{"id":"arxiv/2609.05258","url":"https://arxiv.org/abs/2609.05258","tags":["math.OC","cs.AI"],"score":0,"title":"Ask Before You Optimize: Dynamic Pre-Formulation Clarification for Interactive Optimization","domain":"math.OC","source":"arxiv","authors":["Sihan Ge","Yichen Lin","Chenyu Zhou","Jianghao Lin","Tao Yao","Dongdong Ge"],"reviews":0,"subject":"math.OC","summary":"Large language models (LLMs) are increasingly used to formulate optimization models from natural-language problem descriptions, yet realistic operations research (OR) requests are often incomplete: missing objectives, constraints, or business rules can change the resulting mathematical program. Existing evaluations largely assume a complete specification and therefore overlook whether an agent knows when clarification is needed before modeling. We introduce OR-Clarify, a benchmark for pre-formulation clarification. Each task presents a partial public problem description, withholds structured…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-04","reactions":{}}