GeoReform: Reflective Formalization Evolution for Multimodal Geometry Problem Solving

Source facts

The preprint proposes GeoReform, which treats geometric formalization as an optimizable policy rather than a fixed parser output. It collects failed rollouts, diagnoses representation defects, and mutates how entities, relations, constraints, and targets are selected and presented. On Geometry3K, the authors report an accuracy increase for Qwen3VL-2B from 42.0% to 56.0%.

Analysis

GeoReform identifies representation design as a bottleneck in multimodal reasoning. The feedback loop links parsing choices to downstream failure, though its gains are reported within geometry benchmarks and may not transfer to other diagram domains.

Uncertainty/Doubt

• This is an arXiv preprint. The reported benchmark gains, failure diagnoses, and generalization beyond geometry were not independently reproduced here.