fa1zvn

RL/eval environment for ghostwriting in a client's voice, scored by deterministic rule checks plus an LLM judge.

Multi-turn contract negotiation environment with verifiable reward function

Run OSU-NLP QUEST's released RL tasks + rubric-tree rewards without their withheld cached corpus (live-fetch reproducibility bridge)

Mind2Web web-action prediction with swappable binary vs shaped GRPO reward

Multi-term contract negotiation with a verifiable, Pareto-frontier reward (no judge)