It's always worth being reminded that the game you think you're playing (or designing for that matter), isn't the game you are actually playing. And it certainly isn't the game that the people at the top level are playing.
What do I mean by that? A canonical example is chess. There's two levels to learning chess: the rules of the game, and the consequences of those rules. The first takes maybe five minutes to master, the second takes a life time.
With that in mind, have a read of the following thread on Google about the strategy of playing Angband, written in a somewhat clipped style. I'll quote some of Eddie Grove's initial suggestions to discuss further:
To Win: kill Sauron and Morgoth.
need enough speed they are unlikey to kill you with double move.
need lots of hp and lots of healing, amount of healing depends
on hp and offense. Basically, game is about reaching 18/200 CON
and getting speed+20.
Every additional turn played lowers your chances of winning.
Learn wrong lessons before 2000'. Qualitative change in game.
Time spent before 2000' tends to reinforce bad habits.
Tedium and/or attention might be treated as resources.
Game is about multiplying bonuses and increasing stats.
Offense with avoidance is more important than defense.
Try to ignore monsters with no drops, but must kill
breeders, try to kill fast monsters rather than leave behind.
Free loot on the floor, try to keep as high as possible ratio of free loot vs
monsters killed, so no search mode and try to rest less for less monster regen.
Early game is all about selling. Do not test wands, want money
for each charge. Always id them. Only ones to keep are -lightbeam
and -trapdest and -stone2mud and -telOther. Otherwise, better to
sell for money for enchant scrolls for longbow.
Experience is unimportant. Comes automatically with survival.
Don't kill for exp unless you can do so quickly without using
consumables and not taking enough damage/mana that you would
consider resting. Exception if you will immediately gain a level.
than to spend two hours clearing an orc pit.
Now if you've played Angband but never got very far, the above information is vital. But at the same time, nothing in the game conveys this to you. In fact, much of the user interface of the game is counter intuitive in this sense. The experience counter and the reward of gaining levels implies that experience is important. The game reinforces bad habits, by allowing you to scum for items and gold forever, without reinforcing the good habits that Eddie highlights.
Angband is apart from other roguelikes, that it is much more about strategy and tactics than learning the game environment. In Nethack, a well-written spoiler could potentially help you win the game the first time you play. But in Angband, it is as much about recognising when you are in trouble, as having perfect information about the underlying rules. This comes directly from the hit point mechanic. If you have 600 hit points, 599 damage in a turn is inconsequential (providing you have a 100% reliable escape mechanism), 601 ends the game.
The other recommendations that Eddie makes flow from this, and his preference for winning the game in a timely fashion (also knowing as diving) as opposed to playing it forever (which is equally possible, and sure to be an issue as games take on more procedural content generation techniques).
But looking at the above list, my first instinct is want to change parts of the game to address these problems, because much of the flavour of the game is lost. But it's not fair to call these recommendations problems: they are consequences, just like the consequences of the rules of chess. And as a game designer, you can't change consequences, you can only change the rules that lead to them. And there is only a loose linkage between the two.