Evaluating a counterfactual using a Structural Causal Model (SCM) involves three sequential steps, which are valid regardless of whether the functional relationships are linear or non-linear:

1. Abduction: Use evidence (observations) to estimate the values of the unobserved background (exogenous) variables $U$.
2. Action: Use the do-operator to intervene on the model, setting the value of the antecedent variable (e.g., setting $M = m$), which modifies the structural equations by removing the equations that originally determined that variable.
3. Prediction: Use the modified structural equations and the estimated values of $U$ to calculate the counterfactual values of the outcome variables (e.g., $Y$).