CGD Podcast Ep. 31 – permadeath, structure, the death of game design writing, and more

Hello everyone. Today I’m talking about a new article I read about permadeath/grinding, as well as what I perceive as the death, or at least curving off of, the world of game design writing.

I also read and responded to a Frank Lantz quote (now on the Dinofarm Forums!) on the topic of structure in games and win rates.

You should also check out the game design subreddit if you haven’t already:

(By the way… beware the term “beautiful”.)

As always, you can support the show by visiting my Patreon page.

  • Guy

    There’s no reason why permadeath needs to be connected to any of the things you mentioned. Grinding, illusions and/or dumb challenges are their own thing. Permadeath is just being “abused” in that sense right now. Operant conditioning won’t be going away. Death is just a theme wrapped around running out of a resource (HP). It’s very easy to immediately understand and has it’s advantages.

    Dynamic difficulty adjustments are an illusion and is bad design. If the player knew he would have a significant chance to lose his next game due to a streak of wins, how do you think he’d feel about that?

    A 50% win rate as a goal sounds horrible to me.

  • Peter Siecienski

    >If the player knew he would have a significant chance to lose his next game due to a streak of wins, how do you think he’d feel about that?

    If they’re on a winstreak it either means they’re getting lucky or the game is too easy for them (and thus they have a higher than 50% win chance, which the game is trying to correct with the dynamic difficulty).

    Would you say matchmaking in multiplayer games is bad design as well? Is that an illusion too? Because that’s all Keith’s suggesting, that single player games “match make” you, so that you’re always playing at an optimal difficulty level. Winning shouldn’t be the only concern you have when playing, you should also care that you are being properly challenged, and so you should be excited that the game gives you a lower chance to win after a winstreak.

    It means you’ll have to put your mind/skills to use more.

  • It takes awhile to explain why the 50% win rate is correct, but I’m curious as to why you would say that “50% sounds horrible”?

    >>If the player knew he would have a significant chance to lose his next
    game due to a streak of wins, how do you think he’d feel about that?

    I don’t have to imagine how they would feel about that. I know exactly how it feels because I’ve played *literally any multiplayer game that has matchmaking*.

  • BudgieInWA

    I agree that permadeath doesn’t imply these other bad design elements. A decision that is “guess between A, B and C” is bad whether or not you need to invest some amount of time to be able to make a guess. Permadeath, as compared to quicksave/quickload, can remove trial and error as a “metagame tactic”. With the ability to undo “A, undo, B, undo, C” is a winning strategy no matter how interesting the decision would actually be.

    Games that use the term “permadeath” may use it poorly as described, but they may also use it to support more interesting decision. For example a game might put you in a room with that level ten monster for the first time but it shares a colour scheme with those earlier monsters that were weak to fire, and it’s got big muscles. In a good game this means back up and try using fire (it’s probably weak to fire and probably has short range attacks). with a good dose of probabilistic reasoning in there — it’s not a perfect strategy game decision).

    On the other hand, games that don’t use the term “permadeath” but still adhere to the same principles obviously rely on them. Either to allow hidden information to fulfil it’s role of presenting varied situations or to make a game feel like a game and not like a puzzle. How silly would chess be if players could undo their moves? How much less interesting would the decisions be in Auro if you could undo? I think “very”.

  • Jereshroom

    Keith is not advocating for the sneaky dynamic difficulty found in some modern AAA games, but rather systems that openly let you know that you’re doing well. (Imagine a military shooter where if you do well enough on one level, your military rank is increased and you’re given fewer NPC teammates for future levels.)

  • Guy

    You’re sort of assuming with the goal of a 50% win rate that the player wants to constantly challenge themselves. The player isn’t choosing when to raise or lower the challenge level and although player’s can and do ruin their own fun with some choices, forcing it on the player isn’t ideal.

    Multiplayer matchmaking is more along the lines of the best case based upon the situation. It’s zero sum, so if a significant portion of players were to have a win rate higher than 50%, that’d mean you’d have a significant sum below 50% as well (until they leave). Take a look at League of Legends. You’ll notice that they still offer games vs bots… games they let the player choose the difficulty for and which has a win rate over 90%. A handful of the best player’s in the world achieve a 60-70% win rate; the high win rate isn’t causing them to leave.

