Meele damage is way too random

[kuniqs]

I have bad memories about fighting with giant spiked clubs.

The meele (ranged?) damage formula is veeery random: it compresses a zillion factors into MaxDamage, and then rolls 1dMaxDamage.

Spells, on the other hand, roll lots of dice. Lehudib's Crystal Spear rolls more than 10 IIRC. There are exempts of course, like the lightning spells.

I think that's what makes spellcasters 'easier': you can expect a very high probability that killing the same monster with the same spell will take the same amount of tries, on average.


What I propose: instead of rolling for meele damage and AC one time, do it 3 times and average the results.

[mps]

In general, damage formulas in crawl have way too much variance, which is made worse by the fact that the median damage is often zero and the max, well, surprisingly high. Even totally flat, deterministic damage would be better than the current situation. From what I can tell, this is a common criticism of crawl, but I don't expect it will ever change.

Generally, damage formulas should be dominated by constant terms or be computed with large numbers of dice, not few dice with large numbers of sides.

[Sprucery]

Even totally flat, deterministic damage would be better than the current situation.

I strongly disagree with this. Randomness is good. The level of randomness can be adjusted, if need be (I don't really have an opinion on whether there's too much variance currently or not; hasn't bothered me).

[Sandman25]

Even totally flat, deterministic damage would be better than the current situation.

I strongly disagree with this. Randomness is good. The level of randomness can be adjusted, if need be (I don't really have an opinion on whether there's too much variance currently or not; hasn't bothered me).

Randomness is good only when you know its range and there are only 2 ways to learn it: by surprise which can result in painful deaths and by fsim.
The worst example I could find - Azure Jelly,4 usual attacks of 1-12. Almost ignorable by heavy armour characters, right? I don't think so, two of those attacks can deal extra 15-44 (!) damage each but only if the player received some positive damage from previous normal attack (Adds HD + 1d(HD*2) - 1 cold damage). There were many threads about unexpected deaths to Simulacra who have similar attacks.

[Sar]

(Also it's spelled "melee".)

[mps]

Even totally flat, deterministic damage would be better than the current situation.

I strongly disagree with this. Randomness is good. The level of randomness can be adjusted, if need be (I don't really have an opinion on whether there's too much variance currently or not; hasn't bothered me).

Bold stance: Randomness is good. Who could argue with that? Conversely, that's exactly the opposite of what I meant, which is clearly "randomness is bad."

I like how you follow up by saying you have no real opinion on the question at hand...

[Sprucery]

"Surprise which can result in painful death" used to be a defining aspect of roguelikes :)

Personally I have never checked the damage ranges for azure jellies or simulacra. I have always just treated them as things that can deal a significant amount of cold damage. It's true, of course, that the first time you get hit hard can be a surpise.

[crate]

If you make crawl combat deterministic, then it's pretty boring. (I've been over this in the past and I'm not going to repeat it.) At the other extreme, if you make crawl combat as random as possible, that's also pretty boring (there's basically no skill in it).

This isn't a case where there is clearly a correct amount of randomness. Having some is important, having not too much is important. You are going to need a much stronger argument to support changing it than "randomness is bad" because that's not even true ... and in the absence of a reason to change it, sticking near the status quo makes sense for several reasons: it's less work, and that is the way it's always been. (The latter is a legitimate argument assuming that one goal is to retain the crawl playerbase, and it's one point along the random-notrandom spectrum that has been proven to more-or-less work just fine.)

[Sprucery]

Bold stance: Randomness is good. Who could argue with that? Conversely, that's exactly the opposite of what I meant, which is clearly "randomness is bad."

Sorry, I lost you there. You said you would prefer deterministic damage over the current situation. I would not prefer deterministic damage over the current situation. That is what I wanted to express, be it a bold stance or not.

I like how you follow up by saying you have no real opinion on the question at hand...

Glad you like my writings. I had an opinion on one particular thing you said. Deterministic damage would be worse than the current situation, imo.

[XuaXua]

There was a thread on this a few months back...

[Lasty]

Highly variable damage is a good thing in the context of Crawl, because it means that you rarely have the luxury of being actually safe. The possibility of high damage spikes ensures that monsters who have those spikes can never be taken for granted. On azure jellies and similar monsters that's a particularly good feature, since they're meant to be dangerous indefinitely, but it's a bit problematic on simulacra, since they're mostly slower than the player, leading to kiting tactics. If enemies did fixed damage, you'd be able to predict each fight much more closely, making fights much less interesting.

