My initial plan was that Forgecore would feature randomized level layouts. I gave up on that idea eventually because I realized that memorizing the level layout was a big part of what made a run successful. In facts Forgecore levels are procedurally generated using a fixed seed, so that they are the same for every game.

Also all of the graphical elements are procedurally generated, in some cases with a degree of randomness.

While developing Forgecore, I've found that it's better to ask myself a couple of questions before of shamelessly jamming some randomness into the mix.

The first question is: "Is that really random or it actually has an underlying scheme?"

I've spent some time trying to figure out how to distribute trees on a mountain for a certain ending sequence. I've tried randomizing it in various and implausible ways, but it was looking good just one time out of ten. I was overlooking a fact: trees don't grow randomly.

Trees need light and space to grow. I've ended up placing the trees in a perfect brick layout and applying some randomness just to create some noise and don't let it look too perfect.

Which brings me to the second question: "Even if there's no underlying scheme, is that really random or it's better if it's equally distributed?"

Let's continue with trees, but considering a single one. Let's leave aside the look and let's focus on the structure of branches. The algorithm I used is pretty simple: the tree has an amount of "age" or "energy" and every time it branches the amount is split between the branches. It goes on recursively until it eventually gets a branch with just 1 energy.

I wanted to have both branches that split into 2 other branches and branches that split into 3 other branches. On my first approach I've tried to randomize it, but many times I would get too many consecutive split into 2 or consecutive split into 3 and it was looking fairly bad. So I realized that in that case I didn't want to have it really random but distributed equally. I completely gave up randomness and I've opted for a fixed 3 - 2 - 2 sequence. It works quite well because it is applied breadth first and the repeating pattern doesn't stand out.

I can say that my idea on the use of randomness in game development has changed a lot. I'm a long-time fan of classic turn-based roguelikes and I used to think that procedural generation was all about randomness, but now I feel that it has be used with a great deal of care and with a clear purpose. Nobody likes a bunch of pointless noise.