Decision Audit Trail — excerpt

Artifact for tracing decision logic and accountability

Artifact purpose

The Decision Audit Trail enables reconstruction of: decision inputs, applied logic, approvals of key assumptions and locations of evidence.

Decision trail structure

• input data and their sources,
• decision logic (including ML components / rules),
• high-leverage assumptions,
• accountability (owners),
• events and decisions over time.

Decision Audit Trail table — excerpt

Sample rows showing how to trace specific decisions.

Time / event	Input / context	Logic / component	Accountability	Evidence / record
2026-01-15 10:32 CET Event-ID: EVT-0012	Credit application from mobile channel; customer segment: SME.	ML model v3.4 + credit policy rules.	Model Owner + Policy Owner.	Feature store snapshot + model version record.
2026-01-15 10:32 CET Decision-ID: DEC-0012-A	Score > threshold, no AML risk flags.	Decision rule: "approve if score ≥ X and no AML flags".	Committee / Credit Risk.	Decision log + stored threshold parameter.
2026-01-15 10:34 CET Override-ID: OVR-0004	Manual override after additional assessment.	Manual override with justification.	Senior Credit Officer.	Decision note + approver signature.

Accountability mapping

Each trail element (input, logic, decision, override) has an assigned owner. Responsibility for the decision cannot "dissolve" between the system and the human.

Audit checklist (excerpt)

• can inputs for a specific decision (ID + snapshot) be reconstructed?,
• is the exact version of the model / rules used at that moment known?,
• does an override have a reason, an owner and alignment with policy?,
• is there an escalation path when a decision is disputed?

Final note (SAMPLE)