Hey, sorry it took me so long to reply. It's actually a big topic, so I didn't answer right away to think about it first, and only now I remembered that I didn't answer at all. Here is a list of my thoughts:
- Yes, these dungeon notes are too random, can’t argue with that.
- The generator is inspired by “one page dungeons” and they tend to be like that (because of the format). The idea is that one is supposed to connect the dots somehow to see a “story” behind small pieces of text.
- One of the reasons why it doesn’t really work is the generator was made for a jam and I didn’t have much time to implement this part properly*. After the jam ended, it turned out that people were more interested in the hatching and in the way rooms connect.
- *I did try to make this all more coherent by introducing “themes”. For example, a dominant race is chosen for each dungeon, so if it’s a dwarven dungeon, dwarf npcs are more likely to spawn.
- Another (probably more important) reason is that compelling (even if random) vignettes are needed and text generation is a special skill I am lacking.
- I don’t know python, so I can’t comment on your example, but I encourage you to make it into a full-fledged application :)
- Regarding the approach you suggested, I personally wouldn’t adopt it here. It’s a question of “faking vs. simulation” and I am a proponent of “faking”.
- You see, I believe that if a screenshot (e.g. a map) of a complex system (e.g. a dungeon) is needed, in most cases just “painting” such a screenshot is a better way to make it than creating and running a model of the system.
- Models are harder to implement, fine-tune and debug. Models are prone to getting stuck in a limited number of states including bad ones.
- Of course, you can’t do without simulation if not a still image is needed, but a running thing. Or if a still image is expected to be very closely inspected (e.g. like in Obra Dinn).
As I said, it's a big topic :)