> incremental improvement often fails. why does it fail? I don't see why this has to be a rule.
a frequently annoying counterexample is when they yet again reinvent the wheel, include a new "compatible" implementation of all the old features and some new features, all based on some better design - and most of the old bugs are gone, lots of things just work, lots of new stuff even - but lots of stuff that used to work is now bugged also.