NetHack 4 homepage

About this document: Last modified by Alex Smith, 2013-12-29

Copyright (C) 2013 Alex Smith.

This document is licensed under the NetHack General Public License. See libnethack/dat/license for details.

Explaining the design decisions behind roguelikes seems to be all the rage, nowadays. So here, I'm trying to explain my thoughts behind the philosophy of NetHack and NetHack 4.


In most roguelikes, flexibility is a minor or irrelevant design goal. (Brogue, for instance, is an awesome roguelike, but is intentionally designed to be inflexible; the skill in that game is adapting your character build to the resources you find. That doesn't make it a bad game; it just makes it a very un-NetHack-like game.) By flexibility, what I mean is: if you want to be able to do something, you should be able to do it. NetHack is, and always has been, excellent in this regard; people have thought up interesting new ways to play the game which were not really envisaged by the developers. (Unlike some roguelikes, NetHack's classes are really just starting points, and there are probably too many of them as it is; the issue is that removing or merging any would cause an uproar.)

As such, in order to accommodate everyone, the game needs to be rather easy for experienced players to play without restrictions, so that playing with restrictions (of their choice) is an achievable goal. This is not really a problem; sometimes people just want to breeze through the game and that's perfectly reasonable, sometimes people want a real challenge, and they can do that in whatever way they find most fun. (The game tracks some "conducts" in order to provide a starting point for this. They fall into three main categories: avoiding specific powerful strategies (polyselfless, puddingfarmless, Elberethless); playing with resources limited either in general (wishless, polyitemless) or with specific limitations (artiwishless, illiterate, vegan/vegetarian, atheist); and challenge runs that heavily alter the normal play of the game (pacifist, foodless, permablind, permahallu).) This is the source of NetHack's replayability after ascension; the flexibility given to the player makes it possible to improvise solutions to all sorts of problems that might not have come up before. Even with simple unrestricted play, problems frequently come up that may never have come up before, and require improvised solutions. This is something that I'd like to encourage more of, but it's very hard to balance the game to accomplish that, and I'm erring on the side of not frustrating players.

I'd also like to make a direct argument in favour of choices, even when some of them are usually going to be bad ideas. Although NetHack tries hard to make sure that no course of action is going to be 100% useless (if only due to the presence of unique-deaths competitions where people try to come up with the most creative suicides of characters that they can), something that's 99% useless is perfectly OK to put in the game. There are multiple reasons for this: for people who prefer to play unspoiled, there's the joy of discovery in working out which options are best, and for spoiled players, sometimes that 1% is going to come up, and the game's enhanced by the ability to use unusual solutions to unusual problems. If the best solution to any problem is always an intended solution designed into the game by the developers, the game becomes an issue of memorisation, rather than creativity.

The converse is also true; if there's a choice that's 99% or even 100% of the time correct, it doesn't detract that much from the game. And it probably isn't going to be made in 99% of games, because the player is going to have a goal at reaching a point in the game in which they can make it (and some characters are going to die first); it's serving as an important milestone for those who do that gives a sense of accomplishment (and psychologically, helps cut down on boredom and the feeling of repetitiveness that can pervade long games in any genre). Also, such choices cannot really be eliminated without detracting from the game; forcing the player to take them would eliminate obvious challenge runs (just make the bad choice at that point), and there's probably someone who will try the run and have fun doing so; and nerfing them will make the game more difficult and usually not solve the underlying problem (players will just move to the next-best choice instead, then you'll nerf that one too, and so on). It is important, however, to cut down on newbie traps; if a choice is usually wrong, either it should be obviously usually wrong, or the game should warn people who try to take it.

Finally, to allow this flexibility, the player needs to be able to have a meaningful influence over the development of their character. This is why items like wands of wishing exist. Also, why it's important that the ability to start-scum characters exist, but due to the interface and grindiness issues of that, I'm looking to replace that with some other mechanism that gives the same choices but in a less boring to use way. It's also why different dungeon branches have notably different loot; players who want something in particular should be able to take steps to increase their chance of getting it. Failing to do this is quite commonly observed in roguelikes; it leads to a game where the player has no incentive to go in any direction but the one with the easiest enemies at that point in the game, and as such, the choice is mostly a false one.


