What is really in your legal data?

Before building on legal data, know what a source really holds, where it is wrong or silent, and prove it.

One instance. A research team wanted a file joining every judgment over six decades to its neutral and law-report citations. What we found:

  • Some judgments have no text; others name cases without their report citations.
  • The same report is printed in different ways.
  • One judgment appears under two neutral citations; two decisions are printed as one.
  • Lists disagree on which judgment a report belongs to: the trial decision or the appeal.

What we did. We checked the file against two other lists and traced every gap to one of 15 causes, each with a fix and a number the next build must show. We filled missing texts from our own corpus, put each judgment under one citation and split combined documents. Where we had been wrong, we said so in writing.

What they have. The file, in CSV and JSON; tested corrections for its next build; and a written account of where it is still short, and why.