31 July 2026
CAAP: Extending CAP-Inspired Architectural Reasoning to Accountable AI
CAP and PACELC never addressed AI systems that decide outcomes for people. CAAP argues Accountability belongs in the same architectural class as Consistency, Availability, and Partition Tolerance — with four trade-off configurations, testable propositions, implementation patterns, and alignment to the EU AI Act, NIST AI RMF, and India's DPDP Act.
Key takeaways
- •Accountability is a first-class architectural constraint — not an ethics overlay that can be bolted on after launch.
- •CAAP formalises four trade-off configurations and two testable architectural propositions for high-stakes AI.
- •The framework sits at the architecture layer, distinct from ethics, governance, and regulation above it.
- •Implementation patterns map to EU AI Act, NIST AI RMF, and India's DPDP Act requirements.
Abstract
The CAP Theorem and its PACELC extension provide principled frameworks for architectural trade-offs in distributed data systems. This paper proposes CAAP — extending that reasoning to AI systems making consequential decisions about people in employment, healthcare, and financial services. Accountability is treated as a first-class architectural constraint alongside Consistency, Availability, and Partition Tolerance: it constrains architecture, forces unavoidable trade-offs, cannot be retrofitted after deployment, and can be operationally verified.