Causal Foundry

CausalForge is an autonomous causal-discovery laboratory powered by Meta Muse Spark. In each procedurally generated world, an unknown rule connects four controllable factors - temperature, pressure, catalyst, and field - to an observed response. Muse Spark is not given the rule. It makes scientific decisions through real API tool calls, selects interventions, observes the simulation results, and uses updated evidence to plan its next experiment. A local Bayesian predictive model learns from every observation, while the interactive 3D interface visualizes the authoritative simulation state, experiment history, learned coefficients, uncertainty, API cost, and final comparison with the hidden ground truth. CausalForge demonstrates how AI can move beyond answering questions to actively forming hypotheses, running experiments, and discovering causal relationships. Across our evaluation on 10 unseen procedural worlds, it correctly identified the active causal factors in all 10 runs while using approximately $0.72 in Meta API credits.