There are definitely games whose interest comes from the difficulty of instructing your character precisely in real-time (imagine bullet-hell shooters, for instance). However, games tend to be most fun when they focus on one particular sort of difficulty, rather than making you do multiple unrelated things at the same time; puzzling your way out of a difficult combat is made less fun, rather than more, by having to constantly check for typos as well. As such, in NetHack 4, there's a strong focus on trying to make sure that what the character does is what the player wants them to do. For instance, apparently suicidal actions require confirmation before the character actually does them (with a prefix, not the easily typoed y/n, in cases where the player might be typing movement commands at the time), although the character is allowed to perform them because sometimes they might be important. Likewise, monsters and items whose main interest came from interface issues have been altered to provide an interface-independent effect (with floating eyes becoming mobile obstacles that can't be meleed, and scrolls of amnesia draining skills rather than map memory).

I also try to ensure that information the player needs is readily available, to save time and effort repeatedly asking for it. Almost all monsters have different letter/colour/underscore combinations; the status bar is intentionally mostly grey, with colour for the HP and Pw bars (so that you can easily check your health status out of the corner of your eye); background colours show the remembered locations of traps and stairs, and the attitude of monsters. As such, the need to farlook things should be very low for someone experienced at the game.

There are also interface niceties like travel and autoexplore; however, these are implemented without compromising the corresponding gameplay, so it may often be the case that manual play is better than using the automated commands. It's better to make something like exploration interesting (with nontrivial decisions about which order to clear rooms in, and trade-offs between waiting to heal and pushing on immediately), than eliminate it from the game altogether and tell people to simply tap a key in order to skip past it. The skill in something like exploration isn't currently working as well as it could be, but it's more interesting to try to improve it than just to give up and abandon it to just another task that can be ignored and left to the AI.


I've already covered why NetHack has to be somewhat easy for experienced players. This isn't a balance issue in its own right; there are still plenty of ways to tell players apart, and for a game where people can spend hours or days (or in some cases, years) on a character, win rate for non-challenge games is an uninteresting and reasonably un-fun metric. It's much more important that every game is winnable, than it is that players can easily be distinguished via win rate; playing a game where you're doomed before you start rapidly becomes unfun. For inexperienced players, this is even more important; you can learn to play a roguelike by dying repeatedly to every letter of the alphabet, but that means going through the early game again and again and again and again, getting slightly further each time. And with a game with the scope of NetHack, that's going to lead to frustration and boredom while the player's simply trying to learn the game, which would make it very fun unspoiled.

However, it's just as important to ensure that games do not become routine, and that players are forced to improvise rather than solve the game via memorization and a standard strategy. Some roguelikes do this by limiting the character's options, generating only a minimum resources and leaving the skill in the game about adapting to what the game gives you. This is not a good fit for NetHack, where much of the fun comes from customizing your character; instead, the game is not about adapting to what it gives you, but to what it throws at you. Even after over 10 years with no new versions coming out, interesting problems in NetHack 3.4.3 crop up all the time, and improvising solutions is a major theme in the NetHack community. (A corollary to all this is that problems with only one solution should be avoided, where possible.)

There is an apparent contradiction here. How can we ensure that the game is (almost) always winnable, and yet keeps posing interesting and fresh problems? The solution is to make it easy to back out when in trouble, but at the cost of resources so that the game still retains tension, and also to avoid eliminating strategy from the game (you still have to be prepared, in order to be able to back out, and you're still only going to be able to try a difficult section every now and again). One possibility would be to make it very easy to run away, but having a standard solution to everything difficult is not particularly fun (and additionally, being able to easily run away from a problem often means that you end up never having to face it at all, so you miss the fun of coming back and solving it later). Instead, NetHack has a range of viable escapes, ranging from full-heals to teleports to levports to branchports to digging through the floor to life-saving to Elbereth (which is fine in concept, but which currently tends to lead to tedium because it's simply far too powerful and spammable). As such, a player exploring an area they've never been to before can have multiple tries until they finally figure it out, and yet are never completely out of danger, because they're still losing resources unless they press onwards.

This also means that it's OK if the player can somehow become invincible or effectively invincible; it's a fun goal for some players. However, doing that should take more effort than simply just winning the game; and it should also happen reasonably close to the end of the game. (The game can be counted as won when it happens; you don't want people to expend so much effort into going through the motions to actually win from there.) This goal is harder than it might seem; some players of older versions got infinite stats using a bug, and they still died sometimes (and more frequently still, failed whatever challenge goal they were aiming for). It probably isn't completely impossible, though.

