Governed Auditable Decisioning Under Uncertainty: Synthesis and Agentic Extension

Oleg Solozobov's new paper on arXiv (2604.19112) tackles a critical gap in AI governance: when automated decision systems fail, even compliant infrastructure can't reconstruct what happened. The work synthesizes an operational governance evidence framework combining structural accountability collapse diagnostics, decision trace schemas, evidence sufficiency measurement, and label-free monitoring into an integrated chain. It then analytically assesses transferability across four decision system architectures.

The cross-architecture comparison reveals a clear governance coverage gradient. Deterministic rule engines achieve full DES-property fillability, hybrid ML+rules systems achieve partial fillability, classical ML systems achieve only minimal fillability, and agentic AI systems encounter structural breaks. For agentic systems, Solozobov identifies three specific breaks: decision diffusion (where decisions spread across multiple agents), evidence fragmentation (where audit trails become scattered), and responsibility ambiguity (where no single entity is clearly accountable). Four formal propositions establish boundary conditions for the framework's valid operating envelope, covering gradient, cascade compounding, delegation-depth effects, and extension sufficiency.