← Back to projects

Project

Cross-System Sequential Audit: Three-Agent Recursive Evaluation of Governance Framework Integrity

A governance framework subjected to recursive cross-system evaluation across three AI agents: Agatha (ChatGPT) performs first-order audit, Gemini performs second-order audit, Claude evaluates Gemini's evaluation. Three agents, three passes, recursive depth collapsing into a single auditable governance event.

2026 · Experimental protocol · Protocol designer and researcher

  • triangulation
  • three-agent audit
  • Agatha
  • Gemini
  • Claude
  • meta-evaluation
  • experimental governance
  • AI Governance Master Project
Three-Agent Triangulation Audit: cross-system recursive evaluation collapsing into one auditable event.

Experimental protocol

Method · Experimental Protocol — three agents, three passes, one auditable event

The Three-Agent Triangulation Audit applies the RDAIG method’s triangulation and comparison stage at cross-system scale. Three AI agents evaluate the same governance framework in sequence: Agatha (ChatGPT) performs first-order audit, Gemini performs second-order evaluation of Agatha’s findings, and Claude evaluates Gemini’s evaluation.

The design collapses three recursive passes into a single auditable event. The governance question is whether cross-system triangulation produces more stable findings than single-model recursion — and where the meta-evaluation layer itself introduces new drift, authority excess, or self-referential instability.