Balancing Multiplayer Games, Part 1: Definitions

Balancing a competitive multiplayer game is difficult—really, really difficult. In this article I’ll define the terms that will help us discuss the topic, then in the second and third articles I’ll explain what counts as balance and some techniques to work toward it. Then in the fourth article, I’ll try to impress upon you what deep trouble we’re really in. It's a wicked problem that by its very nature is resistant to complete mathematical analysis. We'll have to do it different way.

First, the terms. Let’s start with balance and depth as defined by my former selves:

A multiplayer game is balanced if a reasonably large number of options available to the player are viable--especially, but not limited to, during high-level play by expert players.
—Sirlin, December 2001

 

A multiplayer game is deep if it is still strategically interesting to play after expert players have studied and practiced it for years, decades, or centuries.
—Sirlin, January 2002

 

 

This definition of balance is pretty good, but there are two concepts hiding inside that term viable options. On one hand, I meant that the game doesn’t degenerate down to just one tactic, and on the other hand, I meant that if there are lots of characters to choose from in a fighting game or races to choose from in a real-time strategy game, many of those characters/races are reasonable to pick. Let’s call the first idea viable options and second idea fairness in starting options, or just fairness for short.

Viable Options: Lots of meaningful choices presented to the player. They should be presented with enough context to allows the player to use strategy to make those choices.

Fairness: Players of equal skill have a roughly equal chance at winning even though they might start the game with different sets of options / moves / characters / resources / etc.

Viable Options

The requirement that we present many viable options to the player during gameplay is what Sid Meier meant when he said that a game is a series of interesting decisions (a multiplayer competitive game, at least).

If an expert player can consistently beat other experts by just doing one move or one tactic, we have to call that game imbalanced because there aren’t enough viable options. Such a game might have thousands of options, but we only care about the meaningful ones. If those thousands of options all accomplish the same thing, or nothing, or all lose to the dominant move mentioned above, then they are not meaningful options. They just get in the way and add the worst kind of complexity to the game: complexity that makes the game harder to learn yet no more interesting to play.

For the sake of depth, we also hope that the player has some basis to choose amongst these meaningful options. If the game at hand is a single round of rock, paper, scissors against a single opponent, there is nearly no basis to choose one option over the other so it’s hard to apply any kind of strategy. And yet a game of Street Fighter might be decided by a single moment when you choose to either block, throw, or Dragon Punch, or a game of Magic: the Gathering might be decided by a single Master of Predicaments decision. These examples at first glance look like the rock, paper, scissors example, but the decisions take place inside the context of a match that has many nuances where each player is dripping with cues about their future behavior. In Street Fighter and Magic, the player does have basis to choose one option over the other, and more than one choice is viable, we hope.

 
 
 

Also for depth, we prefer if the meaningful choices depend on the opponent’s actions. Imagine a modified game of StarCraft where no players are allowed to attack each other. All they can do is build their base for 5 minutes, then we calculate a score based on what they built. There are many decisions to make in this game, and it might have several paths to victory, but because these decisions are purely about optimization—more like solving a puzzle than playing a game—they make for a shallow competitive game. Fortunately, in the actual game of StarCraft, you do need to consider what your opponent is building when you decide what to build.

While we require many viable options to call a game balanced, the requirement about giving the player a context to make those decisions strategically and the requirement that the decisions have something to do with the opponent’s actions are really about depth. They’re worth pointing out though because we should attempt to increase the depth of the game as we balance it, not decrease it.

Fairness

Fairness, in the context I’m using it here, refers to each player having an equal chance of winning even though they might start the game with different options. In Street Fighter, each character has different moves, in StarCraft each race has different units, and in World of Warcraft, each arena team has different classes, talent builds, and gear. Somehow, all of these very different sets of options must be fair against each other.

I want to stress that I am only talking about options that you’re locked into as the game starts. That’s a very important distinction. Options that open up after a game starts do not necessarily have to be fair against each other at all. Imagine a first-person shooter with 8 weapons that spawn in various locations around the map. Two of these weapons are the best overall, 3 are ok but not as good as the best weapons, and the remaining 3 are generally terrible but happen to be extremely powerful against one or the other of the 2 best weapons.

Is this theoretical game balanced? It certainly might be, meaning that nothing said so far would disqualify it. A designer could decide that they want all weapons to be of equal power, but they need not decide that as long as each weapon is still a viable choice in the right situation. It might be fine to have two powerful weapons that players compete over, a few medium power weapons that are still ok, and some weak weapons that allow players to specifically counter the strong weapons. There could be a lot of strategy in deciding which parts of the map to try to control (in order to access specific weapons) and when to switch weapons depending on what your opponents are doing.

By contrast, a fighting game with 8 characters designed by that scheme is not balanced because it fails the fairness test. Players choose fighting game characters before the game starts, but they pick up weapons in the first-person shooter example during gameplay. Being locked into a character that has a huge disadvantage against the opponent’s character is unfair.

Games that let players start with different sets of options are inherently harder to balance because they must make those sets of options fair against each other in addition to offering the players many viable options during gameplay.

