Shoals Surfer
Posts: 287
Joined: Friday, 19th August 2016, 21:21
playing optimally: definition
Since dcss is a fully-deterministic game (rng is only pseudo-random, with initial seeds for map gen and combat and such), like chess, we can construct a full decision tree for each tuple of possible initial rand seeds.
Now we mark each branch as OK if it has a valid path to a win scenario (we could complicate and weight based on #turns to win, etc).
Once we have this collection of decision trees, we can determine the best move for any given ongoing game by
1- Find all decision trees that match what the game has encountered so far (different starting seeds may lead to identical game starts).
2 - Take the decision branch that has the greater proportion of OK's vs non-OK's across all the matching decision trees.
That's optimal gameplay. It exists, and it can be quantified.
When it comes to using the hypothetical optimal player as a deciding factor on game design considerations, I see it a bit like the efficient-market hypothesis - theoretically interesting, but wholly inadequate to describe reality. And for the same reasons - people's motivations are not at all rational.