Guide: Explainability (XAI)

Attach SHAP/LIME values to each decision so auditors can see why the model decided.

Capture SHAP/LIME

xai.pypython

record = client.records.create(
  model_id="credit-model-v2",
  input={"income": 85000, "debt_ratio": 0.32, "credit_score": 720},
  output={"decision": "APPROVED", "limit": 25000},
  confidence=0.94,
  explanation={
    "method": "SHAP",
    "values": {"income": 0.35, "debt_ratio": -0.28, "credit_score": 0.22}
  }
)

Bundle Artifact

explanation.jsonjson

explanation.json
{
  "method": "SHAP",
  "values": {
    "income": 0.35,
    "debt_ratio": -0.28,
    "credit_score": 0.22
  },
  "version": "1.0.0"
}

For deterministic explanations, pin data preprocessing versions and model versions (see Model Cards).