rl-rubric-anchors

Extend RL to open-ended tasks using structured rubrics as reward anchors, enabling fine-grained evaluation of subjective outputs without requiring binary correctness signals.

Independent GitHub snapshot · Automated classification · Original documentation governs