Symmetric vs. Asymmetric Games

Let us call symmetric games the types of games where all players start with the same sets of options. We’ll call asymmetric games the types of games where players start the game with different sets of options. Think of these terms as a spectrum, rather than merely two buckets.

 

Symmetric                          Asymmetric
<---------------------------------------------->
Same starting options                            Diverse Starting options

 

On the left side of the spectrum, we have games like Chess. In Chess, each side starts with exactly the same 16 pieces. The only difference between the two sides is that white moves first. Because of this different starting condition, we shouldn’t say that Chess is 100% symmetric, but it’s damn close. If Chess were the only game you had ever seen, you might think that the black and white sides are played radically differently; white sets the tempo while black reacts. There are entire books written about how to play just the black side. And yet if we zoom out to look at the many games in the world, we see that the two sides of Chess are so similar as to be virtually indistinguishable when compared to two races in Starcraft, two characters in Street Fighter, two decks in Magic: The Gathering, or two armies in Chess 2.

 
 

The more diversity in starting conditions the game allows, the farther to the right of our spectrum it belongs. So asymmetry, as we mean it here, is a measure of a game’s diversity in starting conditions. This is not meant to be an exact science, so there is no specific formula to determine where a game belongs on this spectrum, but it’s a handy concept anyway.

Let’s look at a few examples. StarCraft has three very diverse races so it belongs toward the right side of our spectrum. That said, even if the three races were as different as imaginable from each other, the number three is small enough that we shouldn’t put it at the far right (admittedly, this is a judgment call). Fighting games can have dozens of characters that play completely differently and they tend to have more asymmetry than most other types of competitive multiplayer games.

That said, individual fighting games can vary quite a bit in just how asymmetric they are. Virtua Fighter, for example, is an excellent and deep fighting game, but the diversity of characters is relatively low compared to other fighting games. All characters have a similar template compared to Street Fighter where some characters have projectiles, or arms that reach across the entire screen, or the ability to fly around the playfield. Meanwhile, Guilty Gear, a fighting game you might not have have heard of, has more diversity than any other game in the genre that I know of. One character can create complex formations of pool balls that he bounces against each other, another controls two characters at once, another has a limited number of coins (projectiles) that power up one of his other moves and a strange floating mist that can make that powered up move unblockable. It’s almost as if each character came from a different game entirely, yet somehow they can compete fairly against each other. Guilty Gear is possibly all the way to the right of our chart because it has both wildly different starting options (characters) and many of them (over 20!).

Magic: The Gathering is also extremely asymmetric in the format called constructed where players bring pre-made decks to a tournament. The variety of possible decks is staggering and tournaments usually have several different decks of roughly equal power level, even though they play radically differently.

First-person shooters tend to be very far toward the symmetric side of the spectrum, usually offering the same options to everyone at the start, except for spawning location. Remember that picking up different weapons during gameplay, or even changing classes during gameplay in Team Fortress 2, does not count as asymmetric for our purposes. (Again, because those different options don’t need to be exactly fair against each other.) Also, first-person shooters that do have asymmetric goals for each side often make the sides switch and play another round with roles reversed so that the overall match is symmetric.

Now that we’ve mapped out where some games fit on our spectrum, remember that this is not a measure of game quality. If your favorite games appear on the left (symmetric) side, that does not mean they are bad. If you like StarCraft more than Guilty Gear, you do not need to be upset that Guilty Gear is “more asymmetric.” The spectrum is simply meant to give us an idea about how different the starting options of a game are, not about the depth or fun of the game. That said, I do personally prefer asymmetric games and they are inherently more interesting to me.

No matter where a game appears on this spectrum, it still needs offer many viable options during gameplay to be balanced. In addition to this, the farther a game is to the right of the spectrum, the more it needs to care about balancing the fairness of the different starting options. In the next part of this series, I’ll talk about how we can design games that make sure to offer enough viable options and in the article after that, I’ll explain how we can attempt to create fairness in those pesky asymmetric games.

Balancing Multiplayer Games, Part 2: Viable Options

In the previous article I divided the idea of balance into the two sub-concepts of viable options and fairness. I also defined the concepts of symmetric and asymmetric games, where the more varied the different starting options are that must be fair against each other, the more asymmetric the game is.

How do we make sure we have enough viable options during gameplay?

Yomi Layer 3

The worst thing you can have in a competitive multiplayer game is a dominant move (or weapon, character, unit, whatever). I don’t mean a move that is merely good, I mean a move that is strictly better than any other you could do, so its very existence reduces the strategy of the game. A dominant move also probably has no real counter, so even if the opponent knows you will do it, there’s not a lot they can do.

To protect against dominant moves, we should be aware of the concept of Yomi Layer 3. I wrote a chapter in my book about it, but I’ll quickly summarize it here. “Yomi” is the Japanese word for “reading,” as in reading the mind of the opponent (and it’s also the name of my strategy card game). If you have a powerful move and use it against an unskilled opponent, I call that Yomi Layer 0, meaning neither player is even bothering with trying to know what the opponent will do. At Layer 1, your opponent does the counter to your move because they expect it. At Layer 2, you do the counter to their counter. At Layer 3, they do the counter to that.

Let's look at an example of a Yomi Layer 3 situation in Street Fighter HD Remix. Honda wants to do his torpedo move get close to Ken, but Ken throws fireballs to prevent this.

