiding questions, hints without full disclosure, and prompting alternative lines of reasoning. → Junyi Yao 2026

Evidence¶

Junyi Yao 2026¶

Junyi Yao, Zihao Zheng, and Baichuan Li. (2026). Beyond Helpfulness: A Teaching-over-Solving Diagnostic for Measuring Educational Impact in LLM Tutors. arXiv preprint. https://arxiv.org/abs/2606.16206 doi:10.48550/arxiv.2606.16206

q2 · i? · design · r?

Deterministic rubric-label mapping of the public TutorBench sample (30 scenarios, 324 rubric items across adaptive explanation, assessment/feedback, and active learning). Table 2 reports agency-dimension densities of 0.583, 0.688, and 0.720 and agency-skill densities of 0.139, 0.294, and 0.411 respectively.

"Even under this stricter item-level counting rule, the active-learning subset has the highest concen- tration of agency-oriented criteria. In qualitative examples, rubric language explicitly rewards ask- ing guiding questions, providing hints without dis- closing the full next step"

Discussion¶

Related Claims¶

• Human raters' preferences align with the LLM judge on TutorBench (Pearson r = 0.82 across metric-level win rates) — related
• Within-judge variance of the repeated LLM-judge rubric is negligible, but cross-model-family agreement remains untested — related
• TACTutor shows no degradation relative to its Qwen backbone on external educational benchmarks, with scores rising on MRBench, TutorBench, LongTutor, and DeepTutor — related