Bäuerle and colleagues' 2026-05-07 arXiv v1 study followed 11 external mathematicians using an experimental AlphaEvolve interface over three months. The authors call iterative goal refinement 'intentmaking' and interpretation of results 'sensemaking'; users repeatedly moved between them. They describe difficulty translating mathematical goals into workable objectives and anticipating pathological system interpretations. This is evidence of experts redirecting AI-guided search, not merely accepting outputs. Its formative qualitative design and selected mathematical setting do not establish a population-wide comprehension lag or prove that the interface improves judgment relative to alternatives.

Evidence status: formative_qualitative_findings.