Charter Coherence Protocol: Eliminating Context Rot in Long-Running Autonomous Multi-Agent Enterprise Systems is an open-access, peer-reviewed research paper by Gangadevi Jeyaraj*; James Joseph Raj; Jayasree S; Vishnu Pookkattukunnath Baburaj, published in Volume 15, Issue 9 of the International Journal of Advanced Research in Science and Technology (IJARST), a UGC-approved journal (Print ISSN 2319-1783, Online ISSN 2320-1126).
Gangadevi Jeyaraj*; James Joseph Raj; Jayasree S; Vishnu Pookkattukunnath Baburaj
Enterprises are handing claims, payments, procurement and compliance workflows to autonomous multi-agent systems (MAS) built on large language models, and the policy rules that bound those agents decay as the run grows: current measurements put compliance at 32% after fifty turns, policy violation at 30% after one context summary, and unauthorised actions at up to 20% after a single handoff. We call this decay context rot, locate it at four points of the agent stack, and present the Charter Coherence Protocol (CCP), which eliminates it where it matters. In a simulated enterprise MAS with every form of rot injected, CCP brings rule violations executed at tools from 60–79% of binding steps to zero in every scenario, cuts the violations the model proposes from 60–79% to 6.5–14%, and removes laundered paraphrases, mutated thresholds, stale actions and compaction evictions entirely, at 1.2–1.7 times the baseline token cost. CCP achieves this by treating a governing clause as authority: clauses live in a signed, versioned Charter DB; the runtime places them in a locked window segment by scope; a probe-and-refresh controller measures adherence per clause and tunes the refresh window with the additive-increase/multiplicative-decrease control of TCP; a gate checks every tool call against the full live charter; a directory keeps the copies held by concurrent sessions coherent when a clause changes; and a three-way handshake passes clause identities between agents and transfers obligations on acknowledgement. An enterprise adopts CCP by repointing its agents at the CCP runtime; the orchestration framework, memory store and model stay as they are. The protocol is specified as one algorithm, four state machines and a message sequence, and the reference implementation, charter definitions and test environment are released with this paper.
Keywords: Autonomous agents, context rot, enterprise automation, large language models, long-horizon agents, multi-agent systems, policy compliance.
DOI: https://doi.org/10.62226/ijarst20260929
[1] Anthropic, “Model Context Protocol: Architecture overview,” MCP Specification, 2026. [Online]. Available: https://modelcontextprotocol.io/
[2] “LLM-based agents for cybersecurity: A systematic review of architectures, applications, and open challenges,” J. Cybersecur. Privacy, vol. 6, no. 5, p. 159, 2026, doi: 10.3390/jcp6050159.
[3] P. Laban, H. Hayashi, Y. Zhou, and J. Neville, “LLMs get lost in multi-turn conversation,” in Proc. Int. Conf. Learn. Represent. (ICLR), 2026. [Online]. Available: https://proceedings.iclr.cc/paper_files/paper/2026/file/59f6421e64707225fdf5b28840679a07-Paper-Conference.pdf
[4] P. Robinette et al., “We are what we repeatedly do: Improving long context instruction following,” in Findings Assoc. Comput. Linguistics: EACL, 2026. [Online]. Available: https://aclanthology.org/2026.findings-eacl.254/
[5] S. Chen, “Governance decay: How context compaction silently erases safety constraints in long-horizon LLM agents,” arXiv:2606.22528, 2026.
[6] Y. Hu, Z. Wang, and J. McAuley, “MemoryAgentBench,” in Proc. Int. Conf. Learn. Represent. (ICLR), 2026. [Online]. Available: https://proceedings.iclr.cc/paper_files/paper/2026/file/fd1eff9dd295df50a41f2521942fa31d-Paper-Conference.pdf
[7] Z. Xu, X. Zhang, H. Luo, Y. Jin, Y. Dong, and H. Salam, “MasDrift: Benchmarking authorization preservation across multi-agent architectures,” arXiv:2605.10481, 2026.
[8] A. Menon et al., “Inherited goal drift: Contextual pressure can undermine agentic goals,” arXiv:2603.03258, 2026.
[9] J. Martin and F. Roger, “Classifier context rot: Monitor performance degrades with context length,” arXiv:2605.12366, 2026.
[10] A. Dongre et al., “When attention closes: How LLMs lose the thread in multi-turn interaction,” arXiv:2605.12922, 2026.
[11] H. Zhao et al., “AMA-Bench: Evaluating long-horizon memory for agentic applications,” in Proc. Int. Conf. Learn. Represent. (ICLR), 2026.
