254 primary legal records, each checked against the source itself

How verification works

1. Fetch. Each cited section is requested from the publishing body, honouring the declared character set. Several legislatures still serve windows-1252, and decoding those bytes as UTF-8 corrupts currency and section symbols.
2. Normalise. Markup, HTML entities, non-breaking spaces and curly quotation marks are folded so that comparison tests the words, not the formatting.
3. Assert. Each published figure carries named assertions. Florida’s commission schedule, for example, requires four separate rate clauses plus the definition of compensable value to be present before any rate is published.
4. Record. The HTTP status, a SHA-256 of the response body, the matched excerpt and the timestamp are written to a ledger that ships with the site.
5. Gate. Every release build fails if any source is unverified, if any quotation is not found in the captured text, if the recorded checks no longer match the checks the code defines, or if the record is more than 90 days old.

This is deliberately stricter than citing a statute. A citation tells you where to look. This ledger records that someone looked, when, and what the page said at that moment.

What this does not claim

Re-fetching happens when the verification script is run, not on every page load. The date above is when the sources were last actually retrieved. A statute amended after that date would not be reflected here until the next run, which is why the date is published as prominently as the figures, why the build refuses to ship a capture older than 90 days, and why re-verification runs on a schedule rather than only when someone remembers.

Matching is done on normalised text: HTML markup, character entities, non-breaking spaces and curly quotation marks are folded before comparison, so a quotation matches the publisher’s words rather than the publisher’s formatting. It is not a byte comparison of raw HTML.

Browse by jurisdiction