Paper Identifies Deployment Gaps in Oracle Partition Scores for Algorithm Selection

A new arXiv paper argues that oracle-style benchmarks such as virtual best solvers and best-in-family summaries are often treated as achievable ceilings, even though they ignore the constraints a real selector faces. The author examines decomposed algorithm selection and shows how partition-level scores diverge from what an actual deployed system can deliver. The work calls for evaluation practices that reflect deployment conditions rather than oracle upper bounds.