Orchestrated Inference: Structured Reasoning Through Tool Composition in VDR-LLM-Prolog

The prior papers in this series built a language model architecture with exact arithmetic (VDR-1 through VDR-4), scoped knowledge bases with constraints and provenance (VDR-5), 333 deterministic primitives invoked through compact command tokens (VDR-6, VDR-8), a complete model lifecycle as KB operations (VDR-7), runtime data primitives for working memory (VDR-8), universal dotted-path addressing (VDR-8), and session snapshots with disposable cloning (VDR-8). Each paper added a capability. None of them specified how those capabilities compose into multi-step inference processes. This paper specifies Orchestrated Inference — the pattern by which the language model uses its tools to conduct structured investigations that produce traceable, quantified conclusions. The language model does not reason. It orchestrates. It selects and sequences exact tools — Prolog for logical deduction, Python for numerical computation, pure primitives for data manipulation, operational primitives for external data acquisition — in a loop that produces inferences neither the language model nor any single tool could produce alone. The paper defines the orchestrated inference loop (assess → formalize → execute → store → assess), inference notebooks as standard KB schemas for housing investigations, four inference modes (deductive, inductive, abductive, analogical) with their characteristic tool signatures, external data integration patterns for bringing real-world data into the exact system, and inference provenance that gives every conclusion a complete, queryable derivation chain with exact confidence scores. The central claim is precise: the language model predicts tokens; the tools compute and deduce; the composition produces structured inferences; the KB records everything. Orchestrated Inference is not artificial reasoning. It is a reasoning exoskeleton — external structure that compensates for the language model's computational unreliability while leveraging its strength at pattern recognition, intent mapping, and natural language formalization.