Spells having a different damage model is also interesting because it means that spells and melee feel somewhat different.

[mps]

Wow. No, I did not state a preference. I drew a comparison. I said the current situation is worse than an absurd situation so that the current situation is in need of reform. Saying "I strongly disagree, randomness is good" is a silly response.

Highly variable damage is a good thing in the context of Crawl, because it means that you rarely have the luxury of being actually safe. The possibility of high damage spikes ensures that monsters who have those spikes can never be taken for granted. On azure jellies and similar monsters that's a particularly good feature, since they're meant to be dangerous indefinitely, but it's a bit problematic on simulacra, since they're mostly slower than the player, leading to kiting tactics. If enemies did fixed damage, you'd be able to predict each fight much more closely, making fights much less interesting.

Spells having a different damage model is also interesting because it means that spells and melee feel somewhat different.

If you want the player not to be safe, which is a good goal, reliably dangerous monsters are much better than monsters that are usually not too bad but one in ten encounters suddenly hit you for huge damage.

Anyway, re: an old thread, I go back to my original comment, which is that this is a common criticism of crawl and the prevailing philosophy on it is long-standing. It should change, but it probably never will.

[Sandman25]

"Surprise which can result in painful death" used to be a defining aspect of roguelikes

Personally I have never checked the damage ranges for azure jellies or simulacra. I have always just treated them as things that can deal a significant amount of cold damage. It's true, of course, that the first time you get hit hard can be a surpise.

A good roguelike can be won without spoilers (yes, I know crawl was won this way at least once).
Azure jelly can deal 136 damage per action (272 damage per turn). Do you suggest to retreat at 272 HP when you have no rC+ as Na in robe (GDR 0, AC 40)? Ok, you probably suggest to have rC+. It still can be 132 HP per turn (rC+, GDR 50). Do you retreat from Azure Jelly when you are Gr with 132 HP, AC 100, GDR 50 and rC+?

[Sprucery]

Saying "I strongly disagree, randomness is good" is a silly response.

OK. I try not to be silly. This is not CYC after all. "I strongly disagree, the current level of randomness is better than deterministic damage."

[Sprucery]

Azure jelly can deal 136 damage per action (272 damage per turn). Do you suggest to retreat at 272 HP when you have no rC+ as Na in robe (GDR 0, AC 40)? Ok, you probably suggest to have rC+. It still can be 132 HP per turn (rC+, GDR 50). Do you retreat from Azure Jelly when you are Gr with 132 HP, AC 100, GDR 50 and rC+?

I try to kill azure jellies from afar.

I also don't play optimally and probably I'm in every game in some situations where I could theoretically be killed in one turn. I definitely never calculate possible damage ranges.

[Sandman25]

I try to kill azure jellies from afar.

I also don't play optimally and probably I'm in every game in some situations where I could theoretically be killed in one turn. I definitely never calculate possible damage ranges.

I suspect all players play this way.

[njvack]

I sometimes check Gretell when I think I'm up against something that might surprise me with lots of damage. I've been surprise-killed by things like ettins before.

FWIW, I kind of agree that damage can be a little more spiky then I like. That feeling of "OK, I've melee'd these guys with this character before and it's no big deal" followed by a huge, fatal damage roll frustrates me. It's not that I wanted that last guy to hit me less hard, I wanted those first five to hit me harder.

There are huge number of combats in Crawl where you are actually safe. Almost all of them.

[and into]

Deterministic combat (either in the manner of Westnoth or Desktop Dungeons) would be a bad thing for Crawl, I hope that much is obvious, but that leaves open the question of how much variance, and what kind of variance, is best for the game.

I also think the status quo is pretty good, and there are better measures of this than some very abstract argument about randomness. What is the ratio of "fair" deaths (player made a clearly identifiable mistake that directly contributed to death that could have been avoided) to unfair deaths? Unfair deaths do occur in Crawl, but I think the ratio is very good. The vast majority of deaths involve bad luck and mistakes, in some proportion. I think that's a good place for Crawl to be balanced around.

That being said, most of the examples people have brought up (azure jellies, simulacra) involve specific enemies that arguably have problems. I like azure jellies, and I don't think they need to change, but I can understand the argument that their damage can be very "burst-y" in a way that surprises high AC characters in particular; I think it is fair, especially since these guys are almost exclusively Slime-only enemies, but I understand where the argument is coming from. Simulacra are more problematic for a number of reasons, most of which were covered by Lasty, and yeah those need some work. (Simplest solution is just to remove them as enemies, but another relatively simple solution is to make simulacra faster than their base type, though for balance reasons one might then need to reconsider the damage they do, and the depth at which they can spawn.)

