Validator v0.5.14 released with new model stack

TrajectoryRL released validator v0.5.14, switching the S1 evaluation stack to Qwen3.5-35B-A3B testee and GLM-5.1 judge via OpenRouter. Validators must update `.env.validator` configuration by replacing old `CLAWBENCH_*` variables and restart Docker containers. Meanwhile, community members flagged suspected hardcoded submissions from multiple miners and discussed evaluation errors including JSON parsing failures and iteration limits.

• •Validator v0.5.14 uses Qwen3.5-35B-A3B + GLM-5.1 judge through OpenRouter
• •Required `.env.validator` config changes; old CLAWBENCH variables no longer supported
• •Docker restart needed after updating environment file
• •Multiple miners suspected of hardcoded or copied submission packs under review
• •Evaluation errors reported: JSON parsing failures, 15-iteration agent limits, missing conversation traces