Each user's number of "votes" in the last two criteria will be limited to floor(earned_merit / (10 or 250, depending on the criteria)). If you trust more people than your limit, then you will vote for the people to whom your vote will be the most useful, more-or-less.
I wasn't able to find either an optimal or low-error-approximate solution to
this problem. My current algorithm is sub-optimal in general and could produce results uncomfortably far from the optimal solution, but the current data doesn't actually present a scenario where it matters: my current algorithm is optimal with the current data. Long-term, if I can't find an algorithm that I'm happy with, I could make the trust lists ordered as some have suggested.
When building your trust list, I tend to encourage people not to worry about little details like this, and instead just think about the system in broad strokes. If this results in poor outcomes, then that's a problem on my end.
And I want to insist on my suggestion regarding trust: guest should see some trust.
The main reason that I went for this solution rather than forcing custom lists is that I would like to show some trust indicator to guests. But before doing that, I want to see whether these modifications can actually be made to work. If not, then I may go to the force-custom-lists solution, and that's incompatible with guests seeing any trust indicators.