[thread necromancy - resurrecting a thread that's been inactive for over 5 years.... ]
decsystem10 would agree with you all the way about levels that require users to "build something realistic" to solve the problem as we would in RL.
I do too - most of my solutions start as RL contraptions.
But, I also give kudos to those skilled players who can use the darker aspects of AR to craft unrealistic but nevertheless awesome solutions. Even in AR2, where some of the darkness has been removed, it's still possibly to create mechanical launchers and breakers and strange mechanisms (like walkers and perpetual motion machines) that can solve challenges very cheaply indeed.
There have been a few discussions about what makes a good level. There is a thread somewhere - possibly the one referenced early in this thread - about the various types of challenge, and what makes them good or bad, but my present moderator status makes it impossible to find amongst the spam I get to see.
Plus, as time goes on, it gets harder to find new challenges that haven't been done before.....
There is a general consensus that randomness is a bad thing in level design. If the solution is so variable and random that getting a nice solution is pure fluke, rather than user skill, then it's not a good level. It's one thing to know what you want to do, and be able to build that successfully, but if strange and "unpredictable" events cause that to fail, then the level is less fun and more frustrating. This randomness is not quite the same thing as randomity - which is a deliberately introduced, controllable variability, intended to test how well a solution copes with subtle yet predictable variation in the level parameters. Sometimes, a level has unexpected randomness - usually where there's lots going on, and especially when the dillo(s) begin overlapped with solid structure and anchor points.
Which is a different thing from "robustness" - a solution design that can cope with variability and randomness ("randomity"), and works reliably no matter what goes on in the level. Robustness is a desirable thing for RL, but usually in AR it's the incredible balancing acts and impossible feats of timing that provide the exquisitely elegant solutions for us to marvel at. Such solutions only work for the one set of variables they succeed with. Change any one thing, no matter how slightly, and they will fail. So it's a double edged sword, because robustness usually has predictability and "boring" as companion descriptions. They will cope with randomity, but they will not look inspiring.
It is extremely hard to create levels that are fun and challenging, where users can demonstrate their technical prowess without any "fluke" factors. Ideally, solutions can be pre-planned and executed cleanly, without being derailed by random variability in the solution. If changing one thing causes many other things to change, it becomes a frustrating tweak-fest to find the combination that works. Which is no fun at all. A robust solution that works regardless of variability will be quick to build, and probably appear clumsy, because it works by brute force rather than finesse.
But hunting for the perfect combination where the solution can be predictably tuned is a different matter. For me, that's where the holy grail of level design resides.
Randomity would not improve the challenge for me. Instead, I'd get frustrated with the risky gambits and drop back to the reliable, proven methods. I'd get a robust solution, but wouldn't be satisfied by it. Because some of the most extraordinary challenge solutions are ones that defy the impossible. That achieve the incredible, are simply unbelievable, and could not exist in a level subjected to randomity.