These are good and important points, and I'd add that the rapid churn in the software ecosystem, especially in the Python world, further undermines the ideal of reproducibility, both at the most basic PDF level of "Can I see the visual appearance that the author intended?" and the deeper level of "Can I reproduce the calculations the author made from their empirical data?"
This morning I was reading a very valuable paper on eutectic points of evaporite salt mixtures from 01972, 53 years ago. If you publish something today in electronic form, how confident are you that it will be readable in 02078? Plain HTML and PNG will probably be readable, but it might not be ideal to depend on Web Workers, WebAudio, or the darker corners of the PDF spec.
This morning I was reading a very valuable paper on eutectic points of evaporite salt mixtures from 01972, 53 years ago. If you publish something today in electronic form, how confident are you that it will be readable in 02078? Plain HTML and PNG will probably be readable, but it might not be ideal to depend on Web Workers, WebAudio, or the darker corners of the PDF spec.