Trust-Aware Multi-Agent Traceability: Confidence-Calibrated Knowledge Graphs for Consistent Software Artifact Management

The paper presents a trust-aware coordination framework for multi-agent AI systems in software engineering, utilizing a shared knowledge graph to enhance traceability and decision-making. It introduces a two-stage traceability link prediction pipeline that integrates embedding-based retrieval with LLM-based multi-criteria analysis, and features a consistency protocol for managing agent interactions through confidence thresholds and conflict resolution. The evaluation on an automotive software engineering case study demonstrates the importance of confidence calibration in improving link prediction accuracy and overall system reliability, which is critical for safety-critical applications.