[njvack]

What is the ratio of "fair" deaths (player made a clearly identifiable mistake that directly contributed to death that could have been avoided) to unfair deaths?

I don't mean to suggest I think spiky damage is unfair; I just feel a little frustrated when my thought is "oh, if I had looked in gretell to see what this number was, I would not have stood next to this monster."

I also agree that the status quo is pretty good; this feeling does not happen very often. After a while my brain build up a list of "monsters that occasionally hit like trucks."

[Sandman25]

What is the ratio of "fair" deaths (player made a clearly identifiable mistake that directly contributed to death that could have been avoided) to unfair deaths? Unfair deaths do occur in Crawl, but I think the ratio is very good. The vast majority of deaths involve bad luck and mistakes, in some proportion. I think that's a good place for Crawl to be balanced around.

I agree with the rest of your post. I think ratio is unimportant here. No matter whether we have 1 unavoidable death in 5 or 5000 games, we still should do out best to remove/fix it.
Unexpectedly high damage is a kind of unavoidable death IMHO.

[Sar]

  Code:

Balance
========================================
The notions of balance, or being imbalanced, are extremely vague. Here is
our definition: Crawl is designed to be a challenging game, and is also
renowned for its randomness. However, this does not mean that wins are an
arbitrary matter of luck: the skill of players will have the largest
impact. So, yes, there may be situations where you are doomed - no action
could have saved your life. But then, from the midgame on, most deaths are
not of this type: By this stage, almost all casualties can be traced back
to actual mistakes; if not tactical ones, then of a strategical type, like
wrong skilling (too broad or too narrow), unwise use of resources (too
conservative or too liberal), or wrong decisions about branch/god/gear.
The possibility of unavoidable deaths is a larger topic in computer games.
Ideally, a game like this would be really challenging and have both random
layout and random course of action, yet still be winnable with perfect
play. This goal seems out of reach. Thus, computer games can be soft in the
sense that optimal play ensures a win. Apart from puzzles, though, this
means that the game is solved from the outset; this is where the lack of a
human game-master is obvious. Alternatively, they can be hard in the sense
that unavoidable deaths can occur. We feel that the latter choice provides
much more fun in the long run.

[Sandman25]

Sar,

The section does not mention that crawl is trying to increase number of unavoidable deaths to increase fun. My understanding is that devs are trying to decrease the number while keeping fun. Id weapon on sight, no gnoll packs on D1, display spells for monsters, display hex success chance etc.

[mps]

Deterministic combat (either in the manner of Westnoth or Desktop Dungeons) would be a bad thing for Crawl, I hope that much is obvious, but that leaves open the question of how much variance, and what kind of variance, is best for the game.

I also think the status quo is pretty good, and there are better measures of this than some very abstract argument about randomness. What is the ratio of "fair" deaths (player made a clearly identifiable mistake that directly contributed to death that could have been avoided) to unfair deaths? Unfair deaths do occur in Crawl, but I think the ratio is very good. The vast majority of deaths involve bad luck and mistakes, in some proportion. I think that's a good place for Crawl to be balanced around.

You're turning it into a normative question, which is obviously going to favor the status quo among present company -- i.e. players who are sufficiently satisfied with it to play avidly. It's okay to die to bad luck wrt to spawns, dungeon layouts, even item generation, but bad luck in damage rolls should almost never be the deciding factor.

I say again, this is one of the most cited problems in crawl among people who don't play but have played. I think it might help to take a step back and think about other examples of games where you can face the same situation twenty times and win easily in all but one, where you suddenly go from max health to dead in three turns with the decisive damage loaded heavily into the last turn. Do you think those situations make sense?

That being said, most of the examples people have brought up (azure jellies, simulacra) involve specific enemies that arguably have problems. I like azure jellies, and I don't think they need to change, but I can understand the argument that their damage can be very "burst-y" in a way that surprises high AC characters in particular; I think it is fair, especially since these guys are almost exclusively Slime-only enemies, but I understand where the argument is coming from. Simulacra are more problematic for a number of reasons, most of which were covered by Lasty, and yeah those need some work. (Simplest solution is just to remove them as enemies, but another relatively simple solution is to make simulacra faster than their base type, though for balance reasons one might then need to reconsider the damage they do, and the depth at which they can spawn.)

