TAG ARCHIVE
audit
4 MARIA OS blog articles tagged audit. Safety, accountability, fail-closed gates, auditability, and human-in-the-loop control for AI agents. This canonical topic archive supports search engines and LLM retrieval.
Judgment OS / Decision Intelligence OS
Core MARIA OS research on turning organizational judgment into executable decision systems.
Agentic Company Architecture
Research on human-agent organizations, delegation boundaries, role topology, and governed autonomy.
Responsibility Gates and AI Governance
Safety, accountability, fail-closed gates, auditability, and human-in-the-loop control for AI agents.
Multi-Agent Mathematics
Formal models for convergence, stability, game theory, graph dynamics, and multi-agent evaluation.
Evidence, RAG, and Knowledge Governance
Evidence bundles, retrieval architecture, Graph RAG, knowledge trust, and auditable reasoning pipelines.
Agentic R&D and Judgment Science
Research operations, simulation labs, judgment science, recursive improvement, and experimental AI governance.
Audit Universe Runtime: Agent Design for Executing Audit Procedures as Runtime Operations
Transforming ISA/JICPA standards into executable agent specifications — from sampling strategies to substantive testing, within a MARIA OS governance architecture
Traditional audit procedures are encoded in prose-based standards that resist automation. This paper presents the Audit Universe Runtime — a multi-agent execution environment within MARIA OS that compiles audit standards (ISA, JICPA) into executable agent task specifications. We formalize audit procedures as state machines, design sampling strategy agents with statistical rigor, implement real-time anomaly detection during substantive testing, and prove audit completeness through a formal coverage model. The architecture maps MARIA coordinates to engagement structures, enabling continuous auditing with immutable audit trails and human-agent collaboration gates at every materiality threshold.
Evidence Coherence Spectral Analysis: Detecting Fraud Through Eigendecomposition of Audit Evidence
Using spectral methods on evidence correlation matrices to identify inconsistencies, fabrication patterns, and systemic fraud signals
Traditional audit systems often rely on rule-based checks and statistical sampling, which can under-detect coordinated fabrication patterns. This paper introduces Evidence Coherence Spectral Analysis, a framework that treats evidence sets as vector spaces, builds correlation matrices from evidence attributes, and applies eigendecomposition to identify anomalous spectral gaps associated with inconsistency or fabrication risk. We define a coherence score, relate it to false-discovery behavior, and describe integration with MARIA OS Evidence Bundles. In controlled financial-statement audit experiments, spectral analysis detected 94.7% of fabricated evidence sets while maintaining a false-positive rate below 2.3%, with streaming support for near-real-time analysis.
Audit Stopping Criteria: Mathematical Foundations for Knowing When Enough Is Enough
Defining audit termination conditions through MAX constraints and probability thresholds to minimize False Allow Rate
Every audit faces the same question: when is evidence sufficient to stop? Stopping too early can allow defects to escape into production, while stopping too late consumes budget and attention with diminishing returns. This paper formalizes audit stopping criteria as a constrained optimization problem, derives solutions under MAX constraints and sequential probability ratio testing, and describes integration with the MARIA OS Fail-Closed Gate Engine. In evaluated SOX workloads, the approach reported a False Allow Rate below 0.3%.
Auditable Financial Decision Traceability: Evidence Graph Models for Regulatory Compliance
Formal evidence graph construction and matrix-algebraic traceability for reconstructing every financial decision under SOX, Basel III, and MiFID II
Regulatory reconstruction of AI-driven financial decisions is difficult when logs are fragmented, timestamps drift, or causal links are missing. This paper introduces a formal evidence-graph model where each decision is an immutable node in a directed acyclic graph, linked by typed causal edges with cryptographic evidence bundles. We define `TraceCompleteness` as `TC = |reproducible decisions| / |total decisions|` and report `TC >= 0.997` across evaluated SOX, Basel III, and MiFID II audit scenarios.