Diskurs is a multi-agent reasoning architecture for decision assurance in AI-supported decision processes. It separates generation, evaluation, and decision into distinct layers. Multiple locally operated language models generate and revise typed claims, which are represented in a claim graph. A deterministic, rule-based decision layer derives one of five documented statuses—CONFIRMED, LIMITED, REVIEW, REJECTED, or UNDECIDABLE—without language-model involvement in the final status decision. The record also documents the Completeness Dimension, an underspecification audit of the input question. It identifies missing necessary prerequisites and critical ambiguities as typed gap claims whose rule-based effects can only tighten the decision status. This record contains English and German working papers, a German scientific positioning paper, and two concept figures. It documents revision 2026-07b, aligned with decision core v0.4.1. The architecture is implemented as a research prototype and operational in the documented environment. Empirical validation, calibration, benchmarking, and ablation studies remain pending. No source code, prompts, canonical specification, calibration data, calibrated parameter set, model weights, test data, production configuration, or integration know-how are included.
Bernd Kladders (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: