For and against binary goals in strategy games (and against high score)

keithburgun

Administrator
Staff member
#1
Hi! I'm for binary goals (win/loss) in strategy games, very strongly. Actually I think that non-binary goals (like a high-score system) are probably better suited to another form, like contests or toys. One of the issues we run into is that we just have this single term "goal" that we apply to stuff like "solving a puzzle", "getting a high score" and "winning". But these are all pretty fundamentally different things.

Some good things about binary goals are that they:

1). lets the player create strategy around those goals. They have an idea of how long the game will be and where they need to be to get there, if even loosely.
2). lets the designer build a system around this goal. If we know "winning is THIS STATE" then we can balance the game around that, whereas if the "end state" could be anything from like zero points to 1000 points you have to keep things very loose.
3). gives the player strong, concrete and fundamental feedback for their play. "You won/lost" is a fundamental starting point of the player's assessment of their play in the match. When you know this, you can trying to explain the "why". When you don't have this, you have nothing to explain. It's just some number you got

On a practical level, if you think about how people play high score games, they sort of play them more like toys. They negotiate with themselves during and after the game about how to assess their own play.

What do you think?
 
#2
Hi I'm not convinced that binary goals are strictly better than highscore like systems. However, I readily admit that traditional highscore systems have a lot of problems. Essentially I think that there are yet to be discovered workarounds to all of the flaws of highscore systems, but that their discovery requires active theoretical and practical research.

I think there are several premises that we can (hopefully) agree on from the outset:
When talking only about single player strategy focused games.
1. Goals are good for players, they orient the player towards doing something with the system.
2. Goals are good for designers, they give some structure around which to balance the game.
3. Part of the value of a goal is that it provides the player with information about how good they are at the game once a match concludes.

So hopefully we agree that goals are good.

We can now discuss about what constitutes a 'Good goal'.
I think traditional high score systems, minimally orient the player, do not help designers balance, and provides almost no information about their success.

I think we can conclude that traditional high score systems barely function as goals.

By comparison Keith has established that a binary goals, orient players, helps designers balance games, and provide the player with clear information about their success in a match.

There are some caveats to these benefits.

But first let me address the issues with binary goals as they apply to strictly binary goal games. For example, rogue-likes where the player either defeats the final boss or dies:

2. Making balancing easier with a binary goal isn't strictly a benefit. Balance is still an iterative process where goals are tweaked in conjunction with tweaking the rest of the system. No matter what type of goal you choose you will need to adjust it. However, some types of binary goals are not able to be easily adjusted, so this reduces the number of tools you have to tweak difficulty.
Also the tighter the difficulty of a game, the less variation in player skill it can support. Having a looser goal (or set of goals) gives you more room to support more different types and experience levels of players.
3. Binary win loss gives clear information about the player's success at a match, but is a very poor indication about how skilled they are in general. In a singular binary goal game players often take the results from several matches, or a run of matches, in order to determine their skill.

As such, I'd say that a singular binary goals are okayish.

Many of these issues are address by combining a Elo-like ranking system with a binary goal matches. This type of system, developed by Keith Burgun, and first presented in Auro, replaces a single binary goal with a larger set of binary goals, each of which constitute a different difficulty.
2. This system can support many skill levels of players, since there are easier and harder goals.
3. This system gives better feedback, since the rank at which you can consistently win, gives more detailed feedback than a single binary goal.

The Elo system here essentially acts to match players with a goal thought to be around their skill level.

I think this constitutes a huge improvement over single binary goal systems for strategy games. If we are to think about goal systems along a continuum with traditional highscore systems on one end singular binary goals on the other, Then I think this ranking system's improvements are due at least in part to moving away from singular binary goal games, and towards traditional highscore games. (please read this for a better explanation of this idea ).

However I think there is a lot of unexplored space on this continuum, and I think we should explore this space because of a major flaw with ranked based binary goal systems.
Players face a fundamentally different challenge in different matches based upon where they are ranked. This causes two problems:
A) Many players suffer a large miss match between their skill, and the challenge of the match, especially as they are learning.
B) Ranking systems rely on difficulty modification to vary the player skill requirement for the goals of each different rank. This means matches played at the lower ranks are fundamentally different to those played at higher ranks, and many early lessons are misleading and eventually cease to apply. (see)

I also think that developing a rank system takes time that could have just been spent loosening and balancing around a goal continuum, or series of goal watermarks.

