mlsanityData to 7 Oct 2026

ROZA Graphs: Self-Improving Near-Deterministic RAG through Evidence-Centric Feedback

arXiv 2604.07595 ↗First posted 8 Apr 2026cs.AI cs.CLcited by 0

On the map: LLM reasoning → Knowledge- and tool-augmented reasoning

also in 2 more
Abstract (arXiv)

Language model agents reason from scratch on every query, discarding their chain of thought after each run. The result is lower accuracy and high run-to-run variance. We introduce reasoning graphs, which persist the per-evidence chain of thought as structured edges. Unlike prior memory that retrieves distilled strategies by query similarity, reasoning graphs enable evidence-centric feedback: for every candidate item, the system traverses all incoming evaluation edges across prior runs to surface how that specific item has been judged before. We further introduce retrieval graphs, which feed a planner that prunes consistently-rejected candidates over successive runs. Together they form a ROZA graph: a self-improving feedback loop in which accuracy gains scale with gold-passage reuse (reasoning graph) and efficiency gains scale with candidate-pool overlap (retrieval graph). The base model remains frozen; all gains come from context engineering via graph traversal. We evaluate on MuSiQue and HotpotQA, plus a high-reuse deployment subset. Four findings stand out. (1) Dose-response: accuracy improves monotonically with evidence-profile coverage, reaching +10.6pp over Vanilla RAG at 50%+ coverage on the same questions (47% error reduction, p<0.0001; per-question Spearman ρ=+0.144, p<10-6, n=1,100). (2) Multi-hop scaling: 4-hop accuracy improves by +11.0pp (p=0.0001). (3) Cross-cluster prediction: the cluster-level gain is predicted by gold-passage reuse density (r=0.604, p=0.001, n=26 clusters). (4) High-reuse Pareto dominance: highest or tied-for-highest accuracy alongside 46% lower cost and 46% lower latency. Per-passage decision consistency across repeated runs (N=73 paired probes, K=10 runs each, two model families, three temperatures) rises by +8 to +13pp on a fixed 20-passage context and by +12 to +21pp when the retrieval graph also prunes (all p<0.005).

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Knowledge- and tool-augmented reasoning, one of 17 directions in LLM reasoning. Builds on 20 of its 28 references that are in the map. First posted on arXiv 8 Apr 2026.

Builds on

20 of 28 references.

Closest ancestors · 1

Same direction, newest first

  1. 2023Retrieval-Augmented Generation for Large Language Models: A Survey18 Dec

Other references · 19

  1. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,160
  2. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
  3. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
  4. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,686
  5. 2023Generative Agents: Interactive Simulacra of Human Behaviorcited by 3,500
  6. 2023Voyager: An Open-Ended Embodied Agent with Large Language Modelscited by 2,081
  7. 2023Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflectioncited by 1,800
  8. 2024From Local to Global: A Graph RAG Approach to Query-Focused Summarizationcited by 1,411
  9. 2020Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Stepscited by 1,259
  10. 2021MuSiQue: Multihop Questions via Single-hop Question Compositioncited by 893
  11. 2023Cognitive Architectures for Language Agentscited by 422
  12. 2024Graph Retrieval-Augmented Generation: A Surveycited by 320
7 more
  1. 2025ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memorycited by 253
  2. 2025MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agentscited by 226
  3. 2025LinearRAG: Linear Graph Retrieval Augmented Generation on Large-scale Corporacited by 47
  4. 2025When to use Graphs in RAG: A Comprehensive Analysis for Graph Retrieval-Augmented Generationcited by 46
  5. 2026Confident and Wrong: Silent Semantic Failures in Coding Agentscited by 3
  6. 2026Evaluate-as-Action: Self-Evaluated Process Rewards for Retrieval-Augmented Agentscited by 2
  7. 2026GAM-RAG: Gain-Adaptive Memory for Evolving Retrieval in Retrieval-Augmented Generationcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Why Retrieval-Augmented Generation Fails: A Graph Perspective
  2. 2026Text-Graph Synergy: A Bidirectional Verification and Completion Framework for RAG
  3. 2026Before Reasoning Can Fail: Pre-Evidence Procedural Failures in Agentic RAG
  4. 2025FAIR-RAG: Faithful Adaptive Iterative Refinement for Retrieval-Augmented Generation
  5. 2025Reasoning by Exploration: A Unified Approach to Retrieval and Generation over Graphs
  6. 2025Towards Agentic RAG with Deep Reasoning: A Survey of RAG-Reasoning Systems in LLMs
  7. 2025Retrieval is Not Enough: Enhancing RAG Reasoning through Test-Time Critique and Optimization
  8. 2025You Don't Need Pre-built Graphs for RAG: Retrieval Augmented Generation with Adaptive Reasoning Structures