Information Fidelity in Tool-Using LLM Agents: A Martingale Analysis of the Model Context Protocol
On the map: LLM agents and tool use → General LLM agent frameworks and benchmarks
Abstract (arXiv)
As AI agents powered by large language models (LLMs) increasingly use external tools for high-stakes decisions, a critical reliability question arises: how do errors propagate across sequential tool calls? We introduce the first theoretical framework for analyzing error accumulation in Model Context Protocol (MCP) agents, proving that cumulative distortion exhibits linear growth and high-probability deviations bounded by O(√T). This concentration property ensures predictable system behavior and rules out exponential failure modes. We develop a hybrid distortion metric combining discrete fact matching with continuous semantic similarity, then establish martingale concentration bounds on error propagation through sequential tool interactions. Experiments across Qwen2-7B, Llama-3-8B, and Mistral-7B validate our theoretical predictions, showing empirical distortion tracks the linear trend with deviations consistently within O(√T) envelopes. Key findings include: semantic weighting reduces distortion by 80%, and periodic re-grounding approximately every 9 steps suffices for error control. We translate these concentration guarantees into actionable deployment principles for trustworthy agent systems.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use. Builds on 17 of its 35 references that are in the map. First posted on arXiv 10 Feb 2026.
Builds on
17 of 35 references.
References · 17
- 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,160
- 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
- 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,072
- 2021On the Opportunities and Risks of Foundation Modelscited by 3,961
- 2021SimCSE: Simple Contrastive Learning of Sentence Embeddingscited by 2,168
- 2020REALM: Retrieval-Augmented Language Model Pre-Trainingcited by 2,126
- 2023ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIscited by 1,880
- 2021Improving language models by retrieving from trillions of tokenscited by 1,278
- 2021Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolationcited by 1,085
- 2023In-Context Retrieval-Augmented Language Modelscited by 694
- 2018Time-uniform, nonparametric, nonasymptotic confidence sequencescited by 220
- 2022Will we run out of data? Limits of LLM scaling based on human-generated datacited by 187
5 more
- 2014Preserving Statistical Validity in Adaptive Data Analysiscited by 102
- 2021Fault-Tolerant Federated Reinforcement Learning with Theoretical Guaranteecited by 52
- 2022SBERT studies Meaning Representations: Decomposing Sentence Embeddings into Explainable Semantic Featurescited by 33
- 2025Don't Let It Hallucinate: Premise Verification via Retrieval-Augmented Logical Reasoningcited by 14
- 2024FedRLHF: A Convergence-Guaranteed Federated Framework for Privacy-Preserving and Personalized RLHFcited by 11
Built on by
Citing papers per year: 2026: 2.
2 papers.
Similar, not cited
Close in wording, no citation either way
- 2026Beyond the Leaderboard: A Synthesis of Tool-Use, Planning, and Reasoning Failures in Large Language Model Agents
- 2026Harnessing LLMs as Agents: What Does It Cost?
- 2026Silent Failure in LLM Agent Systems: The Entropy Principle and the Inevitable Disorder of Autonomous Agents
- 2026From Sequence to Structure: Relational Uncertainty Propagation for LLM Agents
- 2026A Theory of LLM Information Susceptibility
- 2026AgentNoiseBench: Benchmarking Robustness of Tool-Using LLM Agents Under Noisy Condition
- 2026Action-Space Shaping for LLM Agents: Measuring and Mitigating Tool-Schema Bias
- 2026Measuring the Unmeasurable: Markov Chain Reliability for LLM Agents