I gave Honda the ability to destroy these fireballs with his torpedo, but only with the jab version of the move that doesn’t travel very far. If Honda can destroy a fireball with it and end up closer, that’s good for him. Another similar option is that Honda can do his flying buttslam move to avoid Ken's fireball and land on him.

Ken can counter either of those by not throwing the fireball in the first place and letting Honda do the jab torpedo or buttslam. As Honda is moving forward with his jab torpedo, Ken can walk forward and sweep, hitting the recovery of the jab torpedo. Against a buttslam, Ken might walk backward a little bit, then sweep to hit the recovery.

I did not need to add anything to allow for Yomi Layer 4 though because Honda can counter Ken’s wait-and-sweep options by simply doing the original, full-screen torpedo. Yomi Layer 4 tends to wrap around like this in competitive games.

Summary of the options:

Honda: torpedo that goes far -OR- jab torpedo that destroys fireballs / buttslam
Ken: fireball -OR- wait and sweep

This type of thing is very common in just about any competitive game. To put it more generally, you and your opponent each have two options:

You: A good move and a 2nd level counter
Opponent: A counter to your good move and a counter to your counter

The designer generally does NOT need to design Yomi Layer 4 because at that point, you can go back to doing your original good move. 

The Yomi Layer concept is a reminder that moves need to have counters. If you know what the opponent will do, you should generally have some way of dealing with that. As you go through development of a game, ask yourself if various gameplay situations you find yourself in support Yomi Layer 3 thinking. If they don’t there might be a dominant move in there somewhere, which is bad.

Local vs. Global Balance

Does every possible situation in a game need to support Yomi Layer 3?
Answer: no.

Does every possible situation in a game even need to be fair to both players?
Answer: definitely not.

Remember that I defined fairness by the overall chance of winning, given different starting options. Think of that as a global term, in that it applies to the game as a whole from the start of gameplay until someone wins. But the local level, meaning a particular situation in the middle of gameplay, does NOT need to be fair. Even symmetric games like Chess are supposed to have unfair situations. When you have 3 pieces left and the other guy has 9 pieces left, it’s supposed to be unfair to you. Or in StarCraft, if we find that two Zealots beat (or lose to) 8 Zerglings—even though they cost the same resources to make--that is perfectly fine. We don’t care if local situations like that are unfair or not, we only care if Protoss is fair against Zerg.

Checkmate Situations

I call a situation a checkmate situation if it means that one player has almost certainly won, even though the game isn’t actually over. For example in Super Street Fighter 2 Turbo, if Honda lands his deadly Ochio Throw against Guile in the corner, he can then follow up with a series of moves (involving more Ochio Throws) that virtually guarantee victory. Human error could change the outcome, but as soon as you see that first move, you know it should be a checkmate.

Are checkmate situations ok? They clearly violate our requirement that there be many viable moves (Honda really only has one option here and Guile has no good options). They clearly violate the concept of Yomi Layer 3. And yet, the answer is that checkmate situations can be ok. It’s sooooo hard for Honda to get close to Guile in this match, that if he does, he basically deserves to do 100% damage. All the gameplay that takes place before the checkmate is pretty good, and even though Honda can do this abusive thing up close, the match is still heavily in Guile’s favor overall.

I’d like to point out the other side of this argument though. Some players think that even though Guile has the advantage in this match, Honda’s ability to repeat that Ochio Throw is too degenerate. They say yes he needs it to win, but the game would be better overall if things weren’t so extreme. If only Honda could get close to Guile a little more easily, then he would not need a checkmate situation.

I think Rob Pardo, VP of Game Design at Blizzard, echoed this sentiment in a lecture he gave at the Game Developer’s Conference on multiplayer balance. He said that “super weapons” in real-time strategy games are generally a bad idea. They leave the victim feeling that there is nothing they could have done (checkmate!). He explained that even though the Terran nuclear missile in StarCraft looks like a super weapon, it has many built-in weaknesses: a ghost unit must be nearby the victim’s base, there is a red targeting dot on the victim’s base, and a 10 second countdown is announced to the victim, giving him time to destroy the ghost to prevent the nuclear missile.

Pardo has a good point and so did the players who complained about Honda. Even though I think checkmate situations can be ok, it’s telling that when it was my turn to make the decisions, I removed Honda’s checkmate situation in Street Fighter HD Remix. In that game, I gave him an easier time getting close to Guile, but replaced his checkmate situation with a Yomi Layer 3 situation so there’d be more viable decisions throughout the match.

Lame-duck Situations

Lame-duck situations are just like checkmate situations, but with one difference: time. Honda’s checkmate situation takes something like three seconds to get through. But consider a similar situation in the fighting game Marvel vs. Capcom 2. In that game, each player has a team of three characters: one on the playfield and two on the bench. Players can call in one of their benched characters for an assist move at any moment, letting them attack in parallel with their main character and assist character at the same time. Or better yet, they can stagger the attacks so that each attack covers the recovery period of the other.

