[
  {
    "id": "mast",
    "name": "MAST / MAST-Data",
    "title": "Why Do Multi-Agent LLM Systems Fail?",
    "source": "https://arxiv.org/abs/2503.13657v3",
    "version": "v3 · 2025-10-26",
    "kind": "Multi-agent traces",
    "scope": "1600+ annotated traces · 7 frameworks · 14 failure modes",
    "summary": "The authors organize multi-agent failures into system design, inter-agent misalignment and task verification categories.",
    "limit": "Author-reported abstract; traces were not downloaded or rerun here. These counts do not measure commercial demand.",
    "use": "Start here when selecting a vocabulary for discussing multi-agent failures. Check the paper and original artifacts before adopting labels.",
    "artifact_status": "Paper metadata only; no executable fixture verified.",
    "observed_on": "2026-10-21",
    "provenance": "Original short editorial summary of the linked author abstract; no full abstract or dataset reproduced."
  },
  {
    "id": "agenterrorbench",
    "name": "AgentErrorBench / AgentDebug",
    "title": "Where LLM Agents Fail and How They can Learn From Failures",
    "source": "https://arxiv.org/abs/2509.25370v1",
    "version": "v1 · 2025-09-29",
    "kind": "Failure trajectories",
    "scope": "ALFWorld · GAIA · WebShop",
    "summary": "The authors describe annotated failure trajectories and a debugging framework that identifies root causes and gives corrective feedback.",
    "limit": "Results are the authors’ report. Linked code/data availability and transfer to a business agent were not checked.",
    "use": "Start here for research on diagnosing a failure within an agent trajectory; inspect the original benchmark before making comparisons.",
    "artifact_status": "Paper metadata only; no executable fixture verified.",
    "observed_on": "2026-10-21",
    "provenance": "Original short editorial summary of the linked author abstract; no full abstract or dataset reproduced."
  },
  {
    "id": "framework-bugs",
    "name": "Agent framework quality study",
    "title": "Where Agent Frameworks Fall Short: Examining Functional Challenges and Usability Concerns",
    "source": "https://arxiv.org/abs/2602.21806v4",
    "version": "v4 · 2026-08-06",
    "kind": "Framework bug reports",
    "scope": "5,669 bug reports · 809 feature requests · 5 frameworks",
    "summary": "The study covers AutoGen, CrewAI, LangChain, LangGraph and MetaGPT, emphasizing interface and execution quality.",
    "limit": "Counts describe the latest abstract’s study scope, not reproduced failures, unique customers or an installed base.",
    "use": "Start here to investigate framework-level API and serialization boundaries. Use the original study for methods and sampling details.",
    "artifact_status": "Paper metadata only; no executable fixture verified.",
    "observed_on": "2026-10-21",
    "provenance": "Original short editorial summary of the linked author abstract; no full abstract or dataset reproduced."
  }
]