The problem is the downloaded pages of the papers themselves, which often completely lack proper bibliographic data (or even header/footer info other than a page number). Compare this with a page from a corporate tech report, where each page might have the title and document number somewhere.

They used to come in magazines, with the date on the front, and authorship metadata on the index.

Nowadays they come as individual papers, but the publishers kept the format... because, why change anything when you have guaranteed profits?