Model Transparency

Explanation

Model transparency provides a single composite score for how understandable a model is overall, combining how well it can be explained, how completely it is documented, and how interpretable its mechanics are - so that a customer-facing lending model can be held to a clear, auditable transparency bar rather than a vague judgement. It matters because regulators expect lending decisions to be explainable to the borrower, and one rolled-up index makes it easy to see whether a model clears that bar. It is computed as Transparency Index = (w₁×Explainability Score + w₂×Documentation Score + w₃×Interpretability Score) / (w₁+w₂+w₃) with weights w₁=0.4, w₂=0.3, w₃=0.3 - a weighted average that puts the most emphasis on explainability. To implement it, define how each of the three sub-scores is assessed (for example explainability from the quality and coverage of post-hoc explanations, documentation from documentation-coverage checks, interpretability from model structure), score each on a consistent scale, apply the fixed weights, and log the sub-scores, weights, and resulting index per model version as evidence on every release and review. The threshold is >0.75 for customer-facing lending decisions; an index at or below 0.75 is a breach that should block deployment or continued use of the lending model and trigger remediation of the weakest sub-score. Aligns with RBI Understandable by Design, NIST MEASURE 3.1 and 4.1, and ISO Clauses 6.2.2 and 9.1.