Researchers propose AdmitOR, a label-free admission mechanism for optimization modeling that uses calibrated external behavioral evidence. Unlike existing learners that admit knowledge by checking against known answers, AdmitOR relies on agreement across value-function traces from different model families and solver stacks. In experiments, AdmitOR outperforms majority vote and execution success in terms of precision and macro accuracy, but fails to generalize to wild streams. The study highlights the need for a more robust admission mechanism that can handle real-world data. --- Why it matters: This work matters because it tackles a critical issue in optimization modeling: how to admit knowledge without relying on labeled instances. AdmitOR's performance suggests potential improvements over existing methods, but its failure to generalize raises concerns about its applicability to real-world scenarios. Source: https://arxiv.org/abs/2608.15565