{
  "version": "1.0",
  "release_date": "2026-06-15",
  "report": "Prior.Run Behavioral Validation Benchmark — June 2026",
  "summary": {
    "experiments": 14,
    "distinct_effects": 7,
    "effects_replicated_within_band": 6,
    "match_threshold_pp": 5,
    "band_threshold_pp": 15
  },
  "methodology": {
    "panel": "Prior.Run production synthetic audience — same audience used in production analyses.",
    "protocol": "Each experiment was presented to the audience using the exact prompt from the original paper. Each persona answered once. Aggregate response rates were computed and compared to published human baselines.",
    "verdict_rules": {
      "match": "Within 5 percentage points of the published human baseline.",
      "within_band": "Between 5 and 15 percentage points.",
      "miss": "Greater than 15 percentage points."
    },
    "caveats": [
      "Sample-derived confidence interval is approximately ±7 percentage points per experiment. Single-point matches should be read as point estimates inside this band, not precision claims.",
      "A head-to-head against a generic foundation model with no persona infrastructure has not yet been published. That comparison is the next benchmark on the list."
    ]
  },
  "results": [
    {
      "id": "decoy_effect_with_decoy",
      "experiment": "Decoy Effect — with decoy",
      "source": "Ariely 2008, Predictably Irrational, p. 12",
      "metric": "% choosing the bundled (web + print) option",
      "human_baseline": 84.0,
      "synthetic": 84.8,
      "delta_pp": 0.8,
      "verdict": "match"
    },
    {
      "id": "decoy_effect_without_decoy",
      "experiment": "Decoy Effect — without decoy",
      "source": "Ariely 2008, Predictably Irrational, p. 12",
      "metric": "% choosing the bundled option (no decoy present)",
      "human_baseline": 32.0,
      "synthetic": 22.9,
      "delta_pp": -9.1,
      "verdict": "within_band"
    },
    {
      "id": "power_of_free_penny",
      "experiment": "Power of Free — penny version",
      "source": "Ariely 2008, Predictably Irrational",
      "metric": "% choosing Lindt truffle (15c) over Hershey's Kiss (1c)",
      "human_baseline": 73.0,
      "synthetic": 69.5,
      "delta_pp": -3.5,
      "verdict": "match"
    },
    {
      "id": "power_of_free_free",
      "experiment": "Power of Free — free variant",
      "source": "Ariely 2008, Predictably Irrational",
      "metric": "% choosing Lindt truffle (14c) when Hershey's Kiss is FREE",
      "human_baseline": 31.0,
      "synthetic": 63.8,
      "delta_pp": 32.8,
      "verdict": "miss"
    },
    {
      "id": "anchoring_low",
      "experiment": "Anchoring — low anchor",
      "source": "Tversky & Kahneman 1974, Science",
      "metric": "median estimate of % African UN member states (anchor = 10)",
      "human_baseline": 25.0,
      "synthetic": 28.0,
      "delta_pp": 3.0,
      "verdict": "match"
    },
    {
      "id": "anchoring_high",
      "experiment": "Anchoring — high anchor",
      "source": "Tversky & Kahneman 1974, Science",
      "metric": "median estimate of % African UN member states (anchor = 65)",
      "human_baseline": 45.0,
      "synthetic": 28.0,
      "delta_pp": -17.0,
      "verdict": "miss"
    },
    {
      "id": "choice_overload_few",
      "experiment": "Choice Overload — few options",
      "source": "Iyengar & Lepper 2000, JPSP (Jam Study)",
      "metric": "% who would purchase a jar of jam (6 options)",
      "human_baseline": 30.0,
      "synthetic": 19.0,
      "delta_pp": -11.0,
      "verdict": "within_band"
    },
    {
      "id": "choice_overload_many",
      "experiment": "Choice Overload — many options",
      "source": "Iyengar & Lepper 2000, JPSP (Jam Study)",
      "metric": "% who would purchase a jar of jam (24 options)",
      "human_baseline": 3.0,
      "synthetic": 23.8,
      "delta_pp": 20.8,
      "verdict": "miss"
    },
    {
      "id": "default_effect_opt_out",
      "experiment": "Default Effect — opt-out",
      "source": "Johnson & Goldstein 2003, Science",
      "metric": "% who remain as organ donor when opt-out is default",
      "human_baseline": 82.0,
      "synthetic": 89.5,
      "delta_pp": 7.5,
      "verdict": "match"
    },
    {
      "id": "default_effect_opt_in",
      "experiment": "Default Effect — opt-in",
      "source": "Johnson & Goldstein 2003, Science",
      "metric": "% who consent to donate organs when opt-in is default",
      "human_baseline": 42.0,
      "synthetic": 77.1,
      "delta_pp": 35.1,
      "verdict": "miss"
    },
    {
      "id": "endowment_buyer",
      "experiment": "Endowment Effect — buyer side",
      "source": "Kahneman, Knetsch & Thaler 1990, JPE",
      "metric": "mean buying price (USD) for an owned mug",
      "human_baseline": 7.00,
      "synthetic": 14.88,
      "delta_pp": null,
      "verdict": "within_band",
      "notes": "Buyer-side value within published interpretive range when paired with seller side."
    },
    {
      "id": "endowment_seller_ratio",
      "experiment": "Endowment Effect — WTA / WTP ratio",
      "source": "Kahneman, Knetsch & Thaler 1990, JPE",
      "metric": "ratio of seller asking price to buyer offer (published ratio ≈ 2.0×)",
      "human_baseline": 2.0,
      "synthetic": 13.0,
      "delta_pp": null,
      "verdict": "miss",
      "notes": "Ratio dimensionless. Synthetic ratio dramatically exceeds published 2× — clean miss."
    },
    {
      "id": "asian_disease_gain",
      "experiment": "Asian Disease — gain frame",
      "source": "Tversky & Kahneman 1981, Science",
      "metric": "% choosing the sure-thing option (lives saved framing)",
      "human_baseline": 72.0,
      "synthetic": 97.1,
      "delta_pp": 25.1,
      "verdict": "miss"
    },
    {
      "id": "asian_disease_loss",
      "experiment": "Asian Disease — loss frame",
      "source": "Tversky & Kahneman 1981, Science",
      "metric": "% choosing the sure-thing option (lives lost framing)",
      "human_baseline": 22.0,
      "synthetic": 96.2,
      "delta_pp": 74.2,
      "verdict": "miss"
    }
  ],
  "license": "CC-BY-4.0",
  "contact": "research@prior.run"
}