[12] Y. Shen et al., “Mem2ActBench,” in Proc. 64th Annu. Meeting Assoc. Comput. Linguistics (ACL), 2026. [Online]. Available: https://aclanthology.org/2026.acl-long.370/
[13] “Faithful, not corrective: Structured handoffs localise but do not repair errors in multi-hop LLM relays,” arXiv:2607.09678, 2026.
[14] D. Louck, “Laundering attacks on agent memory and origin-bound authority (TMA-NM),” arXiv:2606.24322, 2026.
[15] V. Jacobson, “Congestion avoidance and control,” in Proc. ACM SIGCOMM, 1988, pp. 314–329.
[16] D.-M. Chiu and R. Jain, “Analysis of the increase and decrease algorithms for congestion avoidance in computer networks,” Comput. Netw. ISDN Syst., vol. 17, no. 1, pp. 1–14, 1989.
[17] D. J. Sorin, M. D. Hill, and D. A. Wood, A Primer on Memory Consistency and Cache Coherence. San Rafael, CA, USA: Morgan & Claypool, 2011.
[18] N. F. Liu et al., “Lost in the middle: How language models use long contexts,” Trans. Assoc. Comput. Linguistics, vol. 12, pp. 157–173, 2024.
[19] A. Byerly and D. Khashabi, “Self-consistency falls short! The adverse effects of positional bias on long-context problems,” Trans. Assoc. Comput. Linguistics, vol. 14, 2026. [Online]. Available: https://aclanthology.org/2026.tacl-1.15/
[20] Z. Xu et al., “When does divide and conquer work for long context LLM? A noise decomposition framework,” in Proc. Int. Conf. Learn. Represent. (ICLR), 2026.
[21] S. Lee, Y. Jo, M. Seo, M. Lee, and M. Seo, “Lost in the noise: How reasoning models fail with contextual distractors,” arXiv:2601.07226, 2026.
[22] J. Wu et al., “GAM: Hierarchical graph-based agentic memory,” in Proc. 64th Annu. Meeting Assoc. Comput. Linguistics (ACL), 2026. [Online]. Available: https://aclanthology.org/2026.acl-long.1600/
[23] H. Yu et al., “Agentic memory (AgeMem),” in Proc. 64th Annu. Meeting Assoc. Comput. Linguistics (ACL), 2026. [Online]. Available: https://aclanthology.org/2026.acl-long.981/
[24] Z. Zhou et al., “MEM1: Learning to synergize memory and reasoning for efficient long-horizon agents,” in Proc. Int. Conf. Learn. Represent. (ICLR), 2026.
[25] C. Packer et al., “MemGPT: Towards LLMs as operating systems,” arXiv:2310.08560, 2023.
[26] J. Lee and W.-S. Li, “Graph-of-Context: Recoverable fold/unfold and scoped fork for instrumented long-horizon multi-turn LLM agentic workflows,” IEEE Access, vol. 14, 2026.
[27] Z. Xiong et al., “How memory management impacts LLM agents: Experience-following behavior,” in Proc. 64th Annu. Meeting Assoc. Comput. Linguistics (ACL), 2026. [Online]. Available: https://aclanthology.org/2026.acl-long.27/
[28] W. Lam, Y. Li, Q. Zhang, and J. Zhao, “Governing evolving memory in LLM agents: The SSGM framework,” arXiv:2603.11768, 2026.
[29] D. Margalit et al., “Governed shared memory for multi-agent LLM systems,” arXiv:2606.24535, 2026.
[30] “MAS-Guard: Agent charters and a control plane for multi-agent LLM systems,” arXiv:2610.28397, 2026.
[31] M. Cemri et al., “Why do multi-agent LLM systems fail?” in Adv. Neural Inf. Process. Syst. (NeurIPS), 2025.
[32] Y. Xu and H. Wu, “Mitigating LLM hallucination snowballing in multiagent systems via context-aware semantic consistency reasoning,” IEEE Trans. Neural Netw. Learn. Syst., 2026, doi: 10.1109/TNNLS.2026.3655508.
[33] S. Jamshidi, A. Moradi Dakhel, N. Nafi, and F. Khomh, “Hallucination cascade: Analyzing error propagation in multi-agent LLM systems,” arXiv:2606.07937, 2026.
[34] A. Singh and R. Pawar, “The hallucination snowball: Modeling error propagation as state transitions in multi-agent LLM pipelines,” arXiv:2608.14588, 2026.
