Active Causal Discovery Benchmark: Evaluating LLM Agents Under Budgeted Interventions
What this work does
Generates linear-Gaussian SCMs with an observe/intervene/submit protocol and separate skeleton, directed-graph and intervention-efficiency scores.
Connection to discovery
A direct environment for testing causal experiment selection separately from observational structure inference and overconfident edge commitment.
Evidence
Inspected hard-intervention budgeting, partial orientation submissions and the audit showing a surprisingly high structure-blind random-graph floor.
Scope and limitations
The initial dense ladder requires calibration and has small graph sizes, causal sufficiency and linear-Gaussian assumptions. The paper explicitly frames current results as an audit, not solved causal discovery.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.