FORK Reward Mitigates Verbosity in Rubric-based RL

The FORK reward mechanism addresses reward hacking in rubric-based reinforcement learning by jointly optimizing for recall and a precision objective. This approach prevents models from becoming overly verbose or factually incorrect while attempting to satisfy rubric coverage.