When one player is down to their last character, they can no longer call assists. Fighting with just one character against an opponent with two or three characters might as well be checkmate, almost all the time. The problem is that it takes excruciatingly long for the match to actually end. It takes so long, that I call that last portion of the game the lame-duck portion. Other fighting games are exciting right up to the last moment, but a lame-duck portion of gameplay means the real climax is somewhere in the middle, and then players are forced to act out a mostly pointless endgame while spectators lose interest. Yes, on rare occasions someone pulls off an amazing comeback, comebacks also happen in games without lame-duck endings, so that’s not a good argument.

The lame-duck situation was specifically addressed in Marvel vs. Capcom 3 by adding the X-Factor mechanic. That's a mode you can activate once per game that powers you up a huge amount for several seconds. It powers you up more the fewer characters you have left. Activating X-Factor when you're down to your last character is so powerful that you can reasonably hope to defeat one or two of the opponent's characters before your X-Factor runs out.

While a checkmate situation is maybe ok, you should try to avoid game designs that allow for long lame-duck endings. Both Chess and StarCraft have this undesirable property, and it just means that players often concede the game before the actual end. Those games also show that it’s not the worst thing in the world to have lame-duck endings (because Chess and StarCraft are still good games), but you should still avoid them as a designer if at all possible.

Chess 2 avoids almost all those lame duck situations with the inclusion of the midline invasion rule. That rule is an alternate win condition: if you cross the midline of the board with your king, you win. If you are really so far behind that you can't reasonably come back by checkmate, then the opponent will probably have already won by midline invasion. Furthermore, it's easier to make a comeback when that rule exists than if it didn't, so we both eliminated lame-duck situations and allowed for more exciting comebacks with a single new rule.

Explore the Design Space

Design space is the set of all possible design decisions you could possibly make in your game. Whether your game is symmetric or asymmetric, it’s usually a good idea for your game to touch as many corners of the design space as possible. This helps give a game depth and nuance, but also tends to protect you from dominant moves.

For example, in the virtual card game I designed called Kongai, each character has four moves. When a move hits, it has a percentage chance to trigger an effect. For a given character, we could vary the damage, speed, and energy cost to come up with four different moves. If that’s all we did, though, we’d be missing out on a chance for more diversity in the game, and we’d get dangerously close to making some of those moves strictly better than others which would reduce the number of viable options. Instead, I tried to explore the design space as much as possible with different effects. One move can change the range of the fight from close to far, which is usually only possible before the attack phase. Another move deals enough damage to kill every character in the game, but only four turns after you hit with it. Another move can hit characters who switch out of combat, even though switching out usually beats all attacks.

The point is that by exploring the design space as much as possible, it’s a lot harder for players to judge the relative value of moves. How good is a 90% chance to change ranges during combat as opposed to a 95% chance to hit a switching opponent with a weak move? It’s hard to say and depends on a lot of factors, and that’s good because it means each move is likely to be useful in some situation and knowing when is an interesting skill to test. Incidentally, I call that skill valuation.

Players want you to explore the design space, too. When everything is too similar in a game, it feels like one-note design rather than a symphony. The more nuances and different choices you present, the more each player can express his own playstyle.

Wheat from the Chaff

Here’s my favorite quote from Strunk & White’s The Elements of Style:

Omit Needless Words
Vigorous writing is concise. A sentence should contain no unnecessary words, a paragraph no unnecessary sentences, for the same reason that a drawing should have no unnecessary lines and a machine no unnecessary parts. This requires not that the writer make all his sentences short, or that he avoid all detail and treat his subjects only in outline, but that every word tell.


Treat your game design the same way. Yes you should explore the design space, but omit needless words, mechanics, characters, and choices. Although your primary goal regarding viable options is to make sure you’re giving the player enough options, your secondary goal should be to eliminate all the useless ones.

Marvel vs. Capcom 2 has 54 characters, which is ridiculously many. How many are viable in a tournament? I’ll say 10, and I’m being generous. I actually call that a success because coming up with 10 characters in fighting game that are fair against each other is really hard. That said, it does look pretty bad to have more than FOUR TIMES that many characters sitting around in the garbage pile of non-viable choices. Compare this to Super Street Fighter 2 Turbo’s 16 characters, almost all of which are tournament viable; or Guilty Gear’s 23 characters, almost all of which are viable; or Yomi's 20 characters, all of which are tournament viable, and you see what a compact design looks like.

One genre of game is notable for intentionally creating an enormous number of useless options: collectable card games. Even though Magic: The Gathering has many good design concepts and several viable decks at any given time, when we judge it on how much chaff it has—cards/decks that aren't viable in the tournament environment—it's one of the worst situations of just about any game.

MTG’s Mark Rosewater defends the intentional inclusion of bad cards for design reasons, but really it's entirely marketing driven. Their business is to put a collectable barrier between you and the the cards you want so as to sell more packs of those cards. That means mountains of chaff cards are inventible. I think even one single chaff card is a problem that would ideally be fixed, but Rosewater claims that bad cards are ok because they:

a) allow for interesting experimental mechanics that might end up being bad
b) test valuation skills because if all cards were equally good, there’d be less strategy
c) give new players the joy of discovering that certain cards are bad, as a stepping stone to learning the game
d) are necessary because even if they came out with a set that consisted entirely of known good cards from old sets, there’d still be only 8 tournament viable decks and the rest of the cards would not be used.

