{
  "id": "test-03",
  "version": "1.0.0",
  "title": "Test 3: complex multi-prompt virtual-data apps",
  "backend": "virtual",
  "backendLabel": "Virtual business tables; read-only Office 365 Users profile",
  "design": "Five model/effort configurations once each, four identical ordered authoring prompts in each fresh app: build, connect the existing app to Office 365 Users, extend workflows, then change requirements. Independent external review.",
  "evidenceRoot": "evidence\\test-03\\runs",
  "acceptancePath": "benchmark\\test-03\\acceptance.json",
  "acceptanceThreshold": 85,
  "allCriticalChecksRequired": true,
  "authorizationPath": "benchmark\\test-03\\authorization.json",
  "measuredBuildLimit": 5,
  "automaticExtension": false,
  "controls": {
    "identicalBuildPrompt": "benchmark\\test-03\\stage-01-build.txt",
    "promptFreezePath": "evidence\\test-03\\setup\\protocol-freeze.json",
    "identityVerificationPath": "evidence\\test-03\\setup\\tenant-verification.json",
    "browserEndpoint": "http://127.0.0.1:9334",
    "externalReviewInstructionsPath": "benchmark\\test-03\\external-review-instructions.txt",
    "reviewMode": "external-only",
    "nativeSelfReviewAllowed": false,
    "freshTaskAndAppEachRun": true,
    "noConcurrentBuilds": true,
    "promptLimits": { "review": 0, "planned-change": 3, "clarification": 2, "correction": 2, "cost": 12 },
    "profileGate": "Build the app first; connect Office 365 Users through the existing app in stage 2, not a general Cowork connector. Pause remaining runs at a shared unresolved authorization or managed-app connector blocker.",
    "storageRule": "Synthetic related virtual/local tables; browser-local persistence is accepted. No managed-server persistence requirement.",
    "stopRule": "Unresolved common platform/authorization blocker; two defect repairs; thirty observed active-generation minutes per authoring response or ninety per task. No hidden replacement or extra app.",
    "privacyRule": "Read only the current user's profile through Office 365 Users. No real approvals, transactions, other employee reads, directory writes, admin settings, publication or sharing. Keep profile evidence private.",
    "comparisonRule": "Compare within this four-stage complex-app scope only. Preserve Test 1/Test 2 as completed separate cohorts. Selected Cowork settings do not prove internal builder routing."
  },
  "stages": [
    { "id": "stage-01", "kind": "build", "prompt": "benchmark\\test-03\\stage-01-build.txt", "title": "Build the relational virtual-data app first" },
    { "id": "stage-02", "kind": "planned-change", "prompt": "benchmark\\test-03\\stage-02-connect-profile.txt", "title": "Connect the existing managed app to Office 365 Users" },
    { "id": "stage-03", "kind": "planned-change", "prompt": "benchmark\\test-03\\stage-03-change.txt", "title": "Virtual approvals, asset lifecycle, themes and preferences" },
    { "id": "stage-04", "kind": "planned-change", "prompt": "benchmark\\test-03\\stage-04-change.txt", "title": "Approval policy change, revisions, cost centres and responsive refinement" }
  ],
  "runs": [
    { "id": "run-01", "model": "Sonnet 5", "modelFamily": "Claude Anthropic", "effort": "Medium", "effortValue": 1, "replicate": 1 },
    { "id": "run-02", "model": "Opus 5", "modelFamily": "Claude Anthropic", "effort": "High", "effortValue": 2, "replicate": 1 },
    { "id": "run-03", "model": "Opus 5", "modelFamily": "Claude Anthropic", "effort": "Medium", "effortValue": 1, "replicate": 1 },
    { "id": "run-04", "model": "Sonnet 5", "modelFamily": "Claude Anthropic", "effort": "High", "effortValue": 2, "replicate": 1 },
    { "id": "run-05", "model": "GPT 5.5", "modelFamily": "GPT OpenAI", "effort": "Medium", "effortValue": 1, "replicate": 1 }
  ]
}