As far as I know A) is a hugely unsolved problem. There are workarounds, but they are not ideal, especially for new players, and returning players.
Also I strongly suspect that B) will become an even larger issue in games that have scoreless goals, unless they still have some way of increasing the difficulty of the goal that is independent of increasing the difficulty of the rest of the match.

As such, I'd say that a ranked binary goals are decent to good.

Because of all this I am in favour of actively researching par based goal systems that have watermarks or medals that indicate that a player's skill has reached a certain quality. In the mean time I strong recommend games designers evaluate the costs and benefits of ranked based and par based goal systems for the specific structure and pillars of the game you're working on. Weigh the trade-offs, and pick what will work best for your game.

As such, I'd say that par based goals are decent to ????

PS. I think that many types of goals help orient the player and inform them about the structure of the game. And I also think there are other non-goal ways to communicate at least some of this information.

TLDR: Traditional highscore systems are pretty bad, singular binary goal systems are also pretty bad, but there is a continuum of better goal designs between these two extremes. Ranked based systems are one, but more are being developed (ie. par). Each have their trade-offs and so it will pay good developers to choose the goal type that works with their games system.
 
Last edited:

keithburgun

Administrator
Staff member
#3
I hear often that a single player Elo system, like a rank system, is "really hard to do". This is based on what exactly? I attempted it for like the first time ever with Auro, and since then I think @BrickRoadDX has done it a few times, and that's pretty much it as far as I am aware. I have very little doubt that if development on Auro hadn't been cut off we'd have gotten it right years ago. I'm not saying it's easy but it's well within reach.

Someone should explain a par system here so that we have that explained, especially by someone who's in favor it it.

And yeah just to clarify, I was talking about singular goal systems matched to your skill level. Like you can't say that chess has a bad goal because I played against a guy who's way better than me. We *assume* that you'd be playing against someone of a comparable skill level, and we should do the same with single-player.
 
#4
I don't like the idea of par systems (assuming I understand what they are!) because they seem to require that the game has a score goal. Which in turn implies that the game will give points for each different action and mid-arc objective, according to what it deems they're 'worth' rather than the player being allowed full scope to create their own strategy based on understanding of the actual *value towards a goal* of each thing. It's almost like the designer is saying they've solved their own game and doodads are 'worth' 3 and whatsits are 'worth' 5. Independent of any utility and just based on the difficulty of achieving them. Skill then seems like it's going to be all about numerically second-guessing the designer like 'hmm, actually I think doodads are undervalued and are really worth 3.2' etc. Runs the risk of getting spreadsheety? IDK maybe that's OK, I might just be recoiling at the numbers, all the evil, evil, numbers.
 

keithburgun

Administrator
Staff member
#5
When you start a match with a "par" system, are you going for "par+1"? Or just trying to match par? Or par + "as much as possible" (which gets back to some of the original complaints with high score)?
 
#6
I don't like the idea of par systems (assuming I understand what they are!) because they seem to require that the game has a score goal.
Yes par has a score based goal, so to a large extent it lives and dies based upon the quality of the score system design. I think there are advantages and disadvantages to score based goals, they give another knob to tune the system/difficulty. I think what you claim is an implication is wrong, and merely an example of bad score design. I think good score design will be about choosing which of the medium arcs to reward with a point. By comparison good binary goal design will be choosing the longest arc and making it worth one point.

I think the real design key here is figuring out what your game is about. Is it about the long arc, or is it really about a medium arc. Traditional highscore games reward the shortest game arcs, and so they are at best tactical and very repetitive.

Essentially what your doing when you choose which arc is worth points is saying to the player emphasize and optimise this task over all others. Trade them all off against the chosen arc.

@keithburgun You need to be more careful about distinguishing match goals and meta goals. In SP Elo the meta goal is to raise your rank, with par the goal is to raise your par. Within a match the goal in SP Elo is to win, and the goal in par is to maximise score at the end of the game, given the generated layout.

Match goals in par Do Not reduce to a binary, don’t try and force it. Good par systems have defined end points/long arcs, against which points can be traded. Watermarks/Medals give the metagame and the match score context. ie I got gold medal this run, or i finally got my par up to silver. The SP Elo equivalent might be: yay i finally beat a rank 11 match, or I got to rank 9.
 
