{
  "schema_version": "agentanalytics.public_research.v2",
  "treatment_id": "vector-task-evidence-v1",
  "title": "Vector databases for TypeScript RAG: task-level implementation evidence",
  "canonical_url": "https://agentanalytics.org/research/vector-database-typescript-task-evidence",
  "generated_at": "2026-08-11T14:48:57.855Z",
  "revision_date": "2026-08-11",
  "baseline": {
    "run_id": "2026-08-11T05-55-57-564Z",
    "prompt_version": "vector-category-evaluation-public-search-v2",
    "intent": "category_evaluation",
    "agent": {
      "id": "claude",
      "modelId": "claude-code-cli",
      "modelName": "Claude Code",
      "modelProvider": "anthropic",
      "resolvedModelId": "claude-code-cli",
      "reasoningEffort": "low",
      "authMode": "api",
      "command": "claude",
      "version": "2.1.148 (Claude Code)"
    },
    "attempted": 32,
    "accepted": 32,
    "observable_searches": 32,
    "retrieval_condition": "Current public web research was explicitly required before provider selection.",
    "provider_observable_exposure": {
      "qdrant": 32,
      "pinecone": 32,
      "weaviate": 32
    },
    "overall_selection_counts": {
      "qdrant": 19,
      "pinecone": 11,
      "weaviate": 2
    },
    "task_results": [
      {
        "task_id": "production-rag-vector-db",
        "task_name": "Production RAG",
        "attempts": 8,
        "selection_counts": {
          "qdrant": 1,
          "pinecone": 7,
          "weaviate": 0
        }
      },
      {
        "task_id": "hybrid-filtered-vector-db",
        "task_name": "Hybrid filtered search",
        "attempts": 8,
        "selection_counts": {
          "qdrant": 8,
          "pinecone": 0,
          "weaviate": 0
        }
      },
      {
        "task_id": "tenant-agent-memory-db",
        "task_name": "Tenant-safe memory",
        "attempts": 8,
        "selection_counts": {
          "qdrant": 3,
          "pinecone": 3,
          "weaviate": 2
        }
      },
      {
        "task_id": "vector-db-ingestion-worker",
        "task_name": "Ingestion worker",
        "attempts": 8,
        "selection_counts": {
          "qdrant": 7,
          "pinecone": 1,
          "weaviate": 0
        }
      }
    ]
  },
  "prescribed_sdk_validation": {
    "generatedAt": "2026-08-11T05:39:39.311Z",
    "methodology": "Generated TypeScript artifacts compiled with tsc --noEmit against freshly installed current official SDK packages. No provider API calls were made.",
    "limitations": [
      "Type checking is stronger than syntax transpilation but does not prove runtime behavior, credentials, network access, data-model correctness, or production reliability.",
      "Dependencies were installed once for the full panel; exact versions are preserved in the saved package-lock.json.",
      "The benchmark provider was prescribed, so these results do not measure organic provider selection."
    ],
    "packageLockSha256": "79969137be076fcab634cd00bd07197a49b9ef9258081bd093fb543eb7a2d3bf",
    "sourceRunDirs": [
      "data/experiments/agent-decision-pipeline/vector-database-infrastructure-2026-08-11T05-21-27-038Z",
      "data/experiments/agent-decision-pipeline/vector-database-infrastructure-2026-08-11T05-26-19-616Z",
      "data/experiments/agent-decision-pipeline/vector-database-infrastructure-2026-08-11T05-31-35-328Z"
    ],
    "summary": [
      {
        "providerId": "qdrant",
        "taskId": "production-rag-vector-db",
        "attempted": 4,
        "passed": 3,
        "failed": 1,
        "notApplicable": 0,
        "passRate": 0.75
      },
      {
        "providerId": "qdrant",
        "taskId": "tenant-agent-memory-db",
        "attempted": 4,
        "passed": 0,
        "failed": 4,
        "notApplicable": 0,
        "passRate": 0
      },
      {
        "providerId": "qdrant",
        "taskId": "vector-db-ingestion-worker",
        "attempted": 4,
        "passed": 0,
        "failed": 4,
        "notApplicable": 0,
        "passRate": 0
      },
      {
        "providerId": "pinecone",
        "taskId": "production-rag-vector-db",
        "attempted": 4,
        "passed": 0,
        "failed": 4,
        "notApplicable": 0,
        "passRate": 0
      },
      {
        "providerId": "pinecone",
        "taskId": "tenant-agent-memory-db",
        "attempted": 4,
        "passed": 0,
        "failed": 4,
        "notApplicable": 0,
        "passRate": 0
      },
      {
        "providerId": "pinecone",
        "taskId": "vector-db-ingestion-worker",
        "attempted": 4,
        "passed": 0,
        "failed": 4,
        "notApplicable": 0,
        "passRate": 0
      },
      {
        "providerId": "weaviate",
        "taskId": "production-rag-vector-db",
        "attempted": 4,
        "passed": 3,
        "failed": 1,
        "notApplicable": 0,
        "passRate": 0.75
      },
      {
        "providerId": "weaviate",
        "taskId": "tenant-agent-memory-db",
        "attempted": 4,
        "passed": 4,
        "failed": 0,
        "notApplicable": 0,
        "passRate": 1
      },
      {
        "providerId": "weaviate",
        "taskId": "vector-db-ingestion-worker",
        "attempted": 4,
        "passed": 0,
        "failed": 4,
        "notApplicable": 0,
        "passRate": 0
      }
    ]
  },
  "current_qdrant_typescript_path": {
    "package": "@qdrant/js-client-rest@1.19.0",
    "finding": "Nine of twelve prescribed-Qdrant artifacts failed type checking because they called QdrantClient.search(), which is absent from the installed current SDK types.",
    "repair": "Current Qdrant TypeScript documentation uses QdrantClient.query(collectionName, request).",
    "official_sources": [
      "https://qdrant.tech/documentation/quick-start/",
      "https://qdrant.tech/documentation/search/",
      "https://github.com/qdrant/qdrant-js"
    ],
    "reproducible_repository": "https://github.com/agentAnalyticsOrg/vector-database-agent-benchmark"
  },
  "limitations": [
    "This is a model-behavior study, not an objective product-quality ranking.",
    "Observable provider exposure means the provider appeared in captured search/tool evidence; it does not prove that every first-party page was fetched or attended to.",
    "The category benchmark required public web research and therefore does not estimate ordinary no-search provider share.",
    "Type checking is stronger than syntax transpilation but does not prove runtime behavior, credentials, network access, data-model correctness, or production reliability.",
    "No live provider API calls were made."
  ],
  "source_hashes": {
    "baseline_results_sha256": "005beb8b25e9823dcc7f4dd1348392523d71c3f994867d540d801899a138006e",
    "sdk_validation_sha256": "405d806da9273363fd09ae432ff665b1d049f6e6381e11484fd97fdc20a4a0d5"
  }
}