Test statistical differences between reshaping configurations

Test whether the observed task-success differences between the reported Chain-of-Code, iterative-correction, baseline Instruction-Conditioned Motion Control, and modified Instruction-Conditioned Motion Control configurations are statistically meaningful under matched task sets and generation budgets.

Background

The paper compares several reshaping configurations, but the rows use different task sets, prompts, function sets, and generation budgets. It reports wide Wilson confidence intervals and overlapping intervals for several configurations.

Consequently, the descriptive proportions do not establish whether the apparent differences reflect genuine performance effects. The paper explicitly identifies formal testing of differences as unresolved.

References

These intervals summarize individual proportions; testing differences between rows remains future work.

— Instruction-Conditioned Motion Control for Robotic Arms: Language-to-Code Decomposition and Reinforcement Learning  (2610.19617 - Huang et al., 14 Oct 2026) in Section V, Prompting comparisons, p. 5