Assurance Frameworks
Structured evaluation for high-stakes AI systems.
Ducaltus develops governance-aware operational assurance frameworks designed to evaluate deployment readiness, subgroup reliability, threshold instability, remediation progression, and governance-sensitive deployment conditions in high-stakes AI systems.
Rather than relying solely on aggregate performance metrics, Ducaltus frameworks evaluate deployment-state instability, governance escalation conditions, subgroup reliability variation, threshold-sensitive behaviour, and operational deployment uncertainty.
Operational AI Deployment Assurance (OADA)
OADA is a governance-aware assurance framework for evaluating deployment readiness, deployment-state instability, remediation progression, reassessment workflows, and high-stakes AI deployment conditions.
The framework connects fairness disagreement, subgroup reliability, threshold sensitivity, governance escalation, and remediation-aware progression into operational deployment assurance.
Fairness Disagreement Index (FDI)
The Fairness Disagreement Index (FDI) measures disagreement between fairness metrics under varying evaluation conditions and decision thresholds.
The framework was developed to address a key limitation in AI fairness evaluation: different fairness metrics can produce conflicting conclusions about the same model.
FairRisk-FDI
FairRisk-FDI extends fairness disagreement analysis into deployment-oriented risk evaluation.
The framework examines how disagreement between fairness metrics can influence deployment decisions under different operational conditions and domain-specific risk priorities.
The framework evaluates whether AI systems remain operationally suitable for deployment under conditions of disagreement, instability, subgroup variation, and governance-sensitive risk.
Intersectional Fairness Evaluation (IFEM)
IFEM is an intersectional subgroup evaluation framework designed to identify deployment-relevant disparities hidden by aggregate accuracy and single-attribute fairness analysis.
The framework evaluates subgroup reliability variation, threshold-sensitive behaviour, mitigation effectiveness, and operational instability across combined demographic conditions.
IFEM supports governance-aware evaluation of whether AI systems remain operationally reliable across intersectional deployment contexts.
Subgroup Reliability Evaluation
Ducaltus evaluation approaches emphasise subgroup-level analysis rather than aggregate performance alone.
This includes assessment of false positive rates, false negative rates, threshold behaviour, and performance disparities across demographic or operational groups.
Deployment-Oriented Evaluation
AI systems deployed in healthcare, law enforcement, finance, and other high-stakes environments require evaluation methods that account for operational consequences rather than predictive performance alone.
Ducaltus frameworks therefore focus on how systems behave under real-world deployment conditions, including instability, uncertainty, and unequal error distribution.
Future Direction
Future Ducaltus research focuses on operational assurance infrastructure, governance-state orchestration, deployment control systems, reassessment workflows, remediation-aware progression, and high-stakes AI deployment governance.