Four things this does not do,
said as plainly as the things it does.

1. This platform does not verify that a document is what its publisher says it is

There is no authentication step anywhere in this system. None.

When a PDF is submitted, nothing checks that it came from the agency named, that it is the whole of what was released, that its pages are in the order they arrived, or that a single character of it is as the agency wrote it. There is no channel to the agency, no signature to verify, no counterpart copy to compare against. A carefully forged release and a genuine one are the same bytes to this system.

What the platform records is that somebody asserted a document came from a request with a given identifier, filed on a given date with a given agency. That is a claim by the author, published in a form that makes it checkable by anyone who cares to check — which is genuinely valuable and is not the same as verification.

2. Publishing a document does not make it true

Three separate things get collapsed into one all the time, and the collapse is where most bad records journalism lives:

The document exists and came from the agency

What the request establishes, if the release was genuine.

The document says what it says

Readable off the page. Almost never in dispute.

What the document says is true

A completely separate question, and one that a records request cannot answer.

Agencies write down things that are wrong. An incident report is evidence of what an officer wrote, not evidence of what happened. A contract is evidence of what was agreed, not of what was delivered. An internal email is evidence that somebody said something, at a moment, possibly while wrong or lying or joking. A record is a record.

The platform publishes documents and the author’s reading of them. It endorses neither. There is no verified badge, no fact-check, and no editorial layer — and a reader who takes publication here as a signal of accuracy has been misled by something other than this page.

3. The redaction scan is a net, not a guarantee

It runs before anything can publish, it blocks, and it routes minors and medical details to review. It also misses things, and the ways it misses are knowable:

• It reads text. An image-only scan with no text layer is invisible to it.
• It has no idea who is a public official and who is a bystander. A phone number is a phone number.
• It cannot see that a redacted name is identifiable from a badge number, a shift roster and a date on three different pages.
• It does not know what is already public, so it cannot tell you that combining this release with a previous one identifies somebody neither one did.
• It is pattern matching. It fires on things that do not matter and stays quiet on things that do.

The author remains responsible for what they publish. That responsibility is not transferred, shared, or reduced by the scan’s existence, and the product’s language is written so it cannot be read as if it were: a clean scan says “the scan raised nothing”, never “no personal information found”, and never “safe to publish”. Privacy and redaction →

Documents the scan never saw

A document submitted as a URL with no extracted text is a document the scan cannot read. Those drafts are marked unscanned, not clean, and the distinction is carried through to the published post. A post whose documents were never scanned says so where a reader can see it.

4. Nothing here is legal advice

Specifically, and not exhaustively: this platform does not advise on whether a record is exempt, whether an agency’s redactions were lawful, whether republishing a lawfully released record is lawful where you are, whether a person named in a document is a public figure, or what any statute requires of you. It does not assess defamation risk, privacy tort exposure, or contempt. It does not tell you whether the thing you are about to publish is a good idea.

Operational limits worth knowing

• Removal cannot recall what was public. Archives, caches and screenshots are outside this system. That is why the scan sits on the way in. Takedown →
• The vocabulary is fixed and it is an argument. Research that does not fit one of the fifty-five subjects gets filed under the nearest one, and the nearest one is sometimes wrong. Topics →
• A pseudonymous author is not an unaccountable one, but a reader loses something. Institutional standing cannot be weighed when no institution is named; the provenance line is what stands in for it, and it is not a complete substitute.
• The widget can only show what the read API serves. If a post is removed, embedded copies stop showing it — which is the correct behaviour and also means an author’s own page can change without them touching it.
• There is no import from another platform and no bulk export of somebody else’s corpus. Every post here was submitted through the ingest contract by someone holding a publish-scoped token.

What this is genuinely good at

Making records research checkable. A post here carries the agency, the jurisdiction, the request identifier and the dates, so a stranger who thinks it is wrong has a specific next step instead of an argument. And it puts a redaction scan and a human decision between a document dump and a public page, which is the difference between careful publishing and fast publishing in the one subject where the difference lands on people who never chose to be involved.

Everything on this page is the boundary around that. Neither half is the interesting one on its own.

About this site

A static build with no framework and no third-party requests: no CDN fonts, no analytics, nothing loaded from another origin. Every page carries a narrated, captioned explainer served from this origin, with its transcript printed below it in text. The taxonomy on Topics is generated from the same source file the platform itself reads.

Loop footage is credited on each band. Explainer narration is synthesised (ElevenLabs), and the transcript beside each video is the exact text that was voiced. The example records shown on the widget page are placeholders and are labelled as such — no real author, post, agency or figure appears anywhere on this site.