At some point, the immediate cost of fail-and-fix is lower than the total short to mid term cost of not fixing it.

The problem arises when systems are intentionally set up to fail (e.g. USPS pensions, gutting education), because those who set them on a doom course usually do so to offer a privatized "alternative" to pocket the profits.

What course of action doesn't come with a cost? Fixing things well is expensive but fixing things poorly is generally more expensive in the long run.

But the point is that what fixing something well means depends on the context. Total software re-writes are rarely a good idea for large production systems, and software isn't special in that regard.

Yeah, but we don't care about preserving the existing large production system, we explicitly want to switch to a different, more appropriate system.