MAP-Graph: Provenance-Aware Shared Memory for Multi-Agent Workflows

Authors: Yiqi Wang, Zihao Yan, Jiaqi Zhang, Zhangkai Wu, Mingkai Zheng, Zequn Sun, Yanming Zhu and Taotao Cai. Venue/status: arXiv preprint, submitted 11 August 2026. Problem: summaries can hide private, poisoned, revoked or untrusted sources. Contribution: typed provenance graph, permission filtering, path-trust ranking and risk-sensitive action gating. Practical significance: provenance becomes a run-time authorisation signal. Limitations: 2,700 synthetic tasks per method across three domains; no independent or production reproduction. Paper