{
  "version": "1.0",
  "frozen_at_utc": "2026-10-22T06:20:55.973467+00:00",
  "dimensions": [
    {
      "name": "accuracy_and_sources",
      "weight": 25,
      "level": "article",
      "evidence": "Claim accuracy, support from primary sources, faithful citations, dated changing claims, labelled illustrative examples."
    },
    {
      "name": "cluster_strategy",
      "weight": 20,
      "level": "cluster",
      "evidence": "Required coverage, audience fit, distinct reader questions, useful progression, absence of avoidable duplication."
    },
    {
      "name": "practical_usefulness",
      "weight": 20,
      "level": "article",
      "evidence": "Actionable methods, concrete examples, tradeoffs, failures and sufficient technical depth."
    },
    {
      "name": "writing_quality",
      "weight": 15,
      "level": "article",
      "evidence": "Clarity, precision, structure, accessible terminology and editorial polish."
    },
    {
      "name": "coherence_and_links",
      "weight": 10,
      "level": "cluster",
      "evidence": "Consistent terminology and guidance, resolved links and coherent reading journeys."
    },
    {
      "name": "intent_and_metadata",
      "weight": 5,
      "level": "article",
      "evidence": "Specific reader intent, useful titles, accurate descriptions and plausible target queries; no invented search volumes."
    },
    {
      "name": "completeness",
      "weight": 5,
      "level": "cluster",
      "evidence": "All requested articles and supporting files, honest lengths and usable handoff."
    }
  ],
  "anchors": {
    "0": "Absent or unusable",
    "1": "Major failures; extensive reconstruction required",
    "2": "Substantial corrections required",
    "3": "Useful with ordinary substantive editing",
    "4": "Strong; minor corrections required",
    "5": "Excellent against stated requirements; ready for final publication checks"
  },
  "aggregation": "Score article dimensions 0\u20135 per article, then arithmetic mean across all eleven equally weighted articles. Score cluster dimensions once. For each dimension multiply score/5 by its weight and sum to obtain /100. Missing articles score 0 on article dimensions. Do not double-count an article as multiple samples or claim eleven articles are eleven independent model runs.",
  "publication_blockers": [
    "Fabricated citations or empirical results",
    "Materially incorrect guidance",
    "Missing complete article drafts"
  ],
  "comparison": "Score submissions independently before pairwise comparison. Allow ties. Do not force matching of independently selected article titles. Hide model identities, settings, token usage and runtime until quality scores are frozen. Reverse A/B presentation order for the second comparative pass and record disagreements.",
  "editing_effort": "Classify required edits and estimate effort explicitly as an estimate; observed editing time only when timed edits are actually performed.",
  "pilot_limit": "One run per model compares these submissions and does not establish general superiority or statistical significance."
}