  • Guy

    A 50% win rate assumes a lot of things about the player. You’re challenging the player, yes, but you’re also feeding them challenges you (more or less) know they can’t overcome to maintain your 50% goal. Depending on how you achieve your goal, it can be done shady as well.

    Are you displaying the strength of the challenge to the player? Probably not. If they knew they were going to lose for sure, they probably wouldn’t play or at least not engage as normal. You’re hiding it behind the scenes. Multiplayer matchmaking, as I said below, is merely the best that can be done given the constraints. Ever hear of a smurf account?

  • How does “a challenge I know the player can’t overcome” result in a 50% win rate? Sounds like that would result in a 0% win rate.

  • Guy

    Although it depends on mechanics and execution, let’s take the extreme. If I can bench 200 lbs. and you challenge me with to lift 190, I’ll succeed. If you challenge me to lift 210 (barring gains), I’ll fail. When you break down the 50% goal, you have a win (100%) followed by a loss (0%) in some manner of order. If you’re shooting for 50%, by definition, you’ll be feeding the player losses half the time. Just because it’s an algorithm that calculates and increments some manner of difficulty that the player “may” be able to defeat doesn’t get rid of the fact that it’ll eventually even out.

    It works out to 50% because the reverse is also true. Lose enough and it’ll start feeding you games you know they can win.

  • >>you have a win (100%) followed by a loss (0%) in some manner of order.

    Assuming the player learns nothing and doesn’t improve, that’s true.

    >>Lose enough and it’ll start feeding you games you know they can win.

    Well, no, if you start losing a lot, it will adjust the difficulty to a *slightly lower* one at which you should be getting 50%. It’s not like the difficulty just tanks suddenly.

    I would listen to this podcast episode, where I talked about this more:

  • Peter Siecienski

    >forcing it on the player isn’t ideal.

    Well I guess this goes back to Keith’s view on the “game’s are broken toys” thing. We are very used to engaging with games like they are toys, making design decisions in order to figure out what would be the most fun experience for us (as opposed to making strategic decisions, “what the optimal move is”). Keith advocates that the game should do that for you, because it’s the game designer’s job is to figure out what the most fun experience would be, not ours.

    If you don’t agree with that fundamental concept, then you probably won’t agree with much of the stuff on this sight.

  • AName

    I’ve recently started working in the AAA industry as a software developer. I’ve been reading game design blogs for a decade, but this has been my first time meeting and conversing with game designers in real life and now I do it on a daily basis. It’s almost unavoidable, because I’m sitting right next to 10 game designers in an open office and all they ever talk about is LoL, Blizzard games, first person shooters, MMO’s and how games are made.

    By conversing with these people and by reading blogs, I’m starting to detect a pattern of doing too much arm-chair philosophy and knowing nothing about math, complex systems, graph theory, game theory, statistics, layout design or anything that could be beneficial to any kind of design. It’s like all designers ever want is someone to acknowledge their vague assumptions which they pass on as constructive ideas. They know that they’ll never even come close to executing these “ideas”, simply because they can’t even START making a game. They always have their OWN ideas and delusions, they almost always talk ambiguously and make invalid points and focus on whatever irrelevant detail they think they’ve figured out. And worst of all, they almost always philosophize about how to make extremely complicated and polished games, when they can’t even create or repurpose the simplest of mechanics.

    Even David Sirlin is guilty of doing too much babbling. He keeps talking about how he tries to make lean designs and he always ends up making game components which are as noisy as Japanese billboards. His games are really solid, but they’re the definition of NOT lean in terms of mechanics, upkeep, rules, layout design, etc… There’s a disconnect between his philosophies and his design.
    That being said, what I do like about his blog is that he hits the nail on the head – he talks about specific things in games which we all find annoying. He talks about how game designers have approached these problems and why their approach ends up failing. He also talks about better approaches to these problems. It’s usually stuff that we can put to a test straight away. This, to me, seems more constructive than doing philosophy about what the optimal win rate for a game should be. The people reading blogs, for the most part, can’t come up with anything they would ever play and you’re talking about how they can optimize their designs… We’ll definitely have to think about optimizing the win rate… IN THE FUTURE, when we actually know what we’re dealing with here. It’s even better if you can start off with an optimized system and take it from there (like *REAL* designers in other fields actually do), but it’s kind of funny when everyone can’t even smell the ocean, yet they’re already making firm statements about the bottom of the ocean. Come on, people…

  • So you’re saying, “I would prefer it if you talked more about design lessons that can be applied to today’s games”, as in like, how to improve on Zelda or Grand Theft Auto or something?

