Tutoring quality belongs to the base model and agent harness together rather than either alone, as adapter rankings reorder across tiers¶

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? No adapter leads Axis I (learning gain) on more than one base-model tier, so base model and harness interact rather than contribute separably, and single-tier leaderboards mis-rank systems. → Lee 2026

Evidence¶

Lee 2026¶

Lee, U.; Lee, S.; Jeong, Y.; Lee, E.; Shin, M.; Kwon, H. (2026). EduClaw-Bench: A Long-Horizon Benchmark for Pedagogical LLM Agents with Simulated Learners. https://arxiv.org/abs/2608.03206 doi:10.48550/arxiv.2608.03206

q2 · i? · design · r2

Benchmark evaluation of 10 agent adapters across 3 base-model tiers over 55 scenarios, each averaged over 4 students. The authors report that "No adapter leads Axis Ionmorethanonetier", demonstrating that "base model and harnessinteractratherthancontributeseparably."

"No adapter leads Axis Ionmorethanonetier (openclawonSolar-pro3+0.36%,zeroclawonCodex-gpt5.5 −0.08%, metaclaw on Qwen3+0.64%), so base model and harnessinteractratherthancontributeseparably"

Discussion¶

Related Claims¶

• Almost no model-and-harness combination sustains tutoring over the full 30-day horizon, as learning plateaus within 5–10 days — related
• LoRA RFT with a helpfulness-weighted composite reward can collapse an adapter's pedagogy mid-training while responsiveness stays intact — related
• Reasoning-capable model variants and chain-of-thought prompting yield no measurable improvement in either alignment axis for classroom evaluation — related
• DeepTutor improves overall first-person interactive tutoring quality by 10.76% over a Naive Tutor baseline on TutorBench — related