Causal audit traces for high-risk AI decisions
Tionne Smith · Zenodo (CERN European Organization for Nuclear Research) · 2026
A technical brief presenting C-DAG, a causal audit-trace architecture for high-risk AI decisions. C-DAG is implemented by `causal-credit-risk-engine`, a config-driven causal DAG system that generates counterfactuals, replayable audit records, fairness diagnostics, evidence packs, and tamper-evident decision traces to support explainability, oversight, model-risk review, and AI governance workflows. This version expands the original reference architecture with public financial validation artifacts, a benchmark evidence dashboard, public loss-exposure mapping, and baseline outcome-validation results. The work demonstrates how high-risk AI decisions can be converted into verifiable evidence: trace → counterfactual → replay → hash-chain → evidence pack → risk exposure. C-DAG is a public reference implementation and workflow evidence benchmark. It is not a production lending model, a live credit decision system, or a legal/regulatory certification tool.Keywords: causal AI, AI governance, explainable AI, audit traces, model-risk management, counterfactual reasoning, EU AI Act, high-risk AI, deterministic replay, fairness diagnosticsGit: https://github.com/electricwolfemarshmallowhypertext/c-dag