Last edited:

keithburgun

Administrator
Staff member
#7
and the goal in par is to maximise score at the end of the game, given the generated layout.
What is "maximize"? When has one done that exactly? How do you build a strategy around "maximizing"? IMO, you can't, you can just operate tactically and try to not-lose as long as possible.
 
#8
Hi. Since no one has done so so far, I'll give a basic description of what the par system actually is:

Par is a measurement of a player's skill level, which adjusts after each match. After each match the player gets a score, and their par is set to a weighted average of their score and the previous value of par. In other words, par follows this formula after each match:

newPar = score * c + oldPar * (1 - c)

where c is some constant which determines how sensitive the par system is to change; higher c means the par will move more after each match.

Par is intended to give the player an objective measurement of their skill level over multiple matches, just like the "rank" in an elo system.

In a par system you are not just trying to reach par, or par+1. You are just trying to maximize your score; reaching par is no more significant than reaching any other number. e.g. if my par is 20, going from 19 to 20 points should be no more consequential to me than going from 15 to 16. Par only exists to give an indication of your average skill, it is not a binary goal that the player should be shooting for.

One implementation of the par system (and actually the only one, as far as I know) is in my game Brazen Berry Bonanza. Here, the par starts at 10, and c=0.5 (so each match, your par is recalculated as (score + oldPar)/2). However, I've since come to believe that 0.5 is too high, I think a value like 0.25 if probably better.

...according to what it deems they're 'worth' rather than the player being allowed full scope to create their own strategy based on understanding of the actual *value towards a goal* of each thing. It's almost like the designer is saying they've solved their own game and doodads are 'worth' 3 and whatsits are 'worth' 5.
I agree that the way of using score systems that you are describing here is really bad. I'd call the type of thing you are talking about here a "post-hoc score system", which happen when the designer creates a game with some predefined notion of how the game should be played, and creates a score system to try to match that notion. In post-hoc score systems there are often many different things that give different amounts of points, e.g. you get 100 points for killing one type of monster, 150 points for killing another type of monster, 200 points for collecting a collectable, etc. This type of system seems pretty obviously terrible for many reasons.

Thankfully, that is not the only way to use a score system. Whereas a post-hoc system starts with a preconceived notion of strategy and slaps together scoring mechanisms to encourage it, a good score system starts with a basic scoring mechanism and allows the strategies to arise from it. In BBB, for instance, there is only one thing that gives you score, and it only gives you one point: collecting a berry. This is not because I had an idea of how I wanted the player to play and decided that giving the player a point whenever they get a berry was a good way to encourage that way of playing, instead this was just chosen as a basic property of the ruleset, and all strategies arise from it (the same way strategies arise from the goal in a binary system).

What is "maximize"? When has one done that exactly? How do you build a strategy around "maximizing"?
Maximizing your score essentially means that you choose whatever plan you believe has the highest expected value for score at the end of the match. e.g. if you think that plan A has an EV of 20 and plan B has an EV of 30, and you are trying to maximize your score, you will choose plan B.

There is no point at which you are "done" maximizing.

You build a strategy around maximizing score the same way your build a strategy in any game. Win/loss games are also maximization problems: to try to win a match of a win/loss game is to try to maximize your probability of victory. All games, whether they use a score system or a binary system, are about maximization, so if it isn't possible to build a strategy around "maximizing", it isn't possible to build a strategy at all.
 

keithburgun

Administrator
Staff member
#9
@Hopenager
So I think what you're saying is that you know the game length, and let's say par right now is 30 points. I plan a strategy that I think is likely to get me, say, 30-35 points or so. Right? You don't "go for" 50 points I assume, because that will probably be way too high risk. And you certainly don't go for UNDER PAR, right? So I dunno, it kinda just sounds like it's still a binary goal in the sense that there kinda IS some correct "score goal to go for", probably 31 points. It seems less correct to go for something much higher or much lower than that.