I think it's really weird the way this kind of reasoning works. "Sure, simulacra have overly spiky damage, but we can compensate by making them faster and adjusting the damage or just remove them." Here's an idea: How about change the damage formula? Ettins and two headed ogres are another example. Most of the time, they barely even hit you, then one time they hit you three times in a row and you die. This makes no sense thematically or from a design perspective.

It cuts the other way too. Sometimes you miss or do only chip damage for turns and turns, especially in the early game. It's absurd that you can sit there exchanging blows with an adder for 20 turns as a melee character with reasonable skills and equipment for the level. This is a fairly common situation as far as I can tell.

[and into]

What is the ratio of "fair" deaths (player made a clearly identifiable mistake that directly contributed to death that could have been avoided) to unfair deaths? Unfair deaths do occur in Crawl, but I think the ratio is very good. The vast majority of deaths involve bad luck and mistakes, in some proportion. I think that's a good place for Crawl to be balanced around.

I agree with the rest of your post. I think ratio is unimportant here. No matter whether we have 1 unavoidable death in 5 or 5000 games, we still should do out best to remove/fix it.
Unexpectedly high damage is a kind of unavoidable death IMHO.

I like that bit from the game design philosophy of Crawl that Sar posted, but I agree with what you say, in principle. Truly unfair deaths are not fun, and it is a good goal to remove them. By contrast, tough but fair challenges—which we might define as situations in which your character can just survive with careful play, but will potentially die if you make even relatively minor misplays—are very fun.

The problem is that most features in Crawl, even ones that are widely seen as problematic, are not unambiguously 100% terrible. If there is an enemy that causes an unfair death perhaps once in 5000 games, and "tough but fair" situations very regularly (say, in about one-third of the games in which it is encountered), then removing that enemy will actually be a net loss for the game.

[Sandman25]

The problem is that most features in Crawl, even ones that are widely seen as problematic, are not unambiguously 100% terrible. If there is an enemy that causes an unfair death perhaps once in 5000 games, and "tough but fair" situations very regularly (say, in about one-third of the games in which it is encountered), then removing that enemy will actually be a net loss for the game.

After thinking about it, you are right. It's better to unavoidably lose 1/5000 game than to have 4999 boring games.

[crate]

btw in case no one else noticed, all the "spiky damage" things being brought up are af_cold
note that af_cold is not actually melee damage at all! it works differently
now, it seems clear to me that the devs have decided that af_cold being this way is fine, because it's been this way for forever (except the time when it didn't work at all)
but if it really bothers you enough, perhaps you should focus on that
there are some changes I would personally make to af_cold but I think they are unlikely to happen and it's not entirely on-topic so I'm not going to bring them up here

---

As an aside, I suspect in practice player melee damage is actually less random than people think (I'm assuming here that you already hit your attack; I don't really want to comment on accuracy here but it should be obvious that if you want accuracy to matter there's going to be randomness involved if you're going to keep each attack independent of previous attacks) (I suspect monster variable hp is actually at least as large of a factor) but I can't say for sure without seeing a histogram of damage dealt, or at least something like a standard deviation. The extremes are often very different from the average, but I'm not sure that matters much because they're also extremely rare. And it seems to me that the OP was talking about player melee damage (whereas the rest of the posters seem to be talking about monster melee damage, which, since this is crawl, is a different thing--and, I think, more random!).

[Bloax]

I have bad memories about fighting with giant spiked clubs.
What I propose: instead of rolling for meele damage and AC one time, do it 3 times and average the results.

You might notice that I've done exactly this in my little minorly tweaked custom build.
I really like it, especially in conjunction with making accuracy more deterministic (and improvable) at the same time. (Try ogres!)

[and into]

You're turning it into a normative question, which is obviously going to favor the status quo among present company -- i.e. players who are sufficiently satisfied with it to play avidly.

I play Crawl because I like the game overall. I was offering a specific, concrete standard by which to judge how problematic the variance in Crawl actually is. You can disagree with that standard, of course, but I don't really know what you mean by saying I am "turning it into a normative question." If you convince others of your viewpoint, it could become the new normal.

It's okay to die to bad luck wrt to spawns, dungeon layouts, even item generation, but bad luck in damage rolls should almost never be the deciding factor.

