{
  "description": "Historical descriptive comparison of archived OLS coefficients. Both series use India as the country-of-origin reference. The experimental designs differ.",
  "retrieved": "2026-09-18",
  "contrast": "Iraqi rather than Indian origin",
  "units": "probability; multiply by 100 for percentage points",
  "model": {
    "label": "Mixtral-configured run (2024)",
    "run_url": "https://wandb.ai/why-earth/dev-subconscious-ai/runs/1c06ddda-b741-49b2-8512-6eb18cc2bceb",
    "summary_key": "OLS Model Output_method-24-03-11-21-09-19-045",
    "original_reference": "Germany",
    "iraq_original": 0.1230015390866978,
    "india_original": -0.014404658639456,
    "iraq_vs_india": 0.1374061977261538,
    "configuration": "mistralai/Mixtral-8x7B-Instruct-v0.1",
    "fallback_configuration": "gpt-4-1106-preview",
    "response_level_model_identity_verified": false,
    "source_commit": "550c8529459c30e8791de6f877b4d7960b484d7e",
    "choices_retained": 6666,
    "neither_choices_retained": 148,
    "synthetic_personas_retained": 413,
    "ols_profile_observations": 13332,
    "context": "Human-persona prompts set in the United States in December 2011; fully randomized profiles; neither option available. Execution code allowed GPT-4 fallback. Retained responses reproduce the archived OLS coefficients."
  },
  "human": {
    "label": "Archived human-study baseline",
    "run_url": "https://wandb.ai/why-earth/dev-subconscious-ai/runs/v0oecmg6",
    "summary_key": "OLS Model Output",
    "original_reference": "India",
    "iraq_original": -0.106,
    "india_original": 0,
    "iraq_vs_india": -0.106,
    "context": "Archived transcription of a historical human conjoint study, not a fresh estimate of present-day US opinion. Human study used restricted profile combinations and forced choices."
  },
  "limits": [
    "Descriptive point estimates only. No cross-study significance claim or estimated deployment effect.",
    "Matching the reference country does not equalize prompts, populations, choice options, randomization restrictions or estimands.",
    "No training-data or national-culture cause is identified.",
    "Human preferences do not define the intended safety policy.",
    "Historical execution permitted a GPT-4 fallback; these results cannot be attributed exclusively to Mixtral."
  ]
}