    I agree that that would reach more people. But that’s just not where I am. The way to “improve” those things, in my view, is to change them so dramatically that they would be unrecognizable. And I’ve already been developing my theory of game design for years. So yeah, I can see how if you’re stepping into it and hearing “50% win rates” it would sound kind of alien.

    I don’t know how to solve that problem though, other than telling you to watch my video series (3 Minute Game Design) or catch up on my older posts or even read my books. Let me know if I’ve interpreted you correctly.

  • Chris Bateman has lamented the lack of blog-style writing in general, including and especially with regards to game design. As you say, there’s been a shift towards podcasts and videos to some extent, but Chris has also noted that a lot has to do with social media and the shift towards that, which has caused short ephemeral thoughts to be the most propagated things on the internet. A shame, but I am not sure the trend will always be in this direction.

    Chris does still sometimes blog about game design at: though.

  • Van

    That’s not what I’m saying and your idea about the 50% win rates isn’t alien to me (and I don’t disagree with your reasoning about it).
    I’ve been reading your blog for a while, I’ve watched your mini series and I’ve watched some of your earlier YouTube videos about Clockwork Game Design a few years ago. What I find most interesting about your blog, your approach to game design and your 3 Minute Game Design series is that you break things down into elements of game design, much like we have elements of music composition. When we do that, we realize that a lot of what has been done in contemporary games doesn’t support their form. For example, we create games which thrive on interesting decisions and then we make those decisions inaccessible (or even irrelevant) through execution barriers, paywalls, random noise and complexity brick walls. By not paying attention to form, game designers just end up repeatedly shooting themselves in the foot.

    Like you, I’m not interesting in improving today’s games, because they seem fundamentally flawed and thus have a limited potential. I would much rather see a fresh start.

    The focus in my previous post isn’t about how Sirlin talks about improvements to popular games, rather it’s that “It’s usually stuff that we can *test (empirically) straight away*”. I fully realize that not all good ideas can or need to be tested.
    Currently even successful designers struggle with making even the simplest mechanics work on the most basic level and you’re already starting a discussion about how to figure out a universal win rate. It’s a useful discussion, but it’s one that, I believe, needs to happen some time down the line after we’ve figured out how to consistently come up with interesting and solid mechanics. Discussing input and output randomness, information horizon, handicapping, etc. seem much more appropriate for this “early game” in game design.

  • Van

    Or maybe I’m wrong and we need to establish the best win rate before we can begin to make good game design.
    Either way, figuring out the best win rate doesn’t tell us anything about how to make interesting and solid mechanics. And without interesting and solid mechanics, we won’t even have a win rate to worry about, because we won’t have a game. The more I think about it, this discussion about the optimal win rate seems like it should be done post hoc.

  • I just don’t agree that figuring out the best win rate is at all difficult. If you accept the other parts of my philosophy and model for game design, the 50% win rate is pretty obviously correct. It’s very simple: we are measuring whether you have improved or not by comparing you to “yourself”. Are you better than “yourself”(your past average recorded performance)? If so, you’ve improved. If not, maybe you’re making new mistakes and getting worse. If you were to play a clone of yourself, you’d have a 50% win rate.

    It’s not that we “have to figure out best win rate before we can make good game design”, it’s that if you accept the other bits of my philosophy, this bit about win rate comes with it, at least it seems to me. If you object to the reasoning on win rate, that’s fine. But I don’t think it’s any more inherently problematic of a project than any of the other theory work I do.

  • Van

    I think that making a hypothesis about it does seem easy and it does seem like yours is a sound one. At best this is more of an optimization, a useful constraint if you will, one which we will use AFTER we’ve already learned how to make complete games. It’s like we’re car engineers who don’t know which car part goes where, yet we’re already concerned with fine tuning every nut and bolt.