What? In almost every case in which someone dies, it is because they didn't have enough HP compared to incoming damage. If I try to think of a situation in which a bad spawn is the major deciding factor, I think of a creature with "death gaze" that kills you if you lack a "death gaze" specific resistance. That sounds terrible. Maybe you meant something different.

The good thing about having so much stuff rolled is that the player has a lot of influence (both strategically and tactically) over how combat develops, without combat being deterministic.

I say again, this is one of the most cited problems in crawl among people who don't play but have played. I think it might help to take a step back and think about other examples of games where you can face the same situation twenty times and win easily in all but one, where you suddenly go from max health to dead in three turns with the decisive damage loaded heavily into the last turn. Do you think those situations make sense?

The take away from that situation should be, "Damn, I guess ogres can hit for a lot of damage when they really connect," or "even vanilla orcs are dangerous if they have a dagger of electrocution/distortion," which is information you can then apply the next time you play. (Note that brands are IDed on sight now, so one source of large damage spikes in the early game is much less likely to be applied to you suddenly, with no warning and little recourse.)

Ettins and two headed ogres are another example. Most of the time, they barely even hit you, then one time they hit you three times in a row and you die.

Don't let Ettins hit you three times in a row. An early Ettin or two-headed ogre is absolutely a worthwhile situation to use your consumables to get away safely, if they are too dangerous toe-to-toe and you don't (yet) have the ability to take them out at range. (Or your positioning happens to be bad when you first encounter them.) I cannot see how Crawl can be a challenging game if there were no enemies capable of killing you when given three full turns to beat up on you while you fail to kill it or escape. The Ettin cannot move if you don't; you have time to stop and think based on the damage you've taken.

This makes no sense thematically or from a design perspective.

Yes, m'lord.

I think it's really weird the way this kind of reasoning works. "Sure, simulacra have overly spiky damage, but we can compensate by making them faster and adjusting the damage or just remove them." Here's an idea: How about change the damage formula?

Specific problems with simulacra can be fixed by changing simulacra. Or maybe by changing the way (a specific kind of) cold damage works in Crawl. Those would directly address the issue with that particular enemy. If there is a window that doesn't fit well into the frame, I'm going to fix/replace the window, rather than tear down the house and construct a new house that fits around the window. You obviously think there is some inherent flaw in how damage is calculated in Crawl, writ large. I'm sure things could be improved, and I'd actually be curious to hear concrete suggestions from people who actually have concrete suggestions to offer (and I know those who actually develop the game would be open to that, too). But I don't think damage is fundamentally broken, and simply repeating your conclusion isn't going to convince anyone of anything.

[crate]

Rolling for accuracy 3 times doesn't actually do anything, by the way. It just changes the actual hit rate to a different number, since the result is binary (hit or miss). For starting accuracy >50%, this means you actually hit more often than you're supposed to, and for starting accuracy <50% you hit less often than you're supposed to. One could even say this is actually more variation (mathematical sense) than before, since your accuracy distribution is now more skewed toward the extremes (though this is not true if you look at a subset of accuracy values, say from 50% to 100%--in which case there is less variance).

Since in crawl accuracy tends to be >50% in most cases, what you're doing is just making everything hit more often (and making the difference between two displayed accuracy numbers smaller--this is the >50% subset above). This isn't necessarily unreasonable but I think you have to figure out a good case for it if you want it to be changed.

If you want deterministic accuracy what you really mean is you want attacks to have a memory of previous ones: if you hit on your previous attack, then your current one is more likely to miss, and vice-versa (so that even in a short string of attacks you're very likely to hit exactly the most likely number of attacks).

(For AC it does actually change the distribution of results so it does have an effect.)

[Sandman25]

Wrt player melee damage vs player spell damage I think it is fine if player deals 0 damage often with melee attack. It does not drain MP so you can insist on melee or try something else.

[kuniqs]

But it drains HP, doesn't it?

Comparing 2 averaged damage rolls vs 2 AC rolls vs 1 or 2 accuracy rolls makes combat a little more predictable without making it deterministic, especially in the early game IMHO.

[Sandman25]

Comparing 2 averaged damage rolls vs 2 AC rolls vs 1 or 2 accuracy rolls makes combat a little more predictable without making it deterministic, especially in the early game IMHO.

This would be fine too IMHO.

[duvessa]

No thanks, I don't think I would enjoy meleeing hell sentinels or spriggans in that system.

[Sandman25]

No thanks, I don't think I would enjoy meleeing hell sentinels or spriggans in that system.

