Black-box Reasoning Path Planning Protocol: A Methodology for Reasoning Structure Integrity Auditing

Abstract

Aiming at the problems of insufficient interpretability and difficult verification of reasoning structures in black-box reasoning systems, this paper proposes a Zetetic Abductive Audit Engine (ZAAE)—a black-box reasoning path planning protocol for auditing reasoning structure integrity. By defining standardized forward reasoning and backward completion interfaces, the protocol constructs a diverse and redundant set of reasoning paths, and combines stripping tests to realize quantitative auditing of reasoning structure integrity. A relevance-diversity balance mechanism is introduced to ensure the correlation between paths and conclusions while improving the diversity of path combinations, providing a path-level auditing basis for subsequent cross-validation. Through core indicators such as Premise Compliance Rate (PCR) and breakpoint sets, multi-dimensional diagnosis of black-box reasoning behavior from macro to micro is realized. Case analysis shows that the protocol can successfully locate structural breakpoints in reasoning paths, verifying its logical feasibility and effectiveness as a reasoning structure auditing tool. The diversity penalty mechanism helps improve path coverage, providing reliable support for the credibility evaluation of black-box reasoning results.

Keywords