    Also the optimal win rate problem doesn’t seem at all like an apriori type of problem. I don’t know if you’re familiar with the concept of convergence, but what if this low variance system leads to local optimums in our learning? Or what if, due to the human psyche, we’re less effective learners at the optimum win rate. Maybe having the right amount of variance is more important than a specific average? Maybe an optimum win rate isn’t the most engaging or compelling for us and maybe (actually definitely, but whatever) we’re much more efficient learners when we’re engaged? Are those not valid concerns? How would I resolve these concerns without doing expensive and difficult empirical tests? And if we do resolve these concerns and we figure out a perfect win rate, what have we learned about making solid mechanics?

  • Van

    P.S. Those weren’t rhetorical questions.

  • I actually don’t think “50%” is a result of “fine tuning”. I can see how if I had some weird number like 57.4% you might be making such an argument. But 50% is just because of a very simple principle that the best way to get feedback on a win/loss is to have you play yourself.

    >what if this low variance system leads to local optimums in our learning?

    I don’t see any reason to think that a 50% win rate would be responsible for that. Or that any win rate would be responsible for that.

    >we’re less effective learners at the optimum win rate.

    I think that is exactly the opposite of what would logically make any sense, at least given my entire understanding of how interactive systems and feedback works. So to me it’s like “what if down is up and hot is cold, what then?!”

    >Maybe having the right amount of variance is more important than a specific average?

    Are you just proposing arbitrary possibilities? By default, yes “maybe” anything is the case. If you actually are proposing this specific idea, make the case for it please.

    >Maybe an optimum win rate isn’t the most engaging or compelling for us
    and maybe (actually definitely, but whatever) we’re much more efficient
    learners when we’re engaged

    This question contains the assumption that a non-50% win rate is “more engaging” than a 50% win rate. I think you need to demonstrate this.

    I am concerned that the “flurry of questions” model of your comment seems perhaps not terribly interested in answers. Like that last question is pretty easily answerable: if we know that the optimal win rate is 50%, then we can design systems to provide a 50% win rate for all players (which I have done in Auro, and every online competitive game also does).

  • Van

    A “flurry of questions” wasn’t my intention. I would much rather retract some of my statements for now for the sake of focusing on one thing at a time (that and you’ve answered some of my concerns). If having a 50% win rate isn’t fine tuning, then what if you made the mistake of setting Auro’s win rate at 30%? Would it stop functioning as a game or would it simply be less efficient? Or something else?

  • It would be significantly less efficient. 30% win rate means that the game is much harder / “better than the player”. So the big problem here is that losing isn’t necessarily telling you anything about whether the player is improving or not, because the player *should* be losing more often than not anyway. The player has to not only improve, but improve by a significant margin now to even get the “you improved” feedback. If the player improves a small amount, it won’t necessarily even get picked up by the system, because the game is so much harder.

    Just to clarify further, the higher the win rate goes, (i.e. above 50%), the more you run into the opposite problem. With an 80% win rate, the player has to really fuck up badly for the system to even pick up on it.

  • Van

    >It would be significantly less efficient
    This is what I meant by “fine tuning every nut and bolt” – it’s an optimization. I do agree though – winning more than 50% when you’re not playing well, certainly doesn’t do any good and while less than 50% can lead to compulsion, it can also be demotivating. Here’s the thing I don’t quite understand though: if I’m actually always playing wrong, what would happen to my win rate?

  • Yes, I’m trying to advance optimal guidelines for game design.

    “Playing wrong” is relative, of course. If you start with a 50% win rate at some skill level, but then you start screwing up consistently, your win rate will go down. That’s when the system needs to re-adjust itself (become easier) to try to re-establish a new 50% win rate at your new lower skill level. If you start playing better, your win rate rises, and the system should re-adjust again.

    (Just to make another thing clear: you don’t adjust difficulty after every match. You need to get a handful of games to start making a meaningful prediction about whether this player is actually getting better or worse, or if it’s just a fluke. Check the way it works in Auro for, I think, a pretty good execution of such a system.)