Why?

[duvessa]

because i would never do damage or hit them

[Lasty]

It seems like the heart of the requests for change is the idea that you should be able to determine in advance with some precision which monsters are scary to melee and which are not. I fundamentally disagree with this idea. I think a game structured as crawl is works significantly better if every non-trivial fight has a chance to go significantly better or worse than the previous one. If fighting a monster (say an ettin) and winning more or less guaranteed that you could fight and win against all future ettins, you would only effectively have one meaningful fight per monster per game. Also, packs of varied monsters would be much less interesting, since you could know in advance that some of those monsters pose no threat at all.

It being possible to be wrong about how well a fight will go is a cornerstone of crawl's interest, imo.

[johlstei]

viewtopic.php?t=13338

It was specifically about monster hp instead of damage, but same overall flavor. Gonna shamelessly quote myself:

The amount of randomness that is appropriate is a strange, esoteric thing to debate and I don't think there exist any compelling arguments that take the form of "this layer of randomness is too much randomness so we should remove it" without any real reasons given beyond that. Reductio ad absurdum doesn't work here if you don't show why a conclusion is absurd.

What level of randomness feels the best psychologically and does the most good for the balance and longevity of the game? I don't know, but that's the sort of question that needs to be asked and answered if you think it should be changed, and it needs to be answered in an absolute sense because "it's too much right now" doesn't mean much on it's own.

Basically, I think a lot of people who complain about too much randomness are actually complaining about the high end of the damage range - they wouldn't be happy if every damage range was fixed and nonrandom at its current max value. No one complains about the wildly varying damage range of crocodiles, they complain about the scary monsters that can kill you really fast if they roll well. I think that should be a separate argument than arguing about randomness itself. I'm happy with scary monsters that can kill you super-fast, randomness is pretty hard to argue about either way.

[Sandman25]

Basically, I think a lot of people who complain about too much randomness are actually complaining about the high end of the damage range - they wouldn't be happy if every damage range was fixed and nonrandom at its current max value.

Not me, I am complaining about hidden huge damage potential. You kill one monster without taking any damage and then you die in 2 hits to second such monster. Basically it means I would prefer to see 20-60 instead of 0-50 (after applying player's AC) provided the game would show me in some way that I can get 20-60.

No one complains about the wildly varying damage range of crocodiles, they complain about the scary monsters that can kill you really fast if they roll well.

Because flat distribution 0-20 is not a "wildly varying damage range" especially when character has 70+ HP.
Compare with Azure Jelly.

[johlstei]

Would you be okay with halving each monster damage roll, and making the actual damage that plus the average?, That is, every range is goes from 0-y to y/2 - y? I'd be okay with such a change but it would make the game a good bit harder, perhaps in a good way.

[Sandman25]

Would you be okay with halving each monster damage roll, and making the actual damage that plus the average?, That is, every range is goes from 0-y to y/2 - y? I'd be okay with such a change but it would make the game a good bit harder, perhaps in a good way.

Sounds good to me. It would be a nerf for melee characters (especially ones with heavy armour) and a buff for axes.

[mps]

I should probably apologize for the remark about "deterministic damage." While I think it's true, although only by a narrow margin, the actual proposal in the OP -- averaging three damage rolls -- is a smart one and unfortunately, the prospect of deterministic damage is what people have latched onto as a way to promote the status quo. The OP's proposal would be a massive improvement.

Averaging damage rolls (based on the current system) has the following advantages:

Basic large scale parameters of damage distributions stay the same. The average obviously doesn't change, but the variance, understood in the sense of standard deviation, stays the same as well. This means things like healing item generation would not need major overhaul.

Straightforward to implement. Patches already exist.

The tail distribution issues are mostly fixed. Measures of variance that are actually at issue, like the 99th percentile damage go down considerably, but the max damage stays the same. Damage spikes that are currently 1 in 1000 (two rolls of 100 out of 100 in a row) become more like one in a million. On the other hand, the odds of zero damage go down to compensate.

re: crate and the idea that "hit or miss" is binary -- no, it's not binary. Armor in roleplaying games and roguelikes in particular (crawl being an unusually explicit example) does not make monsters "whiff," it makes them score glancing, nondamaging blows. A big part of the issue in crawl is not just the potential for high damage in a turn, but the fact that that potential is necessitated by the high miss percentages. If you do three full combat rolls and get a hit and two misses and average them, that's a glancing blow and accordingly causes less damage than a roll of three hits. Of course, it would've been better to design the combat system to produce more consistent results in the first place, but the ship has sailed there and averaging is a quick, satisfactory fix.

