One of the least appealing aspects of creating an RPG game (tabletop, video, or otherwise), is creating a critical mass of content.

A simple story about a boy seeking vengeance for the death of his family is a good starting point, but very quickly the game master can be bogged down with the creation of minutia. Where did they live? What kind of place is it, and what did they do there? Who were the people who attacked them, and what was their goal?

Aside from details relating directly to the storyline, there is also a very large amount of other detail that might be nice. It's good not to mislead the players, but at the same time not ALL footprints will belong to the villain.

["Paw"] has, in his tenure as a game creator, put together a veritable mountain of charts to fill in these details. And now ["Calin"] is investigating ways to electronically put all this information together in a useful way.

Ideally, a game creator could specify a bare minimum of detail (whatever is relevant to the story at hand), and a world generator could fill in the rest of the environment. Obviously an element of randomness would be necessary, but randomness with a starting seed could ensure that details will not change arbitrarily.

The problem is, how do you begin such a monumental undertaking, of putting together such a generator?

The best place to start is probably to break generation down as far as possible, and move on from there. Here's an example in two dimensions:

drawing:firstexample

We have a line seperating the top from the bottom. The two end points are pre-determined, but the line connecting them is calculated by a computer program. In this case, there is nothing at all random about how it was calculated. If we altered the calculation program to include arbitrary fluctuation, we might get something like this:

drawing:secondexample

If our goal was a terrain generator, we would want to further alter the program to keep the fluctuations within a reasonable distance:

drawing:thirdexample

Now we can assign a few important points like this:

drawing:fourthexample

and end up with a reasonable mountain like this:

drawing:fifthexample

The key it seems, is to first write a calculation that is very mathematical and precise, and then introduce variation such that at any given point the variation is not far from the precise calculation. In this manner we have a result that is neither too arbitrary (like the very spiky image 2 above), nor too cold and uninteresting (like image 1).