{
  "dataset_identifier": "llb-intelligence-2026-08-09-r2",
  "version": "2026.08.09-r2",
  "entity_identifiers": {
    "brand_id": "Immutable internal identifier for a normalized website-domain brand group.",
    "office_id": "Immutable internal identifier derived from the exact Google Place identity; the raw Place ID is private.",
    "firm_id": "Immutable Law Leaderboard source-record identifier retained for lineage.",
    "profile_id": "Stable public profile identifier used by the existing benchmark release."
  },
  "record_relationship": "Brand -> office -> source firm -> city/practice -> aggregate review themes -> snapshot readiness.",
  "theme_metric_fields": {
    "matched_review_count": "Analyzed review texts with at least one explicit rule match for the theme.",
    "prevalence_pct": "Matched review texts divided by analyzed text reviews for the office.",
    "wilson_low_pct": "Lower bound of the 95% Wilson interval for theme prevalence.",
    "wilson_high_pct": "Upper bound of the 95% Wilson interval for theme prevalence.",
    "average_confidence": "Average rule-level confidence across matched texts; not probability that a claim is true.",
    "qa_precision": "Agreement precision against the OpenAI-labeled validation sample."
  },
  "excluded_public_fields": [
    "review text",
    "reviewer identity",
    "owner-response text",
    "Google Place IDs",
    "Google CIDs",
    "street addresses",
    "phone numbers",
    "provider payloads",
    "quarantine pair details"
  ],
  "limitations": [
    "Brand grouping uses exact normalized public website domains plus a small explicit registry; it is not a legal ownership determination.",
    "Theme labels describe language in a sampled review corpus and can miss context, sarcasm, or uncommon wording.",
    "Outcome language is always marked unverified and is not evidence of case results or likely outcomes.",
    "Market percentiles describe only the offices covered in a Law Leaderboard city dataset, not every law firm in the market.",
    "Trend fields remain suppressed until at least two distinct complete observations exist."
  ]
}