If you still need to gain things from contests (e.g. I got into contests in high school mainly because Waterloo promised to pay my undergrad tuition if I bring some metals, and did WF so I could go to the US): unless the problem setters have very poor academic integrity and are also very very very good friends with you, there is no such thing as good or bad problems: there are just problems that you lose to, and problems you don't lose to. I'm saying this because the back-end of contest organization is usually held together by spit-and-glue, so to make it through such system, one essentially has to prepare for worst-case scenarios.
If you are getting into contest organization, the one sentence criteria I like to use is "achieve the intended score distribution, while maximizing the amount of additional information that one needs to communicate to someone who didn't solve the problem so they solve it". This in my experience minimizes the amount of post-contest complaining, and gives contestants more faith in the system. There are two exceptions to this:
Knowledge based problems: I like the idea of having a syllabus e.g. the IOI version, because otherwise the escalation is towards turning research papers into problems. Should such escalation occur, you (or your coach) need to be very very very good friends with the problem setters to even have a chance vs. such problems: there are just too many 20~40 line algorithmic gems that took way too long for people to figure out. I actually don't mind what the syllabus consists of, as long as it's learnable in a reasonable amount of time (6-10 months), and clearly communicated.
Overlaps with math contests: by now, there has been many on-site programming contest problems that were only solved by contestants w. IMO experience. So I've more or less accepted that such overlaps are inevitable. At the end of the day, I feel it's an issue with the ease of problem setting: there are just way more math contest problems out there that can converted into programming (e.g. most of Tugor non-geo probs). Maybe here the solution is to have programming contests that are explicitly more math focused, to double as math contest training as well. Here I find it a bit funny that attempts at making math contest online judges has yet to work out, while programming contestants seem to get better and better at math contests.
Let the flame war begin.
achieve the intended score distribution, while maximizing the amount of additional information that one needs to communicate to someone who didn't solve the problem so they solve it
This is honestly the best line I've ever heard. It is otherwise so disappointing and discouraging for someone to explain a solution in one sentence to a problem you didn't get.
If I manage to UTFG a solution to a problem, is it bad or not and why?
You can explain in one line. Google problem/technique.
UTFG = look it up on Google?
I'd classify that as knowledge based, so it very much depends on whether it's in syllabus.
Also, programming contests are unique in that the problems aren't given formally, but with the most absurd/whimsical stories possible. So in my experience, problems can usually be disguised to the point where they can't be Googled.
I also feel in cases where obscure problems are Google-able, it's usually only those who know the topic that know what to search for. It's actually pretty amazing/sad how often one spends forever trying to re-derive something that's already in the literature when doing algorithms research.