Anyway, guys, please talk about the OP's actual proposal.

[johlstei]

What about monsters that are only deadly when they roll highly? Won't your proposal turn them into popcorn if you chop off their damage maxima? Again, I don't think you are complaining about randomness here, but using it as a proxy to complain about high damage. If the damage range where a monster is deadly is in the top 20% of it's range, then removing the low end tails does little while removing the high end may well neuter the monster completely.

If you are opposed to randomness itself, would you support removing just the low end? It would decrease randomness as you say you want, but I sense it wouldn't satisfy you because you actually don't like the high end as it is, regardless of how random it is.

[crate]

If you have 0 ac I'm pretty sure crocodile damage is actually higher-variance* than azure jelly.

I'm not sure how much AC you'd have to have to make it cross over to azure jelly having higher variance (I could figure it out but I don't feel like doing so). But I think that, as mentioned above, you're really complaining about azure jellies because they can actually kill you, whereas crocodiles, despite having lots of damage variance, are weak for where you find them.

Stone giants have more damage variance than ettins, for instance**--ettins have multiple dice added together--but you don't see players complain about their melee damage distribution as often because it doesn't kill as many players.

*To make any measure meaningful you would want to use a dimensionless number for comparison. Damage has units of, well, "damage" or "hit points" or whatever you want to call it; taking the damage distribution and evaluating std.dev/average gives you an appropriately dimensionless quantity to compare to other distributions (basically the "relative standard deviation", though I expect that term is less familiar). Certainly in absolute terms something like a stone giant has more variance than a rat, but that's not a very useful thing to think about for reasons that I hope are pretty obvious.

**I'm making certain assumptions that I admit I am too lazy to test about how much AC you have, here. I think that AC reduces the variance of any regular (i.e. no attack flavours or elec brand or such) melee attacks (again really looking at stddev/average) but it does so differently for different attacks. Technically it's also possible that weapon variation makes this not true at 0 AC (again, I didn't actually figure this out) but I believe that's not the case.

---

I think a lot of the problem here is that no one has actually suggested legitimate alternative damage distributions***. On the side of people satisfied with the status quo this is obviously fine, but if you think the current situation is problematic you should absolutely provide an alternative, since it should be pretty easy to come up with a distribution that fits what you want, at least in qualitative terms. This, of course, requires that you actually know what you want, instead of just saying "not what currently exists," which is not helpful.

***since this topic has switched over basically entirely to monster damage (which, as I mentioned earlier, is completely different from player damage!) I'm basically asking for alternative monster damage distributions. The current ones are simple, since it's generally just adding some dice together, and a sizable majority of crawl's monster damage is just 1dX where X just depends on what the monster is, unless I've been mislead about how monster damage works for a long time. I'm pretty sure player damage is not this simple (see weapon_damage[2]), and as far as I'm aware there's not any easy way to figure out what player damage actually is without attempting to make sense of the code.

---

edit: I started writing this like 20 minutes ago

[Sandman25]

Can some mod please split the thread into two? Sorry about derailing the thread (again )

[crate]

re: crate and the idea that "hit or miss" is binary -- no, it's not binary.

Um, yes it is, in crawl. We're not talking about other games. When you attack a monster, you either hit and then deal some amount of damage (this may end up being 0, but that's not related to accuracy), or you miss and deal 0 damage. Just barely hitting does not make you deal less damage. Just barely missing does not deal some consolation damage. There are literally two and only two possible results from checking accuracy. Hence it is binary.

There are ways to make this not true, but that is not how it works in crawl. You would have to make some other significant changes first.

[crate]

Also to follow up on johlstei's post above:
Because of the way crawl works, every single monster is only dangerous when they roll highly, unless you're really bad at threat evaluation (okay, this is not perfectly true, but it is largely true; the number of times crawl drops you in some truly unexpected situation should be pretty low unless you worship Xom or have already created some emergency for yourself in some other fashion. Teleport traps and shafts are the exception, not the norm). This works in crawl because you fight thousands of monsters per game, and you must win every single fight that reaches a conclusion, because otherwise your character dies and your save is gone. Changing the edges of monster damage distributions is actually a pretty significant gameplay change (and in general makes the game much easier, if you make the edges of the distribution less common without compensating in some other manner).

