Downward Bloom control is benchmark-dependent: repository-level tasks are easier to simplify than algorithmic puzzles¶

Claim · All claims
Evidence · 1 study · causal 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? SWE-Bench-Verified is the clearest case where simplification is comparatively easier, especially for the general model, while LiveCodeBench is the most resistant setting for downward shifts. → Yi Zhang & Julia Rayz 2026

Evidence¶

Yi Zhang & Julia Rayz 2026¶

Yi Zhang & Julia Rayz. (2026). From Execution to Education: A Bloom-Aligned Framework for Measuring Educational Control in LLMs. Preprint. https://arxiv.org/abs/2607.08009 doi:10.48550/arxiv.2607.08009

q2 · i? · causal · r2

Benchmark-level TZA breakdown (Table 2): general model Lower-target accuracy reaches 0.786 on SWE-Bench-Verified but only 0.072 on LiveCodeBench; the coder model shows 0.630 versus 0.035 on the same contrast.

"SWE-Bench-Verified stands out as the clearest case where simplification is comparatively easier, especially for the general model, while LiveCodeBench remains the most resistant setting for downward shifts."

Discussion¶

Related Claims¶

• The general and coder models deploy distinct lexical strategies when raising cognitive demand, and neither deploys simplification vocabulary downward — related
• LLMs show a directional asymmetry in educational control: they reliably increase cognitive demand but struggle to lower it — a broader claim this one bears on
• Both models target Higher Bloom levels far more accurately than Lower levels, with the general model outperforming the coder model upward — related
• Prompting a learning disability shifts feedback toward word- and sentence-level remediation, simplified vocabulary, and lowered expectations — related