There is old Unix software that can run on modern Unixes, but it depends heavily on not doing much I/O. Once you introduce an I/O requirement, failure is imminent - the protocols change, hardware gets accessed with a new method. You end up needing some kind of emulation. The same was basically true of DOS apps running natively on Windows - it would probably run your old Lotus 1-2-3 or WordPerfect in a window, but not games. Windows itself has had plenty of API churn(albeit with an unusually strong compatibility story), but a large share of new Windows apps are actually Electron wrappers.
The glide slope theory isn't perfect because the life cycles do seem to depend on what you look at specifically, and there may have been a slowing of the curve with greater reliance on open-source and Internet technology. However it does seem to be predictable in describing the limits of software platform economics.
A series of small incremental steps is vastly superior to a complete rewrite, or massive re-engineering project from the ground up every 10 years.