Why abduction matters
Population-level causal effects answer "on average, what would happen under intervention?" — but individual-level counterfactuals require something more: the specific, unobserved background factors that made this case's outcome what it was. Abduction recovers those factors from what was actually observed, anchoring each U node to the value consistent with the facts of the case.
Without abduction, a counterfactual query uses population-average noise — which may be badly wrong for any particular individual. With it, you hold the individual's background fixed and ask what would have happened under different circumstances: the same case, different decision. This is the formal engine behind personalised medicine, individual legal liability, insurance coverage disputes, and any question that begins "but for this decision, what would have happened?"
In a linear Gaussian Structural Causal Model (SCM): Node = intercept + Σ(weight × parent) + U. When a node is observed and parents are known, U = observed − predicted exactly. This anchors the individual's background before applying a counterfactual.