It's also not done because it is mathematically impossible to certify software as 'bug-free' in the general case.
Software isn't like civil engineering where you can mathematically prove that a design is sound.
Software isn't like civil engineering where you can mathematically prove that a design is sound.
All we have in engineering (non-software) is something like safety factors and confidence, and this is done with (usually) rigorous mathematical models as well as loads and loads of testing to fill in the gaps of mathematics (think unknown constant/parameters, assumptions, etc).
None of this is impossible to do for software. There are systems that enable one to do easy/entry level verification (such as something like TLA+), to much more complicated reasoning (something like COQ). This will allow the system designers to gain confidence in if the system will work and gain understanding about under what scenario they will fail. Contrast this with the existing software landscape, which is mostly, at least from my perspective, just let me write some stuff until things do approximately what I want. Even at the top of the ladder, I feel the tests conducted are "adhoc" at best and with none of the rigours that you associate with traditional engineering fields.