MatrixReward: Reward from Rubric Matrix for Open-Ended Generation

Excerpt

arXiv:2610.00389v1 Announce Type: new Abstract: Open-ended query generation lacks standard answers, thus necessitating an effective reward mechanism. Pointwise scoring rubrics provide limited information about the relative quality of sample answers under the same prompt; merging multiple rubric judgments into a single score may also mask the differences between these answers. We propose MatrixReward, which constructs rewards from a rollout-by-rubric win-rate matrix obtained by comparing every pa