The solution to this problem is clear if we only cared about design and not rip-off marketing: print fewer cards. Reason a) is a great one, experimental cards that end up accidentally bad are fine. Reasons b) and c) are just silly. Saying the game would not have enough strategy if bad cards were removed is an insult to Mark’s own game. Saying that new players need to discover the intentionally bad cards is even more silly because this comes at the cost of making sets overwhelming to new players and needlessly unwieldy for expert players. We all know the real reasoning here is to make players buy more random packs of cards to get at the few good ones.

Finally, reason d) is a blatant admission that the game should have fewer cards. Ironically, I’m not even sure d) is true. Maybe printing a large set of all good cards really would lead to more viable tournament decks than the game currently supports. If not though, they should stop printing all that chaff. Here's an example of a card that doesn't need to printed:

You could say that MTG proves that it’s really all about chaff, though. Giving a few viable options amidst a sea of bad ones is probably good business when you sell by the pack. But we don’t see this in other genres and really we just haven’t seen anyone crazy enough to stand up to MTG on this issue and offer a competing card game that’s just as well designed but that eliminates all chaff. (Disclosure: Codex is that game and is in development now.)

League of Legends is also notable for its chaff, though to a lesser extent than Magic: the Gathering. When you sell by the character, it's hard not to make more and more characters forever—way more than would make sense to try to balance. Such a game might only need 30 characters to cover all necessary archetypes and it could then be balanced really well. When you go past 100 in such a game, you then need a bandaid system of character bans at the top end and you're practically guaranteed to have a bunch of chaff characters at the bottom end. But it sure would make more money to sell 100 characters rather than 30.

Lots of chaff and/or having to ban characters routinely is pretty ugly design.

Double-blind Guessing

I used the technique of double-blind guessing in both my Yomi card game and my Kongai virtual card game (that one’s actually a turn-based strategy game dressed up like a card game). Anyway, the idea is to make all players commit to a choice before they know what the others have committed to. This is the same setup as the prisoner’s dilemma.

I learned this concept from fighting games. Though they appear to be games of complete information because you can see everything the opponent can see, fighting games are actually double-blind games. They come down to very precise timing and the moment you jump, you often don’t know that the other guy threw a fireball. You only know that 0.3 or 0.5 seconds ago he didn’t. It takes a small amount of time for the opponent’s move to register in your brian, and though it might seem insignificant, it’s actually critical to fighting games even working as strategy games at all.

Real-time strategy games like StarCraft have the same property, but on a much slower time-scale. You often do not know exactly what the opponent is building in his base at the moment you must decide what you should build. Even if you were able to scout his base, you might be working on information that’s several seconds old, so you have to guess what he did during that time.

If we were to remove the double-blind nature from my two card games Yomi and Kongai, and from fighting games and real-time strategy games, I think all of them would be broken. All those games need double-blind decision-making to be interesting. This design pattern is a way to increase the chances that you have many viable moves in your game because it naturally forces players into the Yomi Layer 3 concept I talked about earlier. Weaker moves become inherently better in a double-blind game because it’s easier to get away with doing them without being countered. I’ve even joked that some matches between the world’s best Virtua Fighter players are “a battle of the third-best moves.” Sometimes the players are so paranoid about doing their “best” option for fear of being countered, they fall back on a third best option that no one would ever counter (though it’s quite a sight when the opponent counters even that!). If no guessing was involved at all, players would not use third-best moves.

Playtesting

Finally, playtesting, especially with experts, is how you figure out where your problems really are. Do the experts ignore some vast portion of you game’s moves? Have they discovered a bunch of checkmate situations that you didn’t know about? Do you see them using a variety of strategies?

How to use playtests is really a whole topic of its own, but here’s a few points to keep in mind. First, be skeptical of them. Gamers tend to overreact to changes and claim that no counters exist to some strategies when counters do, in fact, exist. It can take years to sort out what is really effective in a game, and playtesters during your beta are only on the first few steps of that long journey. If they find what looks like the best strategy in the game, it might just be that they have found a local-maximum. Maybe some radically different way of playing that they have not yet discovered ends up being more powerful. This is actually par for the course in fighting games and it hasn't been much different in my work on tabletop games, either.

That said, playtests are really all you have. Theory is not a substitute for experts playing against each other and trying their hardest to win. I think everyone knows they need playtests, but the hardest question is who do you listen to when all your playtesters disagree, and how do you know when playtesters are wrong about how powerful something is? That question is so hard that I’ll save it for part 4 of this series when I tell you how much trouble we’re really in trying to balance a game at all.

Conclusion

To ensure we have many viable options, building in counters with the Yomi Layer 3 system is a good start. Not all situations need this though, and checkmate situations might be acceptable, but you should avoid their their longer cousins, lame-duck situations, if possible. Explore your game’s design space by offering moves as different as possible because this technique has a good chance of making all moves useful somewhere and it makes it very difficult to determine what the best moves really are. That becomes an interesting skill test for players. Eliminate all the worthless options because they confuse the player and add nothing, but they make you a lot of money in a certain genre. The double-blind guessing mechanic helps keep more moves viable than otherwise would be.

