rl_core-e2-d-rl3000-seed42-verified-v2

A reinforcement-learning-with-verifiable-rewards (RLVR) model checkpoint produced by the Pre-to-Post-2 project, trained for 3,000 RL steps with seed 42 and filtered to verified outputs. It represents a post-training alignment experiment in which a language model is fine-tuned using deterministic correctness signals rather than human preference labels. Source: supabase.