My other objection is that I don't think "guy who is going for ~31ish points (even if we widen this out to like 25-35 points or something)" should have the rank adjusted MORE because he ended up getting 50 points on a single match - and this is kinda at the heart of your issue with true binary goals. My reasons are:

1. Your assumption that 50 is better than 30 because it's a higher number is "contest thinking", or "tactics game thinking". In a strategy game, if you're going for something, you should be rewarded for hitting that thing, not for landing WAY off in either direction, up or down. Even the concepts of "up" and "down" shouldn't apply to strategy games so easily.
2. Variance means there are going to be outliers. BTW this becomes more true I think in a bigger, deeper, more complex strategy game.

I think you've thought for awhile that because you can use the term "maximize" with binary goals and your thing, that they're like the same. And that's missing a lot of nuance.

In these score games, you're maximizing-as-collecting. You're +1, +1, +1, +1 maximizing. Adding. This is actually what you're doing while you play.
In binary goal games, you're not adding stuff up while you play, you're not collecting. You're strategizing, and you're trying to make good moves. Yes, you can describe those moves in terms like "maximizing chance to win", but I think that's just a bad characterization, even if it's true. Let's be totally honest: strategy-judgments is never anything numeric or quantitative, it's very qualitative and heuristic. Whereas the "maximizing" of the score game is totally quantitative and additive, while you're playing.
 
#10
Your assumption that 50 is better than 30 because it's a higher number is "contest thinking", or "tactics game thinking". In a strategy game, if you're going for something, you should be rewarded for hitting that thing, not for landing WAY off in either direction, up or down. Even the concepts of "up" and "down" shouldn't apply to strategy games so easily.
This is a very interesting thought and something I hadn't ever really thought of before (even though I've been a binary goal supporter from the beginning).
 
#11
In BBB, for instance, there is only one thing that gives you score, and it only gives you one point: collecting a berry. This is not because I had an idea of how I wanted the player to play and decided that giving the player a point whenever they get a berry was a good way to encourage that way of playing, instead this was just chosen as a basic property of the ruleset, and all strategies arise from it (the same way strategies arise from the goal in a binary system)
This is persuasive. For a single scored thing certainly. As soon as there are multiple things with different points values to cross-reference I'd worry that the game is prescribing a strategy/solution.

Kind of a related subject, it's why I don't like the trope of 'the shop'. The different tools on sale have different prices, which is effectively telling you how good the designer thinks they are and challenging you to do math all the time and prove them wrong! I'd rather just be offered the spanner or the screwdriver and work out myself which one is best in which situations. I do appreciate though that many players like being presented with a system and 'beating it' by mathematically finding all the little loopholes in the item prices and points awards. It's a certain audience for sure.
 
#12
So I think what you're saying is that you know the game length, and let's say par right now is 30 points. I plan a strategy that I think is likely to get me, say, 30-35 points or so. Right? You don't "go for" 50 points I assume, because that will probably be way too high risk.
This is a misunderstanding of the concept of Expected Value. EV is calculated as the sum of all possible outcomes weighted by their probability, and thus it takes risk into account. e.g. imagine 2 plans:

Plan A has a 50% chance of providing 30 points and a 50% chance of providing 31 points. This plan's expected value is (0.5 * 30) + (0.5 * 31) = 30.5

Plan B has a 85 chance of providing 20 points and a 15% chance of providing 50 points. I assume this is the type of thing you are talking about when you refer to a plan where you could "go for 50", but it would be too risky. This plan's expected value is (0.85 * 20) + (0.15 * 50) = 24.5

Plan A has a higher EV, so you should use plan A, even though plan B has a chance of giving you 50 points.

In a strategy game, if you're going for something, you should be rewarded for hitting that thing, not for landing WAY off in either direction, up or down. Even the concepts of "up" and "down" shouldn't apply to strategy games so easily.
Why? Can you justify this? I have heard you make this claim a lot, but never an actual justification that didn't beg the question by assuming that binary goals were superior to start with.

I think you've thought for awhile that because you can use the term "maximize" with binary goals and your thing, that they're like the same. And that's missing a lot of nuance.
It's not that they're the same, it's that the fact that binary goals are maximization processes invalidates the idea that "all maximization processes are bad". Generally when you speak against score systems you sound like you are talking about Galaga or something, where you just "try to get a bunch of points RIGHT NOW" over and over again. I don't think anyone here has advocated for that; there are far better ways to do a score system. The fact that you are characterizing all maximization processes that way is a problem. The reason I bring up the "binary games are maximization processes" point is to demonstrate how silly it is to paint all maximization processes with that broad of a brush.

