Base Value E[f(x)] 0.620
Net Attributions Σφi -0.340
Model Probability f(x) 0.280
Approval Threshold 0.500
Additivity Axiom Strict 100%

SHAP Waterfall Breakdown (Local Accuracy)

Baseline expectation 0.62 + sum of marginal Shapley contributions Σφi strictly equals output f(x).
Increases Approval (>0) Decreases Approval (<0)

Actionable Algorithmic Recourse

Minimal distance perturbations Δx required to cross decision threshold τ = 0.50 without altering immutable traits.
Why Counterfactuals Matter:

As DhanushKumar notes: Instead of telling an applicant why a 500-tree ensemble rejected them, counterfactuals answer the human question: “What is the smallest realistic change that flips this outcome?”