{
  "schemaVersion": 1,
  "studyUrl": "https://www.universaliqtest.com/ai-iq-test",
  "recordedOn": "2026-11-14",
  "prompt": "You are given one visual pattern-completion question. The upper panel contains the question; the lower panel contains eight answer choices labelled 1 through 8. Select the choice that best completes the missing part. Return a JSON object with exactly one field, \"option\", whose value is the chosen integer from 1 to 8. Do not include an explanation.",
  "questions": 40,
  "optionsPerQuestion": 8,
  "imageSize": [
    1152,
    1552
  ],
  "context": "One fresh context per item; one user message; no tools, browsing, hints or answer key.",
  "optionOrder": "Predeclared seeded balanced shuffle. Each original option appears five times at each displayed label across 40 items.",
  "optionOrderSeed": "universaliqtest-full-2026-11-14-balanced-v1",
  "baselineSettings": {
    "protocolVersion": 3,
    "answerFormat": "strict JSON schema: option integer 1-8",
    "reasoningEffort": "medium",
    "maxTokens": 8192,
    "context": "fresh request per question",
    "tools": "none",
    "optionOrder": "seeded balanced shuffle; each original option appears five times at each label",
    "optionOrderSeed": "universaliqtest-full-2026-11-14-balanced-v1",
    "runsPerModel": 1,
    "modelAnswerRetries": 0,
    "transportRetries": 1,
    "imageSize": [
      1152,
      1552
    ]
  },
  "baselineModelVersions": [
    {
      "version": "google/gemini-4-carbon-20261110",
      "provider": "Google AI Studio"
    },
    {
      "version": "anthropic/claude-fable-5.2-20261109",
      "provider": "Anthropic"
    },
    {
      "version": "openai/gpt-6-astra-20260903",
      "provider": "OpenAI"
    }
  ],
  "scoring": "Map the returned display label back to its original option and grade offline with production key iq-40-2026-09:key-v1. Cutoffs count as unsuccessful slots, separately from wrong returned choices.",
  "calibration": "Separate, ungraded selected-item development. No returned development answer is substituted into an original score. The last profile requested reasoning progress and discarded all reasoning text locally.",
  "recovery": "Original run: one unanswered unbilled HTTP 503 retried identically. Initial calibration: one investigated unanswered unbilled timeout recovered with streaming; later unavailable slots retained without retry. Final progress profile: one attempt per selected question.",
  "review": "Scores were cross-checked locally against the production grader; request/image hashes, timings and receipts were audited. This is an internal study without external peer review.",
  "limitations": [
    "One original run per model; no estimate of run-to-run variation.",
    "Provider-specific effort names and token counts do not establish equal compute.",
    "Public test items; training contamination was not measured.",
    "Static numbered image layout differs from the interactive human test.",
    "No measured human IQ, human percentile or general intelligence ranking.",
    "Calibration was selected for completion failures and is not a representative accuracy sample.",
    "Changing provider and allowance together does not isolate their effects; delivery observations do not establish a universal timeout cause."
  ],
  "provenance": {
    "baselineProtocolFingerprint": "f4425b7fb5a235cd7c11fb84c4849c4f7df3d43064160f289b8434ec9478f59b",
    "calibrationProtocolFingerprints": [
      "a3d759d4258fecfc3dba39d0f437ba7b0b03533584394edbf0d28e45f5445546",
      "8f8ba0bf7d62f00e10dab8b9f25602401cfcab131603877d9d1529492b05c5db",
      "de519ca05127ce18e087812e3b3d500e9d720d9ac4131195bfb15b5dc2eda1c4"
    ]
  }
}