Скачать или смотреть 7 Adaptability Guidance

Adaptability guidance for applying a standard Fairness Audit Playbook to diverse domains, such as healthcare and finance, and across different machine learning problem types like classification, regression, and ranking. The guidance emphasizes that while the core phases of the audit must remain consistent, inputs like historical context, protected attributes, fairness definitions, and disparity thresholds must be rigorously updated for each new application to ensure methodological rigor. Crucially, the text mandates a re-run of the Historical Context Assessment for every domain to identify unique harms, and requires the selection of problem-appropriate metrics, such as using residual parity for regression tasks instead of classification metrics. The document stresses that intersectional metrics, clear accountability, and explicit fairness definition selection are non-negotiable elements that must be preserved in every audit, regardless of the system being evaluated.