[amalcon]

Indeed. Legal documentations are much more shared-source than software, by nature. Most agreements need to be in the possession of multiple parties and their attorneys in a reviewable form, for example, and court filings make the most contentious of such agreements public record.

This is a massive boon to the training data set. GitHub is also massive, but legal has other systemic advantages as well (e.g. being similar to past work is a structural advantage rather than just a practical one).