Another balance issue is that of grinding; some players actually enjoy it, somehow (and I don't want to deny them their "fun"), but it's for the best if it never becomes the best strategy, and if the game never encourages players to do it. This is why freshly generated levels have more items than cleared levels, why monsters get more powerful throughout the game even if you camp dlvl 1 (and you're probably going to starve if you do that, anyway), and why farming gives diminishing or even negative returns on your score. Likewise, shopkeepers have only limited gold, and pay sufficiently small amounts for common items like daggers that the gold you'd get is not even worth the time you take to collect them. (And it's why gold drops off in usefulness past the early game.)

It's also OK if not all races or classes or play-styles are equal. (In fact, it would be rather surprising if they were.) At the moment, pure melee is probably the best strategy (at least one top player will make a pure melee build even with wizards), but that doesn't matter; players are still going to play in other ways, such as spellcasting or ranged combat or riding, purely because that's what they find fun (and aren't games about fun)? Again, flexibility is important; if the game were hard enough that perfect play were required to win, then everyone would be forced into the best strategy, which would take away a lot of choice from the game. With slack in the system, even underpowered characters have a real chance at being made viable, and this adds replayability as well as personalisation to the game.

Finally, I try to give meaningful strategic decisions about where to go and what to do next, and what sort of goals to focus on. This is why things like intrinsics and ascension kit items exist; they give the player something to aim for throughout the whole of the early and mid-game (whilst none of them, apart from the invocation items, are completely necessary, to play into the flexibility goal), and they're valuable enough that they encourage the player to take risks to get them. (This is the main purpose of enemies that are very dangerous missing a certain intrinsic/extrinsic but easy if you have it, such as Mordor orcs and energy vortices; the implementation here could do with improvements, but the goal is important enough that even a bad method of achieving it adds to the game.) It's interesting to note that Gehennom currently fails very badly at this goal, and it's one of the main reasons why it's the least fun part of the game (there are a few others). I'm looking for improvements there, but it serves as a striking example of why this philosophy is important.

Need (or lack of need) for spoilers

There's a lot to discover in NetHack; even after playing it for years, and reading all the spoilers you can find, there's still going to be things about the game that you don't know, and can surprise you. (And this is true even of 3.4.3, which hasn't changed for years; even after ten years, new things about it are discovered from time to time.) As such, I try to allow for unspoiled play. NetHack 4 intentionally spoils information that is hard to remember or discover (the relative strength of weapons, the Invocation sequence), or that is grindy and un-fun to find out (which corpses are edible and which will kill you), while leaving other things there to be discovered; this is why the descriptions of items list their most obvious effect, rather than an exhaustive list of everything you can do with that item (as has been seen in some Angband variants, for instance; it's interesting to note that such variants are often described as having "NetHack envy"). The flexibility in the game also allows for unspoiled play to have a decent chance of survival even when it makes mistakes; it works like just another challenge goal. (Note that this doesn't mean that I expect unspoiled players to win at all quickly; it's OK if it takes months or years of effort to win unspoiled.)

However, although some players love the joy of discovery, many don't; and just as the game is designed to be fun unspoiled, it should be designed to be fun spoiled too. Even spoilers tend not to give players a good feel for the game's strategy; knowing all the information, but without practical dungeon experience, tends to get players killed in interesting ways. And players who know all the information about the game will still have to improvise solutions from time to time (or all the time, if they're trying a challenge run that's never been done before); it's fun when that happens, and they'll have all the information that they need to work out what will happen when they do. (This is one of the main benefits of NetHack communities online; the knowledge of 100 people is better than the knowledge of 1, and can lead to creative solutions when otherwise there'd be a vague guess and a disappointing death.)


NetHack is a very large game, and there's a huge amount to remember. As such, consistency in the way things behave in the game helps make playing more fun by reducing the amount of arbitrary lists you have to memorise. This is why making NetHack match real life or common fantasy tropes when possible is useful; you probably already know what elves are like, for instance, and if there's something that an item "should" be able to do, it probably can. (A fun experiment: find someone who doesn't play NetHack, describe a problem you're having in the game, ask them to suggest a solution. Over half the time, it's going to be something that is actually possible in the game, although you may not have the items for it.) Likewise, the occasional bad pun in the operation of an item is there both for the groans when it's noticed, and because someone who discovers it is unlikely to forget it from then on.