{"name":"mechatron","display_name":"Mechatron","description":"OpenEnv-compatible reinforcement-learning arena for training LLM agents to investigate hidden transformer circuits through tool use, long-horizon planning, and verifiable rewards.","curriculum":{"task_count":11,"difficulty_tiers":["easy","medium","hard","expert"],"themes":["world_modeling","long_horizon_planning","self_improvement_curriculum","scientific_tool_use"]},"tasks":["head-identification","circuit-localization","full-hypothesis","ioi-circuit-discovery","name-mover-localization","previous-token-head","duplicate-token-head","backup-head-hydra","skip-trigram-circuit","superposition-mlp-sae","grokking-fourier-mod-add"],"actions":["ablate_head","ablate_heads","patch_activation","query_logit_lens","query_attn_pattern","submit_hypothesis"],"endpoints":["/reset","/step","/state","/health"]}