Ethics
Composable ethics interfaces and human-review implementation limits.
Engine composition
ProductionEthicalAIFramework is constructed from five implementations: EthicsEngine, BiasDetector, FairnessValidator, TransparencyManager and HumanOversightManager. These supply principle checks, bias analysis, fairness assessment, transparency checks and oversight operations. The framework aggregates their returned scores and flags; it does not independently measure a model or dataset.
AIDecision contains an ID and summary. HumanReviewRequest adds concerns, urgency, required expertise and a deadline. These are transport types for an actual review system, not evidence that reviewers were notified.
The current assess_hitl_requirement always returns required: false. Consequently the built-in human-in-the-loop decision path does not request review based on decision content. Applications requiring mandatory review must provide an enforced policy path before allowing the action.
The separate default compliance framework's AI-bias assessment also returns a fixed zero score. Do not interpret those defaults as a completed ethics or fairness audit.
Source reference
Exact declarations and source provenance · Package features and manifest.