Yes, you can describe those moves in terms like "maximizing chance to win", but I think that's just a bad characterization, even if it's true. Let's be totally honest: strategy-judgments is never anything numeric or quantitative, it's very qualitative and heuristic. Whereas the "maximizing" of the score game is totally quantitative and additive, while you're playing.
I don't think it's a bad characterization at all. The proper plan in a binary game is the one which gives the highest chance of winning, the same way the proper plan in a score game is the one which gives the highest EV of score. Judgements in a binary system are not necessarily any less quantitative than those in a score system. Whenever you try to implement a good plan in a binary game you are implicitly making a quantitative judgement: "this plan has a higher chance of success than the alternatives".

You are right that most of the time you don't think directly in quantitative terms in a binary system, you use more heuristic methods. But the same can also be true for a score system. I'll again use BBB as an example; when playing BBB the player does not have to do very much direct quantitative comparison, play is guided almost entirely by heuristics. I agree that most score games that currently exist are mostly about direct, non-heuristic quantitative comparison, but I am not advocating for "most score games that currently exist".

Additionally, there are binary systems that rely on direct quantitative comparison, too. Take poker for instance: the game is mostly about calculating probabilities and direct numeric comparisons. There is far more quantitative reasoning in poker than there is in BBB, even though poker is binary and BBB is score-based. Being dominated by quantitative thought is not a property of score systems, it is a property of bad systems (a category which, I will readily admit, most currently existing score games belong).
 
Last edited:

keithburgun

Administrator
Staff member
#13
Why? Can you justify this? I have heard you make this claim a lot, but never an actual justification that didn't beg the question by assuming that binary goals were superior to start with.
I don't know, I feel like it's obvious that a strategy is with respect to a certain, specific goal. I actually feel like you should have to demonstrate that a strategy is directed towards "a certain goal, or more". Take any task. Let's say I hire you to dig 100 yards of ditch in front of my house or something. If you come up with a strategy for doing that and you end up digging 200 yards of ditch, am I supposed to reward you more for that? Or did you fuck up colossally? Probably could think of other examples that are even better than this, but hopefully you get my point that more is not always better.

(Kind of a side note, but I wonder if there's this weirdo capitalist "consuuuuuuume" brainwashing thing that might be related to this - just we're all obsessed with more more more and always sees more as better, in every context.)
 

keithburgun

Administrator
Staff member
#14
'the shop'. The different tools on sale have different prices, which is effectively telling you how good the designer thinks they are and challenging you to do math all the time and prove them wrong!
ETO's shop has randomized items which then get a generated "balanced" price, and then that price is slightly randomized.
 
#15
Hard to say without seeing it but I think for me it might have the same (subjective) problem. I.e. as the player learns the game they will need to work out numbers in their mind for what they think sword and other things are really worth, so that they can decide what things the shop is offering a good or bad deal on this time round. But having the price be part of the random gen context of a match definitely seems better than it being fixed.

IDK I think this is probably a personal taste thing, so might be best not to derail the par/goal discussion. As Bucky remarked on discord it's hardly a big deal to do some simple math to evaluate a trade-off. For me though I'd rather keep it away from any kind of numbers as much as possible. I want all the players! ALL OF THEM! Even the ones who are freaked out by even the tiniest hint of math in daily activities. (Which is a lot of people!)
 
#16
I actually feel like you should have to demonstrate that a strategy is directed towards "a certain goal, or more".
That isn't my position, I have no idea where you are getting that from. I am not saying that "more is always better", I don't even think that's a meaningful claim. What is "better" in a system is defined by the judgement the system provides: In a binary system wins are better than losses, all wins are equivalent to all other wins, and all losses to all other losses. In a score system, a score A is better than a score B if A>B, and equal scores are equivalent.

If a binary system has the goal "have exactly 10 points after 20 turns", and I have 11 points after 20 turns, I have failed. Alternatively, a system could have the goal "have 10 or more points after 20 turns", in which case having 11 points after 20 turns would be fine. Both of these are valid systems.

It's funny that you give that ditch digging example, because it seems clear to me that in that situation a binary judgement is a bad idea. If you asked me to dig you 100 feet of ditch, and I end up digging 101 feet, would you tell me "YOU FAILED" because I didn't get exactly 100 feet, or would you tell me "you were 1 foot off"? The latter is clearly more useful, for exactly the same reason a score is more useful than a win/loss bit.
 