[kuniqs]

Let's change LCS - instead of rolling 10d10 for damage (IIRC), it will roll 1d100. I'm sure that would give spellcasting an adrenaline kick.

And then, the only sources of bell-curved damage would be evocables & maybe abilities.

Or, like one feature request said: Fixed chance of instadeath each turn. Allows precise ballance readjustments.

Edit: Personally, don't care if it's easy or hard. What I care is that "Come on you *beep*, why don't you die already?" is not funny for me. I don't want ChessCrawl (but hey, that would make an interesting variant). I like spellcasters more than melee because less varied damage means I can factor the average damage I've seen in my combat tactics, whereas totally random nature of weapon damage (will I kill him in 1 hit, or 5?) makes corridor fighting the only tactic that's worth using.

kuniqs is BORED with melee

[mps]

re: crate

Average of three samples from current distributions is in the OP and it is a 100% complete and achievable suggestion on how to change damage distributions.

No, crate, you would not have to make significant changes. If you average three combat rolls, you get complete misses, complete hits, and two kinds of intermediate glancing blows. Averaging three combat rolls is something so simple there's already a patch that does it.

There are many statistics/parameters/whatever that measure variance. Standard deviation is one, but it's not one that's relevant to this thread. The change proposed by the OP does not change the standard deviation of the damage distributions.

One relevant measure is 99th percentile damage in a given match up. The OP's proposal would make that number go down in most cases, but it also makes measures like 25th percentile go up in most cases.

[crate]

Let's change LCS - instead of rolling 10d10 for damage (IIRC), it will roll 1d100. I'm sure that would give spellcasting an adrenaline kick.

And then, the only sources of bell-curved damage would be evocables & maybe abilities.

Or, like one feature request said: Fixed chance of instadeath each turn. Allows precise ballance readjustments.

This argument doesn't work--what if LCS always did exactly 50 damage? Just as ridiculous. You can only argue like this if going in one of the two directions is ridiculous but the other is not. Again, I'm not saying that other distributions for damage are necessarily worse than what we have, but I am saying that you need to argue that they are better and additionally worth the trouble to implement, and I haven't seen that.

Average of three samples from current distributions is in the OP and it is a 100% complete and achievable suggestion on how to change damage distributions.

No, crate, you would not have to make significant changes. If you average three combat rolls, you get complete misses, complete hits, and two kinds of intermediate glancing blows.

Um, this is not what I read. This is turning every attack into three separate attacks (slightly changing how AC affects the result). That's doable, but not at all how I interpreted it (and is, in fact, precisely a significant change that I mentioned). You can actually already do this in-game--macro tab to tabtabtab, and then only use that to attack. You get your three attacks at a time! Of course, that has some significant changes to your gameplay experience baked in.

In case it's not clear, I assumed "average three accuracy rolls" meant that you roll three times, average the result, and then if that number is greater than some other constant number you get a hit. That does produce the effect I described above, where you just make things move farther away from 50% accuracy.

The change proposed by the OP does not change the standard deviation of the damage distributions.

What I propose: instead of rolling for meele damage and AC one time, do it 3 times and average the results.What I propose: instead of rolling for meele damage and AC one time, do it 3 times and average the results.

edit: I may be mistaken here, I'd have to do some math so I'll say this might be true since I'm too lazy to check atm, but it's not obvious to me that this is the case

One relevant measure is 99th percentile damage in a given match up.

Yes, this is relevant. I'm pretty sure that you can get at this (not this exactly, but a good enough proxy) with looking at stddev/average with the distributions that are actually used for normal melee attacks in crawl (it is absolutely true this won't work for things like af_cold, which I'm purposely ignoring because they're more complicated (and also because it's a separate mechanic)), but I'm too lazy to go more in-depth about it right now. You're free to go ahead and show me my assumption here is wrong, though.

[kuniqs]

I'll clarify myself, since I'm a terrible communicator and people seem to be arguing about wrong thing.

What I want: Roll 2 times for damage (the 'raw' weapon damage, without brands) and halve it. Roll AC once.

Spells that check for AC work just like that, except use (much) more dies.

I want melee to be a little faster - it annoys me to no end to tab3x and still barely wound the enemy. The variance in damage makes predicting skirmishes a pain and pretty much forces you to fight in corridors to stand a chance. It woudn't be that bad if monsters didn't had multiple attacks or abilities/spells that roll many damage die, and have much higher chance to put a dent in your HP for the effort.