And finally, all the theory in the world does not substitute for playtesting.

Balancing Multiplayer Games, Part 3: Fairness

In asymmetric games, we have to care about making all our different starting options fair against each other in addition to making sure the game in general has enough viable options during gameplay. That means each character in a fighting game and each race in a real-time strategy game should have a reasonable chance of winning a tournament in the hands of the right player. For collectable card games and team games like Guild Wars, World of Warcraft’s arenas, and DOTA2, at least “several” possible decks, class combinations, and heroes should be able to win tournaments. Furthermore, we'd hope that there's never a card, class, or hero that must be part of your composition, and that there aren't any that are so bad that they can't reasonably be used in any winning composition. We'd hopefully have much higher standards than even that, but that's just a minimum level of competence to shoot for. (This assumes no player-run banning system either; that's already below the minimum level of balancing competence to shoot for.)

There are some ways we can affect the chances of our success in balancing before we even get to the balancing part. It has to do with having a solid foundation, or a shaky one. 

Self-Balancing Forces

Self-balancing forces fix balance problems without us having know exactly what those problems even are. If you're in charge of balancing a game, it probably has some problems you know about and some problems you don't know about. The ones you don't know about are the bigger danger, so anything that reduces or eliminates those is very valuable. (It reminds me of The Sheathed Sword.)

The fighting game Guilty Gear is a good example of this technique. It has an extremely diverse set of characters, yet somehow manages to be pretty well balanced. I covered how that's possible in this article. The short summary is that they built in failsafes such that every character has protections from getting abused too badly. Every character has guard meter, progressive gravity, green blocking, burst, and more.

Guard meter helps stop combos from going on forever by making your hitstun get shorter and shorter as you get hit by more moves in a combo. Likewise, progressive gravity ends combos eventually by making you fall faster and faster while you're being air juggled. Green blocking lets you push attackers away when you block, which allows you to avoid potentially inescapable lockdowns. When all else fails, the burst feature lets you escape any combo about once per round.

Each of these features solves balance problems that the designers didn’t even know they had. They don't have to know how to do perfect trap of blocked moves that you can never get out of. They don't have to know how to do a juggle combo that goes on forever. They've solved all these problems generally and globally, without needing to know exactly which cases might have caused a problem. This technique is difficult to apply to many games, but it's very valuable if you can.

A Shaky Foundation

You can fail before you even start by having way too many characters or way too much customization.

Yomi has 20 characters, which means it has 190 matchups in the 1v1 mode, not counting mirror matchups. That's very difficult to balance, but possible. You can read about that process here. Imagine if it had 100 characters though. Five times as hard to balance? Actually that's over 26 times the number of matchups. The difficulty in balancing isn't linear though, so it's probably 100x as hard to balance. A card game like Yomi or a fighting game like Street Fighter doesn't need anywhere near 100 characters. 20 or 30 is already a lot, so a game with 100 is setting itself up for balance failure from the start.

Customization is also a dangerous thing. Similar to having many characters, having lots of customization feels like it gives players more options. Not more viable options though, it actually tends to give them far fewer. Consider Yomi again. It has 20 fixed (non-customizable) decks, each with 55 cards. All 20 decks are tournament viable. What if we took the equivalent number of cards (55*20 = 1100 cards) in a different game and allowed full customization to create any 55 card deck you wanted. That game would have over 377,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000 possible decks. That's more decks than there are grains of sand on Earth. More than the number of stars in the galaxy. In fact, it's more than the number of particles in the observable universe. So that's what we're balancing?

What do you think the chances are that the top 20 of those 377,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000 decks are about the same power level? I think it's about 0% chance. In all likelihood there is only one playable deck, or maybe 2 if you're lucky. If you fixed those one or two decks, you would not find 20 decks of equal power level just below it, either. At any given time, at most 1 or 2 decks (or about 0% of total possible decks) would be viable. This kind of massive customization is almost inherently unbalanceable.

Don't try to fix this by having players ban each other's decks or characters. That's a cop out and it feels terrible to identify with a character, to learn its gameplay and have it fit your personal style, then to be told it's banned for some reason. Don't do that to players. Instead, do the hard work of making your game actually balanced and start with a foundation that has some hope of ever being balanced enough that several different starting options are fair. Let's get to the process of achieving that now.

Playtesting and Course-correcting

Suppose you have a game to playtest. You designed a diverse set of characters / races / whatever and made each one coherent and interesting. During that design time, you should have the confidence that you’ll sort out the balance problems in playtesting. All the theory in the world will not save you from playtests, so there's no way you could have gotten it perfect (or even acceptable) yet.

You need to start tuning the game, and react and learn as you go. Do not let a producer turn tuning into a fixed list of items that you are accountable for checking off, one by one. It’s an organic, continuous process that keeps going until you need the ship the game. Playtesting lets you discover things you couldn’t have predicted ahead of time, and you should be open to those discoveries. The goal isn’t to make the exact game you originally envisioned, because your original vision did not take into account all the things you learned from development and playtests. When you or the testers discover nuances or unexpected properties, you have the chance to build around those and incorporate them into the game’s balance.

The Tier List

During the balancing of Street Fighter, Kongai, and my card game called Yomi, I used a similar approach with playtesters. I think this approach doesn’t really depend on the genre, and the key idea is managing the tier list.

