Recently (ok not so recently) I attended a conference by eeegr on decommissioning oil platforms and nuclear power stations. One of the key things that I came away from this was the failure of modern (I mean seventies onwards) companies to keep information. The point being that these companies don't know basic information about the rigs - how many concrete weights there are at the bottom, what chemicals have been dumped. The reason they don't have this are mainly due to the cost of keeping the paperwork.
But they're all electronic now aren't they? A copy electronically and it takes up a lot less space - well, that would be nice now wouldn't it, so why then in practice isn't that happening? Do we all remember the Edinburgh AI Library fire. Sure, the current research (on backup) was saved but why wasn't this information on disk - especially as it's about electronic stuff, why wasn't it stored electronically?
The idea of software rot isn't just caused by technical improvements of the hardware its running on, but also on changing goal posts and adaptation of tools. Suppose a threaded application that used third party library works fine and does not create a race condition, on the changing of this library (for security or otherwise) where the running code changes but the API does not, the call might take longer - perhaps even long enough to suddenly cause race conditions.
And as for pure mathematics - well, how about something as fundamental (and supposedly never changing) as the ratio of the circumference to its radius, and yet over the past few thousand years it did changed. Now, I don't think (and I don't think anyone else does) that the laws of physics have changed, they just didn't have the ability to work it out any better. Now, does that mean that some of the maths that we are doing to day (some of the ground breaking new maths) is also... shaky?