Fundamental Problem of Causal Inference
For each unit, at most one of the two potential outcomes can be observed:
$$Y_i(1) \quad \text{or} \quad Y_i(0),$$but not both. This means causal effects are inherently missing-data problems at the unit level.
The entire logic of randomization, assignment mechanisms, and assumptions such as SUTVA is built around coping with this missing counterfactual information.