{
  "schema_version": 1,
  "builder_plan": {
    "scenario": "An agent looks good in demos, but you cannot tell which product tasks it fails.",
    "change": "Use the study's evaluation idea to inspect a small set of tasks with human-checked answers.",
    "baseline": "Your current evaluator and a manually reviewed reference set.",
    "steps": [
      "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.",
      "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.",
      "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.",
      "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test."
    ],
    "measure": "Agreement with human review, missed failures, false alarms, and evaluation cost per task.",
    "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation."
  },
  "builder_review": null,
  "arxiv_id": "2610.23449",
  "title": "GatewayBench: Unified Capability and Security Evaluation Platform for MCP Gateways",
  "summary": "GatewayBench replaces ad hoc or absent evaluation of MCP gateway middleware with a standardized benchmark of 524 tasks and 11,118 real tools, comparing flattened and hierarchical gateway architectures under controlled tool-surface scaling. Adopting it costs engineering time to integrate the benchmark harness with the team's gateway and agent stack, plus API or inference spend to run the task suite; it provides evaluation only and improves no production metric by itself. Results can fail to transfer because gateway rankings interact strongly with the underlying model and agent harness, so conclusions drawn with one model may not hold for the reader's production configuration.",
  "rationale": "For a team deploying agents behind an MCP gateway, this benchmark is directly usable at small scale via APIs, but its conclusions are model- and harness-dependent, so the reader must validate gateway choice on their own workload rather than adopt the paper's rankings wholesale.",
  "technique": "Benchmark for evaluating MCP gateway designs",
  "family": "avaliacao_de_agente",
  "recommendation": "testar",
  "published": "2026-10-16",
  "as_of": "2026-10-19",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.23449",
  "independent_implementations": 2,
  "total_implementations": 2,
  "stars": 1695,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "GatewayBench replaces ad hoc or absent evaluation of MCP gateway middleware with a standardized benchmark of 524 tasks and 11,118 real tools, comparing flattened and hierarchical gateway architectures under controlled tool-surface scaling. Adopting it costs engineering time to integrate the benchmark harness with the team's gateway and agent stack, plus API or inference spend to run the task suite; it provides evaluation only and improves no production metric by itself. Results can fail to transfer because gateway rankings interact strongly with the underlying model and agent harness, so conclusions drawn with one model may not hold for the reader's production configuration.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.23449",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "exposure_map": [
    {
      "dimension": "quality",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compute",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "latency",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "operations",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compatibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "security",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "data_and_training",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "reproducibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "risks": [],
  "minimum_test": [],
  "open_questions": [],
  "independent_tests": [],
  "lab_reproductions": [],
  "report_available": false,
  "validation_tier": "unknown",
  "evidence_tier": "unknown",
  "infrastructure_basis": "unknown",
  "equations": [],
  "equations_status": "not_fetched",
  "core_kind": "",
  "equations_fetched_at": "",
  "equations_source_url": "",
  "equations_source_sha256": "",
  "exposures_reviewed_at": "",
  "exposures_source_sha256": ""
}