ifeval-obf-rl-run

GRPO-LoRA RL run on the Wood-Labs 49B organism, reward = IFEval-verifiable + RM. LoRA adapters saved every 25 steps + live cheap metrics (VEA-regex, IFEval, RM-reward). Demonstrates VEA obfuscation.

Status: reserved — populating. Companion to rlundqvist/ifeval-obf-rl-preferences and the paper LLM Judges Disprefer Evaluation Awareness.