The term “tier list” is, I think, a term from the fighting game genre. It means a ranking of how powerful each character is from highest to lowest, but it also accepts that such a list cannot be exact. Instead of ranking 20 characters from 1 to 20, the idea is to group them together into “tiers” of power.

A linear list would be bad for two reasons. First, players might think characters X, Y, Z are about the same power level, but would have to list them as something like 6th, 7th, and 8th. That creates data incorrect data that's indistinguishable from if they think character X is far better than character Y. Second, a linear list would not give you an idea about the gaps in power the players are experiencing. Is the #1 ranked character much more powerful than all others? Or maybe the #1 and #2 are similar in power, but they are both higher than the rest? Or maybe #1, #2, and #3 are similar in power? A tier list will show you this, but a linear list won't. Use a tier list, meaning each tier is the set of characters who are a similar power level to each other. 

Remember that if a divine being handed you a 100% perfectly balanced game, that players would still make tier lists. You should accept the existence of these lists from players as a given, and its your job to manage this list. In Kongai and Yomi, I even gave the players a template for the tier list that is most useful for me as a designer. First, I tell them to think of three tiers: top, middle, and bottom. Then I tell them about the two “secret tiers” that I hope are empty.

0) God tier (no character should be in this tier, if they are, you are forced to play them to be competitive)
1) Top tier (don't be afraid to put your favorite characters here. Being top tier does not necessarily mean any nerfs are needed)
2) Middle tier (pretty good, not quite as good as top)
3) Bottom tier (I can still win with them, but it's hard)
4) Garbage tier (no one should be in this. Not reasonable to play this character at all.)

My first goal of balancing is to get the god tier empty. Of course some character will end up strongest, or tied for strongest, and that is ok. But a “god tier” character is so strong as to make the rest of the game obsolete. We have to fix that immediately because it ruins the whole playtest (and the game). Also, the power level of anything in the god tier is so high, that we can’t even hope to balance the rest of the game around it.

My next goal is get rid of the garbage tier characters. They are so bad that no one touches them, and it’s usually pretty easy to increase their power enough to get them somewhere between top, middle, and bottom. If they are somewhere in those three tiers (which gives you a lot of latitude actually), at least they are playable.

Public Tier Lists

I really like it when playtesters all see each other’s tier lists. The debate this spawns is very useful for me to read (or overhear in person) and for the playtesters to sort out their ideas. Sometimes when someone put a character unusually high or low on the list, I dug deeper to find out that player really did know something most of the rest of us didn’t. Other times, that player is just crazy and the rest of the testers are happy to point that out. It’s also good to see what kind of consensus the testers come up with, like if they all rank a certain character as the worst, for example.

The biggest landmark moments in each of the games I balanced was when the tester communities consistently gave tier lists with no characters in the god tier or garbage tier. Once you’ve achieved that, the next goal is to compress the tiers. That means that you want the difference between the best and worst characters to be as small as possible. Notice that that means even if you have the same characters in the bottom tier that you did a month ago, you might have dramatically improved the game if all those “bad” characters are really only a hair worse than the tier above, rather than way worse.

compressed_tiers.jpg

Adjusting the Tiers

In all the games I balanced, I used the same approach of letting the top tier set the benchmark power-level. In Street Fighter, I already had an established top tier as a starting point from the previous game, but in Kongai and Yomi, it was somewhat accidental who ended up in the top tier. But early on, after the god tier was removed and it was pretty clear which characters / decks were top, I allowed that to be the target power level. In other words, the characters in that tier are “how the game is supposed to be.” Again, I didn’t plan exactly who would be here, but I accepted how it ended up and worked with it. So if the top tier is the target, it’s the bottom tier you should adjust the most. If the top tier is the intended power level, you don’t really want to mess up the good things you have going there. Instead, boost the bottom characters up and compress the tiers as much as you can, so you get the worst characters just barely below or equal to the best characters.

There are some psychological factors that I saw over and over again while making these adjustments. The first is that whenever I make a move or character worse (aka “nerfing”), players overreact. Sometimes that top tier creeps a little too high in power, or an otherwise average character ends up having something unexpected that’s crazily good, or a character has a move that really reduces the strategy in the game and needs to lose that in exchange for gaining something else. There’s lots of reasons for nerfs.

I’ll use some made-up numbers to convey the general idea here. Imagine a move is at power level 9 out of 10, and that’s just too good for that character. Time and time again, I saw that if I made the power level an 8 out of 10, playtesters would complain that the move was worthless and put the character down at least one tier. This happened consistently, and even in the cases where 8 out of 10 was still too powerful and it really needed to be a 7. For some reason, players in every game seem unable to grasp the concept that a top tier character who is made slightly worse can still be a top tier character.

This is one of the cases where I think you just can’t listen to the playtesters. Ignore their first reactions to nerfs, let them play it more and get used to it, let them see if they can still be successful with the new version of the move, then take their feedback on that move or character more seriously.

