The Ludite

Want to start an argument? Start talking about religion or politics. This works especially well if you have several outspoken people in the group to get the ball rolling. If you want to maximize the vitriol, make sure that nobody involved in the conversation knows anyone else very well, so that nobody has any common ground or any context for the things people say. People will misunderstand and talk past each other to the point where nobody will make any progress, and the conversation will end with everybody hating everyone else. Mission accomplished!

So how do you start an argument on the internet? …well, pretty much the same way. But do you know how else to start an argument on the internet? Start talking about game balance. Now, there’s a topic to get the blood boiling, and you can’t get through your average forum thread without several salvos of “lol nub” and “git gud.” What’s interesting is that this conversation bears an uncanny resemblance to the religion or politics conversation. Everyone comes from a different place in their understanding of the topic, but in their experience theirs is the right one. There are plenty of people talking past each other, there’s plenty of dismissal, and everyone is definitely outspoken.

But who’s right and who’s wrong? It’s easy to write the whole thing off as opinion, and yet people win and lose games using this information, so there must be some objectivity to it all. In fact, the premise of a competitive game is founded on a form of objectivity: generally speaking, the winner is incontestably better than the loser. So these players have good reason to believe they’re right. Their experience playing the game is dedicated to learning this stuff. Players trying to eke out any advantage they can over their opponents will, as a matter of course, find things that help them to win more easily.

But a player who hasn’t gotten there yet might find the same new techniques overpowering, at least until he learns a little more about how they work and how to counter them. Even the best players often advance their knowledge of the game, and by proxy so does the rest of the community at large. This is a sign of a healthy game-system – if people at all skill levels are still adjusting what they think it means to be a good player, that means there’s still undiscovered depth for players to chart.

But in the meantime, how do we sort out all these conflicting viewpoints? How can the same character be both too strong and too weak? Who is right? Well, would you believe… everyone? That may sound like a plea for, “Why can’t we all just get along?” but it’s absolutely true. The reason it’s true, however, is because we asked the wrong question. We shouldn’t be asking “Who is right?” What we should be asking is, “Who is the most right?” Let me explain.

Climbing the Heuristic Tree

When new players are learning how to play a new game, there’s a lot to take in. This is doubly true of games with a huge character roster, a massive tech tree, or some other form of an excessively large rulebook. The vast majority of early gameplay will be spent learning what things are in the first place and what they even do. Players don’t really have time to formulate much in the way of strategy when they’re confronted with new rules four or five times a game. Tactics that they thought were effective get constantly and effortlessly wrecked by something else they didn’t know existed, and often not even on a strategic level but rather on a shallow, “Oh, that’s a thing,” sort of level. There are a lot of moments of, “Well, obviously I wouldn’t have done that if I’d known he has a massive stun ability!”

This dynamic is why new players seem to suck so much more in Dotalikes than in other games, even if they’re very good at games otherwise. Indeed, even if they’re good at other Dotalikes. You can be the most brilliant player in the world, but you just can’t make good moves if you don’t even know all of the rules, and learning the rules takes a long time when there are maybe 120 characters with at least four abilities and a passive each. Just look at the rules for invisibility in Dota 2 – this page alone has so many unintuitive quirks that learning to play the game can be like studying for an exam in organic chemistry. As a result, the quickest path to improvement is to just go with whatever shallow, first-order strategy you can decipher with the character you’ve got, and then focus all your efforts toward learning new things.

But sometimes a shallow, first-order strategy is actually super powerful. It dominates new players, even if the player using it is new himself. Oddly, it might not really be so