Comment by snyy

We see far better performance with models. Heuristics break on richer content like codeblocks, formulae, quotes, etc. In our testing, our model was 25 F1 points better than Trafilatura.