MaSCoD Multi-Agent Framework for Causal Graph Generation

MaSCoD uses a multi-agent approach to organize structural patterns and third variables before direct-edge judgment in causal discovery. Using GPT-5.4, the framework showed higher mean Recall and F1 in 'Full' mode compared to integrated judgment, though it increased false-positive rates.