Unfortunately, our group did not benefit too much from play test. We were late on development for the first one(but did have a bare bone prototype to show), and a large bug prevented us from showing anything worthwhile during the second one. But, we did have some people try it during the beta once the game was stable.
In our initial prototype, the player was always pointing up on the screen, and rotation would result in the camera rotating to maintain the player’s position. It was meant as a way to disorient the player and lean into the aesthetics of mystery and dread of the game.
It turned out to be a horrible idea. Players would get sickened by the camera movement, which, in retrospect, should have been obvious. Did you ever see a game with top down view with the camera rotating to get the player always facing one direction?
The fix was easy, we simply went for a non-rotating camera, as is conventional wisdom, like in all the top down zelda game for example.
It is easy when you are working on something and see it built for the ground up to forget how somebody not familiar with the material may react to it. We made the rotating camera to disorient the players and it turned out we should have bothered.
Our game had a fairly minimalist style and a narrow view window. Plain rock make for poor landmark to assist player in navigation and we discovered that players were getting lost a lot more than intended.
We only found a partial fix for this. We tried to have the layout of the level be essentially converging corridors. This way the players would feel like they were getting lost in the twist and turn of the reefs, but they in fact were always going forward. It worked moderately, as players would some time backtrack through a different corridor. A better fix could have been to have unexplored portion of the map that lead backward being turned into dead ends off screen, this way the players cannot backtrack for long, but the benefice of this trick would stay.
Playtest should never be overlooked. In fact it is likely that our lackluster data set in that area made us miss some of the glaring problems with the prototype until it was to late to change course.