Is set basically syntactic sugar for deepcopying a struct, mutating the specified field, and then returning that deepcopy? Seems like it could be quite slow.
There is an underlying precarity in the academy that is so deep it almost feels like a natural part of science. This is part of what makes reform so difficult. Early on, rocking the boat feels like career suicide. Later on, if you are lucky enough to become established, you are much less likely to feel like deep reform is necessary: after all, the system benefited you. And even then, the precarity doesn't go away. You still compete for grants like everyone else, and your trainees are attempting to become established. Why should they be the ones to shoulder the risk of, e.g., ignoring the glam journals and exclusively putting out preprints?
It seems to me that something like eLife's model is the best solution to this [0]: you still have a minimal amount of curation, but generally if a paper is written well enough and within the field it won't be desk rejected. Then, it gets published on the site and sent off for peer review. Peer reviewers assess how sound the paper is and pass a judgement which readers can view, as well as provide some recommendations to the authors make it stronger. The authors can then either revise the paper, or do nothing at all. In either case, papers don't languish in reviewer hell and the larger scientific community gets to see it.