The Algorithmic Auditor

The Algorithmic Auditor argues that modern surveillance—though automated in detection and triage—cannot be made exam-proof by “accuracy” or model explainability alone. Instead, it must be auditable as a system: evidence must be traceable, admissible, replayable, and operationally reviewable under regulatory time pressure. The book develops a governance and engineering blueprint for audit-ready supervision of algorithmic surveillance pipelines, covering control ownership (RACI), evidence-pack design, evidence standards, sampling and escalation workflows, offline validation and backtesting under regime shifts, robustness checks for data and labels, reproducibility workflows, uncertainty signaling in examiner narratives, and versioning/drift/rollback mechanisms. Across these topics, it reframes auditing as a production of defensible artifacts—built into the surveillance lifecycle—so examiners can verify what was done, how decisions were made, and why the firm’s conclusions satisfy sufficiency, relevance, and reliability constraints.