TheMeInTeam wrote:Playing well SHOULD prevent dying.

I think I did not explain myself clearly enough (or maybe I am wrong). I try again.

I know that deaths with very good play are really, really rare in crawl. I will argue that they are too rare.

I will call a strategy which leads to literally 100% win rate optimal, since I only care about win rate now. (Of course maybe no such strategy exist: then the one with the highest achievable winrate is optimal.) I do not know if anybody knows what an optimal strategy is in crawl - I have absolutely no idea. I merely know that some moves are better than others.

If in a game somebody knows an optimal strategy I think it will not give great replayability for her. There must be some tension not only in the form of a one-armed robber game, where simply rolls determine the outcome (the game will degrade to this if you follow the optimal strategy) but also from not knowing which moves are the best.

Chess and go are good games because nobody really knows what the optimal strategy is. (Yes, we made AI that beats every human - but most likely they still do not play optimally.) And more importantly: I do not know what the best strategy in go is, therefore I keep wondering in every game what to do.

Now how can we get whether your requirement ("Playing well SHOULD prevent dying") is true for a given game? I think the only answer is to make the game shallow, that is, that we

do know what an optimal strategy (or at least a strategy we can show is very close to optimal) is, and show that it always (or not always) wins. So we know your condition is true/false for a given game because we know too much about it: it is not as interesting to play anymore!

A good game for me is one where there is a place to improve, so when we are not really sure about the answer for your question.

And I think all of this is still true if we replace "literally 100% winrate" with a winrate which is very, very close to it, since as humans we do not play that many games to really make a difference.

Of course, this goal is not easy to achieve. It is easier to create a game which is more shallow, so there clearly exists an optimal strategy, but make it hard in different ways:

1. More opaque: the game does not gives enough info about the task to the player (spoilers needed)

2. More obscure: there is arbitrary complexity like 500 different monsters with 500 different effects which mostly differ by some numbers, so you need to calculate a lot if you want to know the correct action.

These are not good. A good goal for a game I think:

1. Transparent: gives you info about the game itself, no spoilers needed

2. Simple: the required info fits into a human's head, so no lexicon and constant rule checking (xv) is needed

3. Hard: nobody knows what the optimal move is, so nobody can achieve true 100% winrate

I think that this hard goal is easiest to achieve (or get close) with a random game that does

not have a known strategy that gives you 100% winrate. In other words, we do allow deaths with "good" play exactly to create a game where we do not know what the best strategy is: maybe these deaths could be reduced even more.

Now lets look at crawl for some dangerous situations: centaur/killer bee etc. on D2, or Grinder on D3, etc. I do think that there is a non-zero probability that a very good player dies in these situations. However, by allowing it we also created a race: which player can minimalize this probability? Can we find a strategy that gives even better results (higher winrates)? If we would know the answer for your question, that would mean that we know a strategy that always survives these situations, therefore the game is not very interesting anymore.