This is why years ago I switched to a static site generator. Using something like Django was fragile. My provider would change some library and poof - the site is down. I don't use my site often (post on it once every so many months) - so I simply switched. Never gone down since then.
A few days ago I opened a Jupyter notebook I had authored in 2015 - and it failed to reproduce my results. I relied on pandas, and pandas has gone through API changes. Fortunately, it wasn't hard for me to search around for the newer versions of those APIs and fix things - but this is only 4 years after I authored it. Can I have any confidence that 20 years from now I could reproduce the results - or even find documentation for the APIs I used to understand exactly what I was doing?
I mean sure - I could put the OS and all the libraries needed to reproduce the work in a container and be guaranteed[1] it'll run later, but the only people who will do that are those who need to for regulatory compliance[2].
[1] Not really. The software to run the container may no longer be supported, and the people with the knowledge may charge more than most can afford to get it to work.
[2] Something we're dealing with at work these days. Writing stuff for the automotive industry and they expect that any software + tools should work 15-20 years from now. Same level of expectation for availability as any other car part. Now that there is so much SW in modern cars, this is a good opportunity for being a consultant 15-20 years from now!