Is there an (easy) option to make it fail if there are more broken links than there were before, though?

I am not aware of an easy way. But I also think this is not the right way to go. One can split these broken links into three groups:

1. the ones in the implementations-section: you suggested to remove these sections,
2. links to images showing nicely the complexity, like O(n),
3. all the others.

Two first groups are quite easy to fix (the second one just needs a guideline, so all of the formulas will be written in an uniform way). And I think there are not that many in the 3rd. I would suggest no broken links policy.

I am not aware of an easy way. But I also think this is not the right way to go.

Well, the idea was that if someone opens a PR, we would like for CI to check whether they are introducing any new broken links.

Two first groups are quite easy to fix (the second one just needs a guideline, so all of the formulas will be written in an uniform way). And I think there are not that many in the 3rd. I would suggest no broken links policy.

Yes, I agree. Though the policy should probably be "no broken links, and try to use permanent links (such as archive.org) that are unlikely to break". Most links were probably not broken when they were added, but broke later on.

I've opened a PR to update our policy.

I would merge this as it is, remove all of the broken links (as mentioned above: some of them should be easy) and then make this workflow fail if any broken link is found. In the meantime one needs to manually inspect the log of this workflow.