Cause of GIL performance degradation on coefficient-rich equations

Determine whether the Guided Inquiry Learning method’s requirement for early hypothesis proposal causes its performance degradation as equation complexity and the coefficient search space increase.

Background

The paper evaluates IL and GIL agents on increasingly realistic mathematical equations, including equations with intercepts and numerical coefficients. Performance declines substantially when coefficients are introduced, and the authors observe that free-form inquiry can slightly outperform GIL in this setting. They hypothesize that GIL’s enforced early hypothesis-generation phase may be poorly suited to the larger search space and greater uncertainty of coefficient-rich equations. Establishing whether early hypothesis proposal is the causal source of this degradation would clarify how guided inquiry should be adapted for more complex symbolic-model discovery tasks.

References

We hypothesize that this difference arises because GIL requires early hypothesis proposal during exploration, which becomes increasingly challenging as equation complexity and the coefficient search space grow.

— Evaluating Inquiry Learning in LLM-Based Agents: Hypothesizing, Reasoning, and Exploration  (2610.21604 - Knežević et al., 15 Oct 2026) in Appendix, Section “Equations with coefficients”