Causal Identification from Counterfactual Data: Completeness and Bounding Results

Columbia University researchers Arvind Raghavan and Elias Bareinboim have published a breakthrough paper establishing fundamental limits for causal inference from counterfactual data. Their work builds on their previous 2025 research that first characterized 'counterfactual realizability'—the concept that certain Layer 3 distributions in Pearl's Causal Hierarchy can be directly estimated through experimental methods. The new CTFIDU+ algorithm provides a complete solution for determining which additional counterfactual quantities become identifiable when researchers have access to these experimentally-obtainable Layer 3 distributions, answering a critical open question in causal AI.

The researchers prove CTFIDU+ is complete for identifying counterfactual queries from arbitrary sets of Layer 3 distributions, establishing what they call 'the fundamental limit to exact causal inference in the non-parametric setting.' For counterfactuals that remain unidentifiable even with this new data access, they derive novel analytic bounds that can be tightened using realizable counterfactual data. Their simulations demonstrate practical improvements in bounding previously unquantifiable 'what if' scenarios, with implications for medical trials, policy analysis, and any domain requiring precise causal understanding from limited experimental data.