Reference
What gets published, where the text comes from, how often it is re-read, and what this method cannot tell you.
A jurisdiction appears in these tables only after its statute has been fetched from a published source and recorded. A jurisdiction we have not verified is left out. It is not filled in from a secondary summary, and it is not inferred from neighbouring states — a gap is visible, which is the honest form of a gap.
Right now that means all 51 jurisdictions in the recording-consent table. Three of them — Indiana, Mississippi and Tennessee — are a special case worth stating plainly: no copy of their statutes can be fetched and re-read at all, because each state publishes its code only through a system that serves nothing to a plain request. Those three are published from a frozen archive edition, are labelled as such on the page and in the data, and carry a flag saying that re-checking them cannot detect an amendment. Every run still probes for a live government source for them, so the day one appears they stop being a special case.
The preferred source is the government's own copy — a state legislature, a revisor of statutes, an official code site. Where a state serves no machine-readable copy of its own code, a published statutory source is used instead and the entry says so in plain words, on the page and in the JSON. You should be able to tell at a glance whether you are looking at something read off the state's own site.
Sources are requested with an identifying user agent and a courtesy delay between requests to the same host. Where a site declines that request, that source is treated as unusable and another is found. Nothing here is collected by working around a site's access controls.
Every source is re-fetched weekly. The fetched page is reduced to its text, narrowed to the part containing the statute, and hashed. That hash is compared against the one recorded when the entry was last accepted.
A changed statute is never republished automatically. When the text behind an entry moves, the automated check stops and raises it for review; it cannot quietly update what this site asserts the law to be. That is deliberate: the failure this design most wants to avoid is publishing a changed rule with the confident presentation of a verified one.
Fetching the right page proves we read the right statute. It does not prove our one-line summary of that statute is correct, and it cannot account for case law that narrows or expands the text, for a court's reading of an ambiguous term, or for a statute that was amended in a way the published copy has not yet caught up with. Automated checks corroborate the wording; they are not legal analysis and are not offered as any.
Each entry has a section headed the part that catches people out. None of it is verified. It is editorial judgment — our reading of how the statute works in practice — written rather than fetched, and it is marked as such on every page it appears on and flagged in the data as nuanceIsEditorial and nuanceVerified: false.
The court decisions named in those sections are now checked. Each one is looked up against the court record on CourtListener, weekly, and the entry confirms that the case exists and carries the reporter citation, court and year we give it, with a link so you can read the opinion yourself. A decision that fails that check is not published as a citation — it is named as a claim that did not survive checking.
That check found a real error when it was first run: a case this site named as Glass v. State is in fact State v. Glass. The citation, court and year were right and the parties were the wrong way round. It is corrected, and it is the reason the check exists — a case name carrying a volume and a reporter number reads as more authoritative than the statutory text beside it, so an unchecked one is the most damaging thing that could sit on these pages.
What that check still does not establish is what any of those cases held. Confirming a citation proves the case exists. It proves nothing about whether the court decided what we say it decided — that is a reading of an opinion, and no name-and-year match can stand in for it. So the holdings stay editorial, stay labelled, and are flagged in the data as caseHoldingsVerified: false even where every citation on the page has passed.
The division is meant to be usable rather than decorative: the statute, its source and its date are checked and you can re-check them yourself from the link. Everything we say about the statute is ours, and is labelled.
So: this is a map of where to look, kept current, with its own age printed on it. It is not advice, and it is not a substitute for a lawyer licensed where you are working.
A correction with a citation is the most useful mail this project gets, and it goes into the next revision. Send it here.
The reference tables are compiled and maintained by Andrew Bouse, the writer and publisher of The Insurance Investigator's Playbook. The field reports in the book are David Shackelford's; this reference layer is not his work and he is not responsible for it.