[lifeisstillgood]

That last part is key.

I suspect some form of X-prize approach would be worth trying - seed out a large number of small ISVs for government software (from license management to submarine firing systems).

Spend the money not in the requirements (which as you say are usually written by a non developer with minimal experience of the problem - after all if they were an expert they would have other more useful things to do) and spend the money on building automated test rigs.

I know I sound trite, and this is not even the best way (which is to organically grow solutions to problems), but we have a small window - software is eating everything and we will need to write the first generation of software that covers all politics and I damn well want that to be open to everyone to read. The software that runs governments will essentially be law. We should be free to read our laws