COVID-19 Records

About this site

This is a technical demonstration. It was built to test whether a structured knowledge format can serve as a retrieval layer over a large, messy document set — how well search, citation and summarisation hold up when the source material resists all three. It is published anonymously and has no connection to the Committee, to anyone named in these records, or to any party to the events they describe.

The corpus was chosen because it is a hard test. It is public and citable, but it is also 5,247 pages across 14 releases, much of it scanned images with no text layer at all, mixing 6 kinds of record — chat logs, diary entries, email, interview transcripts, attachments and news articles — with inconsistent attribution, heavy duplication, and a document density that runs opposite to analytical importance. Those are the conditions a retrieval layer has to survive. Its visibility also means the site sees real use, which a test of this kind needs.

The site takes no position on what these documents show — on who was right, on what was known when, or on what any of it proves. It asserts one thing only: that each record here reflects the source page it cites.

Published anonymously, it gives you no grounds to take that on trust — which is why nothing here asks you to.

Every figure on this site decomposes into records, and every record into a page you can open yourself. Nothing here asks to be believed — only checked.

Agents: the same notice is served at /llms.txt and /api/guidance, keyed authorship.