MAIVA: Multi-Agent Integrity Verification Architecture

MAIVA (Multi-Agent Integrity Verification Architecture) is a research prototype for Byzantine-resilient trust aggregation and action authorization in autonomous multi-agent systems. The system integrates:• Hierarchical Multi-Attribute Authority (HMAA) individual authority computation• SAAP Byzantine-resilient swarm authority aggregation• Three-layer anomaly detection using thresholds, CUSUM temporal accumulation, and enrollment drift monitoring• Graduated action authorization aligned with DoDD 3000.09 governance principles• Formal design specification in TLA+ The artifact is a self-contained browser application (2,396 lines) requiring no installation. It includes 37 inline self-tests, Monte Carlo evaluation scenarios, and sensitivity analysis tools. This Zenodo deposit contains the runnable prototype, documentation, formal specification, and configuration files used for evaluation. Live simulation:https://burakoktenli.com/maiva-simulation Source repository:https://github.com/burakoktenli-ai/maiva