What you describe is very close one of my first jobs when I worked for the government (100 proposals, one week, pick 4 winners, summary comments for all). It's not so hard to pick out the "good, but risky" proposals. (Another way to split up your proposal list is to point out that 80 of the proposals will be a re-hash of the same stuff, 30 of the proposals will be nonsense and 10 proposals will actually be about something unique and relevant.)
The most common reason for a creative proposal failing is simply that the program manager wasn't ready for it. You don't want to surprise a program manager because they have to properly prepare the bureaucracy around them to support your project *before* they get your proposal.
When a review committee makes a decision, there are still several government people who have to sign off on that decision before the money flows. There will always be at least one lawyer and one accountant with veto power over a committee selected proposal.
The last thing a program manager wants to do is end the fiscal year with money in their accounts. That can get them demoted or fired. They meet with their support staff sometimes for a year ahead of reviewing proposals to make sure everyone knows what's coming. Slowing things down, or failing to execute a grant, because of administrative surprises is very, very risky for a program manager. There's strong pressure to select institutions who have already worked with the office, and projects that fit well with the briefings given to everyone before proposals were solicited. For unusual ideas, it's better to convene a workshop and spend the next year developing a program around it (by which point all the usual suspects are involved).
Now it used to be that universities themselves funded research, and government scientists used to have broad authority to assign funding, and defense contractors had to spend 15% of their budgets on exploratory research, and we didn't have postdocs... To change things back requires a lot.