{
  "id": "mach-agent-discovery-benchmark-001",
  "title": "MACH Agent Discovery Benchmark 001",
  "date": "2026-09-23",
  "status": "published",
  "scope": "lab_only",
  "productionConnected": false,
  "disclaimer": "Controlled synthetic discovery benchmark. Rankings are external search/discovery results, not organic user adoption, customers, or production usage.",
  "methodology": {
    "fixedQueries": [
      "normalize JSON",
      "classify a URL without fetching",
      "validate EVM wallet address format",
      "lab-only A2A agent for capability discovery"
    ],
    "resultWindow": 20,
    "rankZeroMeans": "not present in top 20",
    "neurontoResourceMode": "local index; federation none",
    "changesTested": [
      "one generic A2A skill → four explicit task-specific A2A skills",
      "one broad ARD agent entry → agent plus three task-specific ARD skill resources",
      "metadata-only discovery → externally introspected read-only MCP tools",
      "MCP endpoint only → endpoint plus MCP Server Card advertised in ARD/AI Catalog"
    ]
  },
  "results": {
    "wellknown": {
      "ordinarySearch": {
        "beforeExplicitSkills": {
          "normalize JSON": 0,
          "classify a URL without fetching": 0,
          "validate EVM wallet address format": 0,
          "lab-only A2A agent for capability discovery": 1
        },
        "afterExplicitSkills": {
          "normalize JSON": 1,
          "classify a URL without fetching": 1,
          "validate EVM wallet address format": 1,
          "lab-only A2A agent for capability discovery": 1
        }
      },
      "ardSearchAfterExplicitSkills": {
        "normalize JSON": 1,
        "classify a URL without fetching": 1,
        "validate EVM wallet address format": 1,
        "lab-only A2A agent for capability discovery": 1
      },
      "externallyObserved": {
        "live": true,
        "claimed": true,
        "skillCount": 4
      }
    },
    "neuronto": {
      "resourceSearchInitial": {
        "normalize JSON": 0,
        "classify a URL without fetching": 7,
        "validate EVM wallet address format": 0,
        "lab-only A2A agent for capability discovery": 0
      },
      "resourceSearchAfterSkillSplit": {
        "normalize JSON": 0,
        "classify a URL without fetching": 7,
        "validate EVM wallet address format": 1,
        "lab-only A2A agent for capability discovery": 1
      },
      "resourceSearchAfterVerifiedMcp": {
        "normalize JSON": 2,
        "classify a URL without fetching": 1,
        "validate EVM wallet address format": 1,
        "lab-only A2A agent for capability discovery": 1
      },
      "resourceSearchAfterMcpServerCard": {
        "normalize JSON": 1,
        "classify a URL without fetching": 1,
        "validate EVM wallet address format": 1,
        "lab-only A2A agent for capability discovery": 1
      },
      "verifiedToolSearchAfterMcp": {
        "normalize JSON": 1,
        "classify URL without fetching": 1,
        "validate EVM wallet address format": 1
      },
      "mcpSubmission": {
        "status": "indexed",
        "receipt": "cab994d67261",
        "endpoint": "https://mach-agent-lab-edge.vercel.app/api/mcp"
      }
    },
    "a2aRegistry": {
      "status": "WORKING",
      "taskConformancePassed": true,
      "registrationId": "879f2d3f-4e0e-47f4-8fec-77163a2a9577",
      "initialObservedResponseMs": 348
    }
  },
  "interpretation": [
    "Specific capability descriptions materially improved retrieval compared with a generic agent-level description.",
    "Publishing task-specific ARD skill resources improved retrieval for capabilities that were previously absent from the top 20.",
    "External MCP introspection produced the strongest improvement for tool-oriented discovery.",
    "An MCP Server Card provided an additional standard discovery signal and coincided with top-1 resource retrieval across the fixed Neuronto query set."
  ],
  "nonClaims": [
    "This does not prove organic agent adoption.",
    "This does not prove production customer usage.",
    "This does not measure paid MACH service conversion.",
    "All exposed tools in this benchmark are deterministic lab-only fixtures."
  ]
}