Search papers, labs, and topics across Lattice.
This study audits the effectiveness of a general-purpose helpfulness rubric in distinguishing between direct answer-giving and pedagogical guidance in LLM tutoring. Using a pre-registered design, the authors compare conversational and pedagogical policies across three tutor models, revealing that while helpfulness ratings are judge-contingent, pedagogical assessments show consistent rank separation. The findings indicate that general-purpose helpfulness is not a reliable signal for pedagogy, suggesting the need for dedicated evaluation rubrics in tutor assessments.
General-purpose helpfulness metrics fail to reliably signal effective pedagogy in LLM tutoring, revealing a critical gap in evaluation methods.
LLM tutoring poses a measurement problem: can a general-purpose helpfulness rubric distinguish direct answer-giving from pedagogical guidance? We audit this signal in a pre-registered study. Within each of three tutor bases, we compare conversational and pedagogical policies instantiated with the same underlying model and paired with one fixed weak simulated student. Deterministic detectors measure answer leakage and next-turn independent work. Claude Opus 4.8 is the frozen, condition-blind primary judge. After the Opus scores were fixed, GPT-5.6 Sol was prospectively specified for a post hoc robustness audit of the same 1,179 confirmatory answer-phase tutor turns under the frozen helpfulness and pedagogy rubrics. On the primary base under Opus, the policies do not differ significantly in helpfulness but are perfectly rank-separated under the pedagogy rubric (Cliff's $|\delta|{=}0.10$ vs. $1.0$). Across the two judges, pedagogy contrasts retain their direction where detected, whereas the helpfulness ordering is judge-contingent, reversing between judges on two of three bases. In an Opus-only ablation, seven primary-base policies span $2.3$ points in mean judged pedagogy within a $0.25$-point band of mean judged helpfulness. Separately, answer-revealing turns are followed by less independent student work on every base, a result that is judge-invariant by construction. In this controlled setting, general-purpose helpfulness is not a reliable pedagogy signal. Tutor evaluation should pair pedagogy-targeted rubrics with deterministic process measures.
