On one hand, it would be nice because it's annoying to read papers with a high level description of the method but no raw data or source code, so I have to gues

On one hand, it would be nice because it's annoying to read papers with a high level description of the method but no raw data or source code, so I have to guess the details, and sometimes the most important part is hidden in the details. In computer science there are some requests for source code / docker, but in other ares it's uncommon. > and thus combat the typically extortionate pricing for journal access On the other hand, the main problem is how to evaluate researchers and compare different areas. To make everyone not angry, the method is to count papers, just count them without reading. So Journals sell stamps. It's actually not so simple, because there are bad journals, so there is impact index, citations, h-index, lost of good and bad journals, number of pages, and in some cases evaluators even read the papers. So it will be very difficult to get/keep a research position if someone publish only in your site (or only in arXiv).