Because the way papers are reviewed is highly manual and relies on individual reviewers who are acting as volunteers and don’t work full time doing review work.

Because the way papers are reviewed is highly manual and relies on individual reviewers who are acting as volunteers and don’t work full time doing review work. This makes it very difficult to implement systematic checks of the sort that would catch minor pieces of retracted work that aren’t central to this paper’s results. To fix this systematically you’d either need automated systems or human-executed “checkbox processes” with training. Of course you could implement those systems, but they’d be costly either in terms of further centralizing the peer review process behind companies like Elsevier or taking critical human time away from reviewing the actual results. And the benefits if you implemented these review systems might be highly limited, plus you’d probably run into a host of additional problems. On the “highly limited” side, many papers have a related work section that doesn’t list primary results that are relevant to the paper’s findings, but instead mainly serves as a reference guide. You might be writing a paper on (here I’m making up silly analogies) LED light bulbs, and you’d have a set of references that begins with “a line of work looks at incandescent light bulbs”. An accidental citation to a retracted work there is sloppy, but mostly scientifically irrelevant when no result in the paper actually refers to anything in those citations (which is why these citations slip by manual reviewers, who focus their valuable review time on citations that affect the scientific results in the paper.) You also have the more annoying problem that citation formats aren’t standardized, so it’s easy for a retracted paper to be cited in a format that an automated review system will miss. Google Scholar throws all the might of Google’s systems at this, and still treats the preprint version and conference version of my papers as different works. (This might be the right treatment or it might be the wrong one, but it’s murder on retraction scanners if the author cites the preprint.) Plus then there’s the additional burden of dealing with unimportant retractions that are issued after a paper that cites them has been published, which should be addressed but largely just cause annoying work for scientists and publishers and (as the article notes) are aggravating because it’s difficult for readers to distinguish a minor correction regarding an unimportant reference from a major one that affects the findings of the work.