[35] A. Mazhar, S. Suri, and S. Galhotra, “Trace-level analysis of information contamination in multi-agent systems,” in Proc. ACM Conf. AI Syst. (CAIS), 2026.
[36] M. Alhwikem, S. Khan, and F. Jaskani, “SPI-Net: Prompt-injection-resilient data integrity verification for agentic LLM networks via cryptographic attestation and activation monitoring,” Symmetry, vol. 18, no. 9, p. 1565, 2026.
[37] H. Wang, C. M. Poskitt, and J. Sun, “AgentSpec: Customizable runtime enforcement for safe and reliable LLM agents,” in Proc. 48th IEEE/ACM Int. Conf. Softw. Eng. (ICSE), 2026.
[38] Y. Lin et al., “VIGIL: Defending LLM agents against tool stream injection via verify-before-commit,” in Proc. 64th Annu. Meeting Assoc. Comput. Linguistics (ACL), 2026. [Online]. Available: https://aclanthology.org/2026.acl-long.443/
[39] N. Kekatos et al., “Formal runtime verification for tool-using LLM agents: An offline same-benchmark study on AgentDojo and STAC,” arXiv:2610.09793, 2026.
[40] K. Joshi, T. Finin, A. Joshi, and L. Kagal, “Deontic policies for runtime governance of LLM agents,” arXiv:2606.19464, 2026.
[41] G. Zhang et al., “AgenTracer: Who is inducing failure in the LLM agentic systems?” in Proc. Int. Conf. Learn. Represent. (ICLR), 2026.
[42] X. Li et al., “MARCH: Multi-agent reinforced check for hallucination,” in Proc. 64th Annu. Meeting Assoc. Comput. Linguistics (ACL), 2026. [Online]. Available: https://aclanthology.org/2026.acl-long.1828/
[43] J. Li, V. Magesh, and V. V. Veeravalli, “Principled detection of hallucinations in LLMs via multiple testing,” in Findings Assoc. Comput. Linguistics: ACL, 2026. [Online]. Available: https://aclanthology.org/2026.findings-acl.1705/
[44] J. Liu, B. Jaiyen, R. Veras, and O. Mutlu, “RAIDR: Retention-aware intelligent DRAM refresh,” in Proc. 39th Annu. Int. Symp. Comput. Archit. (ISCA), 2012, pp. 1–12.
[45] M. K. Qureshi, D.-H. Kim, S. Khan, P. J. Nair, and O. Mutlu, “AVATAR: A variable-retention-time (VRT) aware refresh for DRAM systems,” in Proc. 45th Annu. IEEE/IFIP Int. Conf. Dependable Syst. Netw. (DSN), 2015, pp. 427–437.
[46] J. Woodruff et al., “The CHERI capability model: Revisiting RISC in an age of risk,” in Proc. 41st Annu. Int. Symp. Comput. Archit. (ISCA), 2014, pp. 457–468.
[47] J. H. Saltzer, D. P. Reed, and D. D. Clark, “End-to-end arguments in system design,” ACM Trans. Comput. Syst., vol. 2, no. 4, pp. 277–288, 1984.
[48] S. Murali, T. Theocharides, N. Vijaykrishnan, M. J. Irwin, L. Benini, and G. De Micheli, “Analysis of error recovery schemes for networks on chips,” IEEE Des. Test Comput., vol. 22, no. 5, pp. 434–442, 2005.
10.62226/ijarst20260929
PAGES : 2681-2695 | 9 VIEWS | 4 DOWNLOADS
Gangadevi Jeyaraj*; James Joseph Raj; Jayasree S; Vishnu Pookkattukunnath Baburaj — “Charter Coherence Protocol: Eliminating Context Rot in Long-Running Autonomous Multi-Agent Enterprise Systems.” International Journal of Advanced Research in Science and Technology (IJARST), Volume 15, Issue 9. DOI: 10.62226/ijarst20260929.
Gangadevi Jeyaraj*; James Joseph Raj; Jayasree S; Vishnu Pookkattukunnath Baburaj | Charter Coherence Protocol: Eliminating Context Rot in Long-Running Autonomous Multi-Agent Enterprise Systems | DOI : 10.62226/ijarst20260929
| Journal Frequency: | ISSN 2320-1126, Monthly | |
| Paper Submission: | Throughout the month | |
| Acceptance Notification: | Within 6 days | |
| Subject Areas: | Engineering, Science & Technology | |
| Publishing Model: | Open Access | |
| Publication Fee: | USD 60 USD 50 | |
| Publication Impact Factor: | 6.76 | |
| Certificate Delivery: | Digital |