Editorial Standards

Law Leaderboard publishes data-backed comparison pages. The editorial standard is simple: every important claim should be traceable, scoped correctly, and phrased to match what the site actually measures.

Evidence Hierarchy

  1. Structured facts from the current dataset.
  2. Public source documents and official citations.
  3. Generated narrative that restates stored facts without extending them.

Entity Identity Standard

A law-firm brand and a physical office are different entities. Each normalized record receives an opaque, stable Law Leaderboard ID. Raw Place IDs, CIDs, addresses, and phone numbers remain in the private resolution layer and are not included in public intelligence files.

Only an exact Place ID or CID match can automatically link an office. Domain, name, address, phone, brand relationships, and proximity are supporting evidence only. Conflicting or incomplete matches must be quarantined instead of silently merged.

Count And Metric Discipline

The site must distinguish between public Google review counts, retrieved review records, and text reviews analyzed. A page should not collapse those counts into a single number when they mean different things.

Review-derived signals such as communication mentions, recommendation rate, or case-type mentions describe what appears in the current sample. They do not independently prove specialization, likely case value, or attorney quality in a complete sense.

Legal Content Standard

State-law content is for orientation. Official statutes, rules, and current case law remain the controlling sources. Legal summaries should be linked to primary sources wherever practical and should avoid confident claims when source coverage is weak.

The platform should be authoritative for review-based law-firm market intelligence and useful for legal orientation. It should not present itself as the ultimate authority on substantive law without source-backed legal coverage.

AI Usage Standard

  • Allowed: classification, extraction, summarization from stored facts, comparison generation, and readability rewrites.
  • Not allowed: invented legal guidance, invented firm strengths, unsupported specialization claims, or confident prose from thin data.
  • Generated sections should remain subordinate to visible facts, citations, freshness signals, and limitations.

Review-theme version review-themes-v1.0.1 was independently checked with the pinned model gpt-4o-mini-2024-07-18 on a deterministic 483-review sample. A manual positive-evidence check inspected 27 records: 15 model agreements and all 12 model-disagreed positives for the three publishable themes. API judgments are a QA reference, not ground truth, and sampled precision does not establish complete recall.

Review Theme Publication Standard

A theme can appear in public office metrics only when the office has at least 50 analyzed review texts, the theme has at least three matches, average label confidence is at least 0.80, independent QA precision is at least 0.80, and at least 10 predicted positives were reviewed. A theme that misses any gate remains in the private QA layer and is suppressed publicly.

The current public set is Communication, Responsiveness and delays, and Outcome language (unverified). Outcome language reports what a reviewer claimed; it is not evidence that a result, recovery, settlement, or verdict occurred. Read the full QA table and limitations.

Publishing Threshold

  • A page should have a direct answer block, source-backed metrics, freshness signals, and limitations.
  • A page should not publish if provenance is missing or core counts conflict.
  • A derived dataset should not publish unless its manifest byte counts and SHA-256 checksums match every distribution.
  • A theme should not publish when any office sample, match-count, confidence, or independent-QA gate fails.
  • A trend should remain suppressed until at least two distinct, complete profile observations exist.
  • FAQ schema should not be used for weak, generic, or non-responsive answers.