Post-training improves TACTutor's auxiliary guardrail flags on TACTBench: accept rate rises from 0.603 to 0.872 and leak-or-overhelp falls from 0.346 to 0.026¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? On the same 78 diagnostic contexts, TACTutor's accept rate increases from 0.603 to 0.872 and its leak-or-overhelp rate falls from 0.346 to 0.026 relative to the backbone. → Dongjie Yang 2026

Evidence¶

Dongjie Yang 2026¶

Dongjie Yang, Siyan Lin, Leixian Shen, Rui Sheng, Huamin Qu, and Zixin Chen. (2026). TACT: Taxonomy-Aligned Post-Training for Pedagogically Adaptive English Tutoring. https://arxiv.org/abs/2608.03952 doi:10.48550/arxiv.2608.03952

q2 · i? · design · r2

Auxiliary judge-flag rates on the TACTBench diagnostic runs (Appendix Table 6), reported as guardrails separate from TACT_Overall. The article states accept rate increases "from 0.603 to 0.872" and leak-or-overhelp falls "from 0.346 to0.026"; no significance test is printed.

"This post-training raisesTACT_Overallfrom0.629forthebackboneto0.832;the auxiliary flags also improve, with accept rate increasing from 0.603 to 0.872 and leak-or-overhelp falling from 0.346 to0.026."

Discussion¶

Related Claims¶

• Taxonomy-guided post-training raises TACTutor's TACT_Overall on TACTBench from 0.629 (backbone) to 0.832, the highest among evaluated models — related
• In a blinded study with 50 learners, TACTutor receives the highest overall mean rating (5.54/7) and is rated above its backbone across all four teaching-behavior dimensions (+0.39 overall) — related
• TACTutor shows no degradation relative to its Qwen backbone on external educational benchmarks, with scores rising on MRBench, TutorBench, LongTutor, and DeepTutor — related