SWE-Critix: Post-trained general-purpose verifier for evaluating coding agent trajectories

The paper proposes SWE-Critix, a post-training pipeline for training a general-purpose verifier for coding agent trajectories. Given a task and its execution trajectory, the trained verifier predicts a completion label and generates natural language analysis explaining its judgment. The pipeline combines supervised fine-tuning (SFT) with reinforcement learning (RL), training a series of 30B models on CodeForge-Preview trajectories. On a held-out test set, classification accuracy improves from 57% for the base Qwen3-30B-A3B reasoning model to 66% after SFT and 78% after RL. The authors release the models, dataset, and training code.