A new arXiv preprint introduces DecisionRobust, a benchmark for JEV-style decision models. These models provide typed interfaces that map task instructions and observations directly to choices, judgments, and scores.

The benchmark spans text and vision, aiming to measure how robust such models are as they are increasingly used for automated decision-making. The abstract positions DecisionRobust as a response to the need for systematic robustness evaluation.

Since this summary relies on a single source, there are no conflicting results or independent claims to compare. The abstract does not specify the exact evaluation tasks or robustness metrics used.