What question did this study set out to answer?

The central aim is to create a taxonomy that distinguishes different types of connections in knowledge graphs based on their origin to enhance trust.

February 8, 2026Open Access

The Tri-Color Trust Model: A Provenance-Typed Edge Taxonomy for Trustworthy AI-Augmented Knowledge Graphs

Key Points

The central aim is to create a taxonomy that distinguishes different types of connections in knowledge graphs based on their origin to enhance trust.
Developed the Tri-Color Trust Model to classify connections into three categories based on epistemic origin.
Implemented a provenance tracking system with audit trails for edge creation and modification.
Defined lifecycle transitions for edges that change based on user or validator actions.
Introduced path-aware conflict detection to track and explain reasoning chains causing conflicts.
Established a formal taxonomy that distinguishes between user-created, AI-suggested, and verifier-flagged edges.
Demonstrated the ability to track edge origins and transitions, improving transparency in knowledge graphs.
Enabled conflict detection through analysis of mixed-color subgraphs, clarifying how inconsistencies arise.

Abstract

This technical note specifies the Tri-Color Trust Model, a provenance-typed edge taxonomy for AI-augmented knowledge graphs. The model assigns every connection in a knowledge graph one of three colors based on its epistemic origin: Blue (explicit, user-authored), Green (neural, AI-suggested via semantic similarity), or Red (supervised, flagged by formal constraint validation). The taxonomy addresses a fundamental trust problem in AI-augmented knowledge management: users cannot distinguish between relationships they created, relationships an AI inferred, and relationships a formal verifier has flagged as inconsistent. This conflation leads to trust erosion, knowledge graph pollution, and uncritical acceptance of invalid AI suggestions. The Tri-Color Model provides: (1) a formal provenance tracking system with attribute-level audit trails recording creation source, modification history, and validation status; (2) lifecycle transitions defining how edges change color through user promotion, dismissal, or validator intervention; (3) integration with Colored Petri Net supervision where Red edges carry structured violation data from CPN guard functions; and (4) path-aware conflict detection that traces reasoning chains through mixed-color subgraphs to explain how conflicts propagate. The model is implemented in the AETHER-GRAPH system using React, Dexie.js (IndexedDB), and Transformers.js for client-side embeddings. This document establishes priority for the following novel contributions: (a) to our knowledge, the first formal provenance-typed edge taxonomy distinguishing user, neural, and symbolic origins in knowledge graphs; (b) lifecycle state machines governing edge color transitions with explicit promotion and dismissal semantics; (c) path-aware conflict detection through mixed-provenance subgraph traversal; and (d) integration of CPN guard functions as the formal mechanism generating Red (supervised) edges.

Read Full Paperexternally

AI से पूछें

Bookmark

View Full Paper