Methodology / Reviews and records v2
No score of any firm.
Only what people reported.
PI Data Hub publishes no model rating, ranking or grade of a law firm. It publishes a record of what has been observed, and verified reviews from the people who worked with the firm.
Two distinct things
The record: which fields we hold about a firm, and when each was observed.
Verified reviews: ratings and written accounts from clients, referring attorneys, firm staff and vendors, each email-verified and moderated.
What happened to the scores
Until 2026-09-14 every profile carried three stored model signals (scale, acquisition, marketing) and a record-coverage score, each out of 100. They were screening heuristics computed over an imported record. They were read as a grade of the firm, which is not what they measured and not something they could support, and no reader could act on them. They were removed from every public surface: profiles, the directory, the rankings, the API and the copyable research prompt. The underlying fields are still computed and stored; they are simply not published.
What replaced them is the only rating this site publishes: the average of verified first-party reviews, shown once a firm has at least one, and ranked only once a firm has at least five.
| Field | Points |
|---|---|
| Firm identity | 8 |
| City | 6 |
| State | 6 |
| Public website | 10 |
| Public phone | 8 |
| Street address | 10 |
| Practice areas | 12 |
| Attorney count | 8 |
| Office count | 4 |
| Website technology | 6 |
| Chat detection | 4 |
| Spanish-language detection | 4 |
| Consultation-offer detection | 4 |
| Source observation date | 10 |
Limited coverage: below 50. Partial coverage: 50–79. Broad coverage: 80–100. A complete record can still be outdated or wrong; completeness does not establish accuracy.
An explicitly recorded negative observation, such as “chat not detected,” counts as documented information. An absent field is unknown. Imported zero values for headcount or historical signals are treated as ambiguous defaults.
How a review is verified
Anyone who worked with a firm may review it: clients, referring attorneys, firm staff and vendors. Every reviewer confirms an email address by clicking a link, and every review is read by a moderator before publication. Attorneys supply a bar state and number, checked against the state bar. Professionals other than clients review from a work domain. Staff of the reviewed firm cannot review it.
FTC 16 CFR 465 applies: nothing is offered in exchange for a review, no review is suppressed, and a published review is removed only for a stated policy violation, never at the firm’s request. A firm may post one public response.
What a review asks
A review is a matter record, not a star rating. It asks 32 questions about the case itself: what kind of case, the city it was handled in, who actually ran it, how the fee worked, how it ended, how long it took, and what band the recovery fell in. It also asks 6 one-to-five ratings. The question set is versioned as data (data/review-ontology.json, version 2026-09-14.1) so the form and the validation behind it can never disagree.
What is published, and what is not
Published: the written review, the reviewer’s role, the one-to-five ratings and the firm average.
Not published: the reviewer’s name and email, and every answer about the case itself, including any amount. Those answers are collected to build market aggregates and are released only across five or more reviews, so no single matter is identifiable. As of 2026-09-14 no case answer is published at all.
The original calculation inputs and component weights are not preserved with every imported score. We display valid stored values as historical model signals; we do not invent a component breakdown or describe them as calibrated predictions.
What the profile can support
Expectations are generated from the fields in that firm’s record. Each section pairs its context with supporting observations and questions to verify. Business and client perspectives are kept separate.
A practice-area listing does not prove case suitability. A detected chat widget does not establish response speed. A Google rating does not establish legal ability, and nothing here forecasts a settlement or outcome.
Sources and dates
Records combine directory imports, firm websites, enrichment observations and verified corrections. A source observation date appears only when one is recorded. Processing dates are labeled separately; rebuilding a page does not make its underlying evidence fresh.
A correction date applies to the fields checked at that time. Google snapshot dates are separate. Imported Google ratings are attributed and remain separate from PI Data Hub’s first-party reviews.
Rankings
One ranking dimension exists: the verified review average. A firm needs at least five published first-party reviews to appear; below that it is unranked, not ranked low. The 1–5 average maps linearly to 0–100 for its letter grade: A is 80–100, B is 60–79, C is 40–59, D is 20–39, F is below 20.
Rankings use the selected national or state cohort before the search filter is applied, so searching a name never promotes it to rank 1. Equal values share a rank and the next rank skips the tied positions. Imported Google ratings are excluded; they are a separate, attributed snapshot.
See the ranking →Gemma writeups and entities
Gemma creates source-linked research drafts from available record facts and collected public sources. Published claims retain source references; professional relationships require explicit supporting evidence. These drafts are not consumer reviews or independent verification. Missing evidence stays unknown, and generation dates do not refresh the underlying observations.
Bar profiles have their own visible record-completeness rubric. News entries link to official publishers. The entity explorer retains types, source dates and documented relationships without treating a shared name or co-mention as proof of identity or affiliation.
Research prompts
Every profile contains a prompt with the firm identity, available facts, evidence gaps and research questions. It asks an analyst or research tool to verify primary sources, cite findings and leave unsupported claims unknown. Customizing the objective happens in your browser.
The prompt is a research starting point; it is not a completed assessment. Review its output and cited sources before making decisions.