I figure the goal was to classify each TX as either unlawful or legal. Alas, only 23% managed to get one of those tags, whilst 77% were left unclassified, meaning, I figure that they could not trace nor tie these to either unlawful or legal senders/receivers.

If the wording in the article is as intended (and I’ve checked multiple sources and found nothing better), it means that the phrasing should state that “at least 2% were deemed unlawful”, since there is an immense unclassified 77% of the TXs in the study sample.

We could play the devils advocate here, and just play around with the classified TXs. That would lead to saying that the classified sample group is of 46.000 TX (2%+21% of 200.000). Out of this classified sample universe, 8,70% would be unlawful (4.000 TXs / 46.000 TXs) and 91,30% legal (42.000 TXs out of 46.000 TXs). I don’t particularly like this focus, but it goes to show that the unclassified group is important in size, enough to make the numbers shift wildly from the “only 2%” range provided by the title of the article.

Note: It would also be interesting to know how the 200K TX sample was selected, in order for it to be representative and sparce. For comparison reasons, the current number of BTC TXs/day is around 348K, and the historical total of nearly 444M.