Methodology
How a claim gets
onto this site.
A site that argues a position has to be harder on itself about facts than a site that does not. This page describes the checking, including the parts that are enforced by a machine rather than by good intentions.
One registry, and a build that fails without it
Every source the site cites lives in a single file, and every citation marker in the prose refers to a key in that file. The set of valid keys is a type: a marker pointing at a key that does not exist is a compile error, and the site will not build.
That turns “do not invent information” from an intention into a failing build. It does not catch a citation pointing at a real source that does not support the sentence, and nothing automatic can. That is why the pinpoint locator is shown on every marker, so you can check the exact question or section for yourself.
49 sources in the registry.
The primary documents are stored, not just linked
Government web pages get reorganised and PDFs get replaced in place. A link on its own is not a durable record of what a document said on the day we read it, so the full text of the primary documents is extracted and committed alongside the site.
Those extracts are also what claims about absence are checked against. When this site says the statute never uses the word
lunch
, that is a full-text search anyone can repeat against the same file.3 with the full text committed.
Verification dates, and going stale
A verification date means a person opened the link, confirmed it resolved, and confirmed the quoted content still says what we claim. It does not mean an automated check pinged the URL.
After 180 days without that, a source is flagged on the sources page as overdue. The flag is public rather than internal, because a verification date that quietly ages is worse than none: it looks like a guarantee.
180 days before a source is flagged.
The registry is published in full on the sources page, with the date each source was last checked by a person.
Not knowing is a state we display
Three labels, all of them publicTwo labels appear in the prose where a passage is not carried by a source. One marks our own reasoning: arguable, but not checkable against a document. The other marks a factual question we could not answer from the sources we have, which is an invitation to send us the document.
Most of them are on the opposing case page, because the strongest objections turn on operational facts that no state document reports: how long a device retrieval window takes, and where recording incidents actually happen.
The same principle governs the district tracker. A blank cell in a policy table reads as “no restriction” to someone scanning it, which would be inventing information by leaving it out. So a district appears only once its row can carry a link to that district’s own published policy and a date, and every unknown cell says which kind of unknown it is.
The survey instrument
Questions before resultsEvery question is published in full before any result is quoted from it. A leading instrument produces numbers that get discarded the moment somebody reads the questions, and on a site that already argues a position, that is the fastest way to lose the argument.
It collects no name, no email address and no school. Geography is recorded no finer than a county and age as a grade band, because a school plus a grade plus a sentence in someone’s own words identifies a person in a small district. Results are published as aggregates with the response count beside every figure, and free text is read by a person before it appears.
What the survey is for is not our idea either: it is the review the Department tells districts to run.
Revisit and update the policy annually to reflect stakeholder feedback and emerging digital trends.Communicating Policy