  • Van

    I’ve had some limited experience implementing tracking and dynamic difficulty. I’ve worked on a 2d racing game where I had to implement a few dynamic handicap mechanics (more nitro and better weapons for the trailing player, etc); also dynamic break assist and dynamic handling boosts, depending on the player’s cornering performance. Among other things, I had to track the variance in player performance, in order to avoid rubber banding and the 2d racing equivalent of Elo hell.

    Having had some experience with that, I would say it’s not for beginners and definitely not for laymen.

    >Yes, I’m trying to advance optimal guidelines for game design.
    I don’t presume to be an authority on effective discussions, I just think that we’re at a point where we first need to have something TO optimize.

    Concepts such as dynamic difficulty are complex and are difficult to discuss with beginners. So much so, that such discussions often become counterproductive, because it just makes people stubborn (from my observation). I just think that discussing simpler concepts, which can be implemented relatively quickly even by relatively inexperienced designers, seems to me to be more productive.

  • Van

    Put simply – I think people simply lack entry-level ideas, with which they can do anything at all. Dynamic difficulty is definitely not entry-level. Maybe you’re trying to start a discussion with more experienced game designers, in which case I’m just adding noise, so I’ll leave it at that.

  • In my view, the “during gameplay” dynamic difficulty adjustment is really, really different from the “between matches” adjustment. I advocate for the latter in every case and the former in NO cases.

  • Van

    Yes, it quickly becomes apparent that making real time adjustments just makes the game insanely difficult.

  • Conor

    Everyone loves arguing about the 50% winrate so I thought I’d write a post for my blog on my thoughts around it.

  • Thanks!

  • Isaac Shalev

    You have made a claim about 50% win rates that I don’t think holds up. Your claim, if I understand it, is that a 50% win rate enables players to balance negative feedback with positive feedback such that they learn most efficiently how to play the game.

    My objections fall into two broad categories. The first is about the efficacy of the 50% win rate as compared to other win rates. The second is about whether maximizing efficiency of learning to play is something to strive for or a meaningful marker of good game design at all.

    The first issue is really one of education. Losing/failing is much more educational than succeeding, as nearly any teacher or coach will tell you. Losing tells you to try something new and different. Winning tells you to do the same thing. But if you lose too often, you will get so frustrated that you might just quit. It stands to reason that the “ideal” win rate, in the sense that it is most efficient, is the lowest win rate at which players do not get frustrated and quit. The actual value of that win rate will vary based on player ages, personal factors and cultural factors, but it is almost certainly lower than 50%.

    My second objection is that the 50% win rate is any significant marker of game quality (or of matchmaking system quality). As others have pointed out, the win rate is a statistic compiled out of distinct experienced. If I regularly alternate between stomping on, and getting stomped by, my opponents, I likely won’t learn much, nor will I have a lot of fun. You may be doing some shorthand here and saying “the right opponent is the one against whom you would win 50% of the time.” Let’s assume that’s the case.

    Even in this case, it doesn’t stand to reason that the experience that I’m looking for at all times is the perfectly matched experience. This goes back to the ‘broken toys’ issue. If I want to learn to play better, I may want to play someone significantly better than me. If I want to prove myself, I might prefer to play someone my equal. If I want to experiment with new ideas and strategies, a weaker opponent would be a better lab. The designer should consider the quality of those experiences in designing a game too, not just the optimal head-to-head.

  • >if I understand it, is that a 50% win rate enables players to balance
    negative feedback with positive feedback such that they learn most
    efficiently how to play the game.

    Not… exactly. Actually one issue with the way I have maybe been communicating this, that your comment made me realize, is that people might think that I’m saying players should win 50% of the time. What I’m really saying is that going into a match, a system should try to balance itself to give the player a 50% chance to win as best as possible. Since players are constantly learning new things and improving, it’s quite likely that their actual win rate will be slightly above 50. But we constantly do our best to draw it *towards* 50.

    So it’s not about “giving them 50% wins and 50% losses”. It’s “giving them a balanced challenge” – a challenge based on their “last recorded skill level”. That way we can see if they’ve gotten better or worse.

    I agree that 50% on its own is not a useful metric for figuring out if a game is good. It’s similar to “is the game balanced” – games do need to be balanced, but the fact that a game is balanced, alone, tells you almost nothing about how good a game is.