The other psychological effect to know about is what happens when you increase a move’s power. I learned about this Rob Pardo’s lecture on balancing multiplayer games at the Game Developer’s Conference, and I tried it on all the games I balanced, and I think Rob is right. He said that if you have a move that you’re not really sure how to balance, make it too powerful. If you make it too weak, then you run the risk of no one using it at all. Then, when you slightly increase its power, none of the testers will notice or care. They already decided that move is weak. Then if you make it slightly more powerful still, they still won’t care. Even when you inch it up past the reasonable level of power, it’s hard to get it on people’s radar and that makes it really hard to know how to tune the move.

Instead, Pardo said to start with the move too powerful. Then everyone will know about it and care about it. I did exactly this with T.Hawk, Fei Long, and Akuma in Street Fighter HD Remix, because I had trouble figuring out their power levels. Each one of those characters was the best character in the game at some point in development, and that meant I got lots of feedback from testers about these characters. It also gave me a sense of where the top of the scale even was. Sometimes my “too powerful” versions of a character would end up waaaaay too good, or sometimes just barely too good. By knowing where the upper limit was, it helped me pick appropriate power levels more quickly. That said, I did have to deal with the inevitable cries that follow all nerfs, but that just goes with territory here.

Illusions in Tiers

Another point from Rob Pardo’s speech on multiplayer games was not to balance the fun out of things. I’m very conscious of this as well. Don’t just think about the game as some abstract set of numbers that has to line up. You also have to think about how people will perceive it and whether it’s actually fun. Pardo said that he likes the player to feel like the tools they have are extremely powerful, even though they are actually fair.

An example of this in one of my games is Tafari, the Trapper in Kongai. Tafari’s main ability is that the enemy cannot switch characters while fighting him. Switching characters is one of the game’s main mechanics, so fighting him is like playing rock, paper, scissors with no rock. It seems, at first glance, ludicrously powerful. But from the start, I gave Tafari several weaknesses and he loses many fights if he ends up having to fight on even footing. He’s best when you bring him in against an already-weak character to finish them off.

I knew Tafari was not too powerful. I tested him with many experts and they tended to rank him as middle tier once they got the hang of him. As we added new testers over time, probably nearly 100% of them claimed that Tafari was too strong. I refused to change him though and after a year of testing, the best players still ranked him as middle tier, while inexperienced players still ranked him as top. Tafari is an illusion.

I’m telling you this because you have to be very careful with feedback in cases where you intentionally made something feel more powerful than it actually is. It’s a success if you can pull that off though, because Tafari makes the game more interesting, creates lots of debates, and at the end of the day, he is balanced.

Counter Matches

In addition to the tier list, you should also be thinking about all the specific matchups. Street Fighter HD Remix, for example, has 17 characters and 153 possible matchups. For the version of Street Fighter before HD Remix, experts tend separate the characters into four tiers (none of them are god tier or garbage tier), and they place Guile in the respectable second tier. Even though that means Guile’s power level is acceptable, he is severely disadvantaged in two specific matches: Vega and Dhalsim. Is it ok that an overall good character gets countered by two specific characters? Not really.

If these were weapons in an FPS or units in an RTS or characters in team-based fighting game, then it might be acceptable. You pick up weapons in an FPS after the game starts, so their balance doesn’t need to meet the hard requirements of an asymmetric game. And units in an RTS and characters in team-based fighting game are examples of local imbalances, which are fine (it’s the races and teams that need to be balanced). But in Guile’s case, you lock in your choice of Guile at the start of the game, then you are stuck with him the entire game, so it really is a problem if he has some bad counter matches, even though players rate him fairly highly overall.

It’s really tricky to adjust anything in an asymmetric game though. How can we help Guile in just the Dhalsim match without affecting all the other matches? There’s no easy answer here, but I advise you to really solve the problem, rather than copping out.

My real solution to this problem was two-fold. First, for reasons unrelated to this particular match, I changed the trajectory of Guile’s roundhouse flash kick. This happened to help a bit against Dhalsim’s fireballs, so we’ll count that as a lucky accident. Second, one of Guile’s problems is that Dhalsim’s low punches can go under Guile’s Sonic Boom projectiles and hit Guile from across the screen, with no repercussions. I changed Dhalsim’s hitboxes so that Dhalsim now trades hits in this situation, rather than cleanly hits. This change has virtually no effect on any other match, so it’s a real solution to the problem.

A cheating solution would have been to special case this match and give Guile more hit points. This sounds attractive because you don’t have to worry about messing up other matches, but this non-solution feels really artificial. It messes with players’ expectations and intuitions about how many hit points Guile has.

A similar cop out would be to create a giant table in an RTS of every unit versus every unit and special case how much damage they all do to each other. Again, it messes with player intuition about how damaging each unit is, and creates and invisible, wonky system. I know you’re going to be tempted to use these types of special case solutions when balancing asymmetric games, but try your hardest to avoid them.

Conclusion

Start your design with some self-balancing forces and fail-safes if you can. Then go wild and create all your game’s diversity, then start the long road of playtesting. As you learn more from playtesting, change your course as you go. Start keeping track of tiers, first by fixing the god tier, then by fixing the garbage tier. Then compress the tiers so that even the bad characters are only slightly worse than the best characters. Finally, fix all the counter-matches you can by actually solving the puzzle, and avoiding cop out solutions.