Hello Jeff,
your generation method seems to be a very efficient approach to the very big task of automated telling of game histories. I'm actually working on something in a similar space myself — a tool for developers to help keep a game's history consistent as it grows — so seeing your approach was genuinely interesting to me.
As I understood your system: your story consists of single episodes that are of a certain type (war, natural disaster, peace, ...). The probability of choosing a certain episode type is at first equally distributed, and later the chosen episodes influence future episodes by adjusting these probabilities to create a story arc.
How did you decide the amount of influence that certain episodes have on future episode type probability? Has a generated history ever produced something that didn't actually make sense — and if so, what did you do about it? How does this history influence CultGame during gameplay?
