Method

Dossiers on this site are opinionated research notes. I choose the question, the sources and the framing. I work with frontier models throughout, but a dossier is not simply an output from an LLM asked about the issue in question. I rewrite, cut and reorder, over many rounds, until the text says what I think. Typically, a dossier will either build on my work already published elsewhere or is developed in the process of my work on another publication.

Legislative documents of the kind I rely on are not clean data, even when they come as official PDFs. Extracting the text reliably and in a structured way currently still requires a lot of manual work, even when assisted by frontier AI.

So a legislative tracker is built in rounds. Frontier models extract each text, and a separate pass, starting fresh, checks the result against the source. I resolve what they disagree on and anything that affects meaning. Every correction is recorded and reapplied whenever the tracker is rebuilt, and each new kind of error becomes something later work is checked against.

• Dossier versions: each dossier has a version number, and every public version has a short record on the Updates page, also carried by the RSS feed.
• Legislative tracker updates: I update a tracker when a new text appears, and its overview names the latest text it covers.
• Corrections: I correct errors as soon as I confirm them. Corrections that change substance appear in the update record.

• Downloads: every dossier can be downloaded as a Markdown file, and dossier comparisons as CSV.
• API: corporate subscribers can have API access on request.
• Stable links: every provision and dated text has a stable address, and older addresses redirect, so citations keep working.

Send possible errors, omitted materials or methodological questions to research@eutechreg.com.