#17
Having a win/loss goal doesn't rule out more detailed information about the end state. 'Game over. You lost by digging the ditch 1ft too long'. 'Game over. You lost by digging the ditch 100ft too long'. 'Game over. You won by sacrificing a knight to isolate the king and mating with two unusually mobile rooks'. 'Total control of the top lane'. 'Collecting a lot of berries early game compared to late game, according to this time distribution graph (game displays graph)'. Strategy games actually could do more to try and break down the match like that. In some games you can watch a replay.
 
#18
I think if you design a game with a good match length arc that really makes sense and works then you should use binary win-loss. But, I think there's a lot of potential games that have really good mid-size arcs, that are strategic in nature (especially when constrained by a fairly simple match length arc) and for those game we should use par-like systems.

I find the arguments about trenches and well really any game state that has numbers larger than 1 thoroughly unconvincing. I'm going to need to see something pretty rock solid to be convinced that when given a fixed amount of resources, producing a better than strictly necessary outcome is bad. Efficiency is something that is fundamentally important to all aspects of our lives in the real world, and so I expect my games to reward players for their efficiency, not punish them. A game that says be very efficient, but not too efficient, be good, but not too good, to be quick, but not too quick, just doesn't make sense to me.
 
Last edited:
#19
Imo, one notable quality of par systems is that can cause a state where the player would rather play more conservatively in order to lose small, if there's a ~5% chance of winning vs a ~95% chance of losing big. And as Keith suggests, it could create temptation for a player to select strategy/tactics that let them win big, rather than safer plays with a stronger chance of winning small.

Using Starcraft as an example, it certainly isn't bad to overwhelming crush your opponent. But when players are rewarded for how decisively they won, it creates the unique conditions above (imo, not inherently good or bad).

I don't see this as something to be avoided inherently. To my own sensibilities, this seems less suited to long (15+ hour) scenarios (like Civilization), and more suited to shorter scenarios (a level/round in a game). In tournament play, it could result in games that are less exciting to watch, if a player decides to resign early to prevent a larger loss if they try to play out a risky game state.

---

But, stepping away from par systems, I'd like to offer my own version of a score system based on placement (basically how most racing games work) that I think offers an elegant balance. My example is based on a turn-based strategy game I'm developing.

As the game unfolds, players are assigned points for 7 score tracks (1: the quality of life of your citizenry, 2: your current research output, 3: the total cost of all your military units, 4: your current net income per turn, 5: total value of aid you've given to other nations, 6: the average ecology value in your territory, and 7: the average of your relationship scores with every other nation).

Each score track is turned into a percentile and a placement. So for example, a player might be in the top 80th percentile for ecology, putting them in 4th place for ecology relative to other players/nations, and in the 40th percentile for military, putting them in 8th place for military relative to other nations. Each player has an overall placement/ranking (1st, 2nd, etc.), determined by the combination of all their score tracks. And of course, these values change as the game progresses.

In other words, a Mario Kart-like score system applied to a 4x. The game ends when one country/player has an overall score that's X% higher than everyone else's overall score. And the X% score required to win gradually decreases as the game goes on, so the game is always moving toward a conclusion.

This frees players from having to pre-select their precise victory path (the way Civilization rewards players for beelining/focusing on a military, culture, or tech victory), and gives them full flexibility of strategy. It also lets games continue or end organically, maintaining tension consistently.

I want to presence that a binary win/lose state is certainly ideal for some games (for example, Starcraft), and has real advantages. But it could also reduce the number of viable strategic approaches in games where we want to let the player assert themselves in different ways, and give them several unique variables for victory. I think the tech/military/diplomacy/culture win conditions in Civilization is a good example of how players are punished for using strategies that don't cater specifically to one of those win conditions; in that context, having a choice of several binary win/loss states really flattens the depth of the strategic space.
 
#20
I think if you design a game with a good match length arc that really makes sense and works then you should use binary win-loss. But, I think there's a lot of potential games that have really good mid-size arcs, that are strategic in nature (especially when constrained by a fairly simple match length arc) and for those game we should use par-like systems.
Can you expand on why you think this? I think that we should use a score system for basically all games (or at least, single player strategy games), regardless of arc structure.