Originally posted at procgen.wordpress.com
This article discusses considerations when you are implementing procedural generation (procgen) systems.
If you are dealing with runtime (real or load-time) procedural generators in a game, you need to create rules in a way to guarantee that the game objectives are not compromised. The right rule set and logic can sometimes be tricky to fine tune and narrow down.
As part of guaranteeing the previous line, runtime ProcGen systems require you to thoroughly test its results. The more freedom you give your ProcGen system to create content the more you need to test it. You really want to make sure the resulting content is plausible, so that nothing weird is generated or something that stops you from completing the game's objective.
The question is: do all these trees look alike? Is this room memorable? Or will the user distinguish similar type content? If you want them to, you should make sure you know what aspects and details help you to achieve this perception. GalaxyKate brought this to my attention and you can read her interesting post on it (and her experience on spore) on her Tumblr here.
[subheading]ProcGen by necessity[/subheading]
There are some types of games that require you to use ProcGen systems, like Infinite runners and Roguelikes, or games that learn and adapt with your inputs and gameplay. There are also other good to have situations like creating special effects or textures (like with Substance).
If you have a ProcGen system or tool, it could help you to prototype game levels that require content in scale, like open-worlds. If you are using ProcGen at design-time you can use the resulting content as your starting point.
Another thing you need to worry about is the sense of reward, if the generated content somehow influences it. Normally, you'll want the game to give you a sense of accomplishment, be rewarding and fun. Again, testing is important. Tanya Short from Kitfox Games talked about it at GDC 2015 (below).