What I mean is: Should I design my levels as I would to motivate a real player to try again and again?
Game balancing is long and time consuming, rare are the commercial games, even with teams numbering in the dozens, achieving it...

What you should try to avoid (if possible) is not to aggravate the player, like if you make a half-second mistake 5 minutes in the game, you have to replay the whole 5 minutes, at which point there is still that half-second where survival is decided... So the first time the player gets killed there, give the player a clue to what killed him (by having the camera look into some "interesting" way, having the boss parade its weakness, prompting a text hint, etc.).
Though what is not easy is that to you, the game developper, that half-seconde decision point may not exist, because you know from where to attack the boss, you know the telltale signs and you know *what* to do (even when you try to play naive... f.i. in AirBlast last year, I found during testing that certain people had a whole lot of trouble surviving enemy missiles in the frontal dogfights, while I personnally was untouchable except by the highest-level enemy AIs... I knew exactly how and when to break off or use decoys, knew the maths behind it, knew you could take advantage of missile inertia, etc. while "normal" players had to discover that by themselves through trial and error).

Or should I design the levels as a "blind rush", where you reach the end without any danger, just like using a cheat mode?
IMO that's a bit as bad as "too tough", except instead of aggravating the player, you're boring the player :roll:

There should be a challenge, but split-second failure shouldn't force you into replaying the same "initial" sections of the game again and again. One way is to give the player "lives", so that he respawns a certain number of times nearby, can try several strategies to beat the hot-spot, and hopefully find a way to beat it.
That doesn't mean he'll be able to complete the whole game that way, but that when the "lives" run out and he has to restart, he'll know how to beat that hot-spot (ie. won't be stuck forever on the same spot).

Having levels "unlock" themselves (for direct play from the start menu) is also good IMO (otherwise the first levels become a chore when you've to go through them dozens of time, however nicely designed they are).
To have the player revisit early levels willingly, you have to change the challenge (harder difficulty level with more enemies, "time trial", different weapon selection, etc.).

All of that takes time - a lot of it - so if you absolutely want the judge (and players) to reach some "special" point, make sure to have a walkthrough, hint, or cheat mode... just in case

So how much time will the judges take to try going through the whole contest entry?
Not sure, though personnally if I have to go through the same lengthy section repeatedly, only to die at the very same spot, without any savegame to reduce the 'lengthy' or hint of a way to beat that spot, I'll likely try to cheat my way through. :lol:

Overall, I guess the maximum amount of time will depend on all the other entries: if almost all entries can be mastered in 1 hour f.i., then an entry that would take 10 hours to master would have to be quite above the rest if it wants those 10 hours to be spent for it.
I say "master" though, rather than gameplay length (which may be infinite): there are games that are easy to master, and that can be (re)played many times because they have a level of challenge sufficient enough not to get bored, but controled enough so that you know you can beat it, if you just pay enough attention.