Traceability

Explanation

Traceability checks whether each decision the AI system makes can be reconstructed end-to-end from a complete audit trail - the inputs used, the model version, the features and scores, the rules applied, and the final outcome - so that a regulator or a customer dispute can be answered with hard evidence years later. In banking this is critical: a customer denied credit can demand to know why, and the bank must retain that record. It is measured as Trail Completeness = % of decisions with complete audit trails, applied to credit decisions with a 7-year retention requirement. To implement it, instrument the decision pipeline to emit a structured, immutable record for every decision capturing inputs, model and feature versions, intermediate scores, applied policies, and the outcome, written to tamper-evident storage with enforced 7-year retention; run periodic completeness reconciliation that flags any decision whose trail is missing a required field. Log the completeness measurement itself and any gaps found. The threshold is 100% for all banking decisions, meaning every single decision must have a complete trail; any decision lacking one is a breach that should trigger investigation, gap remediation, and review of the logging pipeline that allowed the omission. Aligns with RBI Understandable by Design, NIST MEASURE 3.1 and 4.1, and ISO Clauses 6.2.2 and 9.1.