Verification methodology
Every field we publish carries a citation, a retrieval date and a verification status. This page describes how each of those is produced, and states the limits of what we will assert.
The record
The unit of the dataset is a record: one regulatory claim, scoped to one jurisdiction, resolved against one primary source.
A record is not a summary of a statute. It is an answer to a specific question — the bond amount, the licensing authority, the permitted activity — expressed so that it can be true or false rather than approximately right.
The bond must be in the amount of $25,000. Cal. Bus. & Prof. Code § 7071.6 · Contractors State License Board, Bond Requirements
Records are versioned. When a requirement changes, the prior record is retained with its own retrieval date rather than overwritten. The history is part of the dataset.
Verification statuses
Four statuses. A record carries exactly one.
| Status | Meaning |
|---|---|
| Verified | The claim is supported by the cited primary source, read directly, on the stated retrieval date. |
| Contradicted | The claim is inconsistent with the cited primary source. Used only where the source text is unambiguous. |
| Stale | The claim was accurate at a prior retrieval date and the underlying source has since changed. |
| Unverified | No primary source resolves the question. We do not infer, and we do not fill the gap from secondary sources. |
A dataset that answers every question is a dataset that guesses. Roughly one field in nine in our current holdings is unverified, most often where a state devolves authority to municipalities. Reporting that honestly is the product.
Source hierarchy
Sources are ranked. A claim is resolved at the highest available tier, and the tier is recorded.
| Tier | Source |
|---|---|
| 1 | Statute and regulation as published by the state legislature or code publisher |
| 2 | The administering agency's own published requirements — licensing board, department, commission |
| 3 | Agency bulletins, formal guidance and rule notices |
| — | Everything else. Trade press, aggregators and competitor pages are used to locate questions, never to answer them. |
Where tiers conflict — an agency page that no longer matches the statute it implements — the record cites both and is flagged for review. That conflict is itself a finding, and in practice it is one of the more common ones.
Retrieval discipline
Three rules govern every retrieval.
- The source is read, not recalled. Claims are resolved against fetched text at verification time. Nothing is asserted from prior knowledge of the subject.
- The date is recorded, not implied. Retrieval dates are absolute. A record without one is not publishable.
- Quoted text is verbatim. Where a statutory phrase decides the question, it is quoted rather than paraphrased, because paraphrase is where accuracy is lost.
Change detection
Verification is not a one-time event. Records are re-checked on a rotating schedule weighted by volatility: bond amounts and fee schedules more often than licensing authority structure.
When a source changes, the record moves to stale and enters the review queue rather than updating silently. Silent updates destroy the audit trail, which is the part of the dataset that has value.
Human review
Retrieval and comparison are automated. Two things are not.
Status assignment on any contradicted record. Marking a published claim as contradicted is a consequential assertion about someone else's work. It is reviewed by a person against the source text before it is issued.
Scope judgements. Whether a question is answerable at state level, or has been devolved, is a reading question rather than a retrieval question.
What we do not cover
The exclusions are as load-bearing as the coverage. Any vendor claiming complete municipal coverage in a devolved state is describing something they have not built.
StatuteData is a research and verification service. Records describe published regulatory text as retrieved on a stated date. They are not legal advice and do not certify compliance.
Corrections
If a record is wrong, we want to know, and the correction is public. Send the record ID and the source you believe governs. Corrected records retain the superseded version and the date of correction.
We hold our own published claims to the standard described on this page. Every figure on this site carries a citation and a retrieval date, including the ones that make us look worse.