Causal Sensor and Actuator Placement for Identifiable Interventional Effect Estimation

Sensor and actuator placement is often guided by objectives such as improving prediction accuracy, monitoring system behavior, or estimating model parameters. Although these goals support effective modeling and forecasting, they do not guarantee causal validity for answering interventional or counterfactual
questions, particularly in the presence of confounding, partial observability, or feedback-driven data. This work formulates placement as a causal design problem driven by identifiability and interventional information efficiency. Sensor selection is formulated as choosing measurements that enable identifiable
adjustment or proxy sets for the target causal queries, while actuator selection determines intervention channels that maximize expected information gain under feasible excitation constraints. Over-instrumentation is characterized as redundancy
with zero marginal causal value-of-information, enabling principled pruning via submodular optimization. The resulting joint greedy co-design strategy yields minimal sensor and actuator configurations that support identifiable and statistically efficient causal inference, together with tractable information-based surrogates for practical implementation.