Facebook pixel tracking

The Science and Information (SAI) Organization publishes open-access peer-reviewed journals in computer science and artificial intelligence.

Contact Info
Website thesai.org
Follow Us
Contact Info
Follow Us
Research Article | Open Access |

Read-First and Role-Gated Context Federation for Multi-Agent LLM Systems: A Controlled Multi-Model Comparison with Point-to-Point Messaging

Author 1: Ryan Valdivia-Segovia Author 2: Adrian Mamani-Quispe Author 3: Karim Guevara Author 4: Norka Bedregal-Alpaca
International Journal of Advanced Computer Science and Applications (IJACSA) · Vol. 17, No. 10 · Published 2026

DOI: 10.14569/IJACSA.2026.0171094

Abstract

Multi-agent systems based on large language models (LLMs) commonly rely on point-to-point message-passing, which increases integration complexity and leaves fine-grained access control and traceability to additional mechanisms. This study evaluates the Graph Context Protocol (GCP), a read-first context-federation approach with role- and capability-based access control and native per-read provenance. GCP is compared with two Agent2Agent (A2A) arms: the base protocol as shipped, and an instrumented variant whose policy enforcement point applies the same role and delegation policy and writes an audit log. Agent logic, models, tasks, and evaluation criteria are held constant; only the interoperability layer changes. Five behavioral scenarios, one scoring semantic leakage of a useful but forbidden node, were run on six models (three local, three hosted; 2,475 runs, none failed). A structural analysis shows that a hub-style GCP needs O(n) integration edges against O(n2) for a full A2A mesh, but A2A through an orchestrator is also O(n), so this reflects topology, not protocol. GCP denied all 443 unauthorized attempts and instrumented A2A all 408, whereas base A2A executed all 392; the instrumented arm matched GCP on containment, provenance, and leakage in every model, although a hermetic test shows that a caller omitting its operation declaration bypasses its delegation rule. The governance gap thus reflects the presence of an enforcement point and audit sink, which GCP provides natively and A2A requires each deployment to add. Token cost is comparable between GCP and instrumented A2A; task success is indistinguishable for four of six models and lower under GCP for the smallest hosted model, where the cause is the wording of the tool description. A software-org gap for llama3.1:8b did not reproduce on rerun.

Keywords

How to Cite this Article

Valdivia-Segovia, R., Mamani-Quispe, A., Guevara, K., & Bedregal-Alpaca, N. (2026). Read-First and Role-Gated Context Federation for Multi-Agent LLM Systems: A Controlled Multi-Model Comparison with Point-to-Point Messaging. International Journal of Advanced Computer Science and Applications, 17(10). https://doi.org/10.14569/IJACSA.2026.0171094

Valdivia-Segovia, Ryan, et al.. "Read-First and Role-Gated Context Federation for Multi-Agent LLM Systems: A Controlled Multi-Model Comparison with Point-to-Point Messaging." International Journal of Advanced Computer Science and Applications, vol. 17, no. 10, 2026, https://doi.org/10.14569/IJACSA.2026.0171094.

@article{Valdivia-Segovia2026,
  title     = {Read-First and Role-Gated Context Federation for Multi-Agent LLM Systems: A Controlled Multi-Model Comparison with Point-to-Point Messaging},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {10},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Ryan Valdivia-Segovia and Adrian Mamani-Quispe and Karim Guevara and Norka Bedregal-Alpaca},
  doi       = {10.14569/IJACSA.2026.0171094},
  url       = {https://doi.org/10.14569/IJACSA.2026.0171094}
}

Open Access — licensed under a Creative Commons Attribution 4.0 International License. Unrestricted use, distribution, and reproduction in any medium, even commercially, as long as the original work is properly cited.

Call for Papers - Important Dates

Volume 17 No. 12, December 2026
Paper Submission Due November 25, 2026
Review Notification December 15, 2026
Publication Date December 31, 2026