Concerns about the dungeon-randomization method!

Nice to hear there are people interested in this.

Perhaps what was not obvious from our presentation was that the dungeon path algorithm described isn't the whole picture.

That path is only the path between two nodes in a much larger graph of nodes with edges between them.

Nodes and edges can be arranged in pretty much any arrangement. We can have multiple paths from start to finish, as you said, and we can control to what extent those paths cross and overlap (or not; they can be completely disjoint too).

So we can make a linear path with nominal wrong turns or dead-ends, but we are by no means limited to it. As a side note, the dungeons in the first difficulty tend to be more linear and straightforward, while in later difficulties they become more maze-like and epic. We're still balancing this.

The real fun, though, is when we have edges that don't represent paths the player walks along.

There's no technical reason we can't have multiple start/end points, but this doesn't work as well thematically, and it isn't obvious how the levels inter-connect. It can also cause confusion about which exit to take.