Corpus-Driven Expert Consultation via Regime-Isolated Multi-Interpreter Structural Synthesis

This paper presents an architectural mechanism for answering domain-expert queries against a knowledge corpus using multiple computationally independent interpretation agents operating under regime isolation. Unlike retrieval-augmented generation (RAG), multi-agent debate, and ensemble methods, the architecture introduces three novel mechanisms: (1) evidence-bound structural synthesis, where answer confidence derives from cross-interpreter agreement on shared corpus evidence rather than model probability scores or text similarity; (2) divergence as a first-class output, where contradictory interpretations of the same passage produce a structural ambiguity report rather than forced consensus; (3) regime-isolated consultation, where no agent sees any other agent's output, preventing the collapse of independent observation into social convergence. System intelligence resides in the architecture — not in any specific model. Models are interchangeable interpreters, not sources of truth. Formal basis derives from the Navigational Cybernetics 2.5 framework (Ax.15, Ax.22, Th.36–37, Th.67). Part of the PETRONUS research corpus.