{
  "version": "astra-four-model-unanimity-claude-opus-5-5-20261023-v1",
  "source": "https://eternalsearch.net/omega/opus-generated-confirmation-20261023/report.json",
  "sourceSha256": "2ad47f9ef43b45c90b0bf471d8d51ef888d6e7aee28445599c187857b4406d6f",
  "generator": "claude-opus-5-5",
  "design": "Preregistered rule recorded in the frozen cohort protocol before any forecast call for this cohort; the rule itself was chosen earlier after inspecting the Grok-generated votes. No laboratory validation",
  "reference": "Grok/N retrospective path on 2026-10-23: 240 input, 53 after Astra, 19 after unanimity, 14 of 19 article matches, 14 of 40 sensitivity",
  "endpoint": "At least two of three original judges assign article-correspondence score >=3",
  "handling": "unambiguous extracted saved responses",
  "first": "gpt-6-astra",
  "members": [
    "claude-sonnet-5-5",
    "gpt-6-astra",
    "gemini-4-argon-high",
    "grok-4.5"
  ],
  "minVotes": 4,
  "cohort": "Fixed common complete-case cohort of the four predictors from the audited report; technical omissions remain exclusions",
  "coin": "Original saved tosses on identical hypothesis identities, without repeat tosses",
  "uncertainty": "Pointwise descriptive 95% paired original-task cluster percentile intervals; 2000 draws, seed20261022; no multiplicity correction",
  "workload": "One model/hypothesis forecast per unit; reuse the saved first-stage vote in the panel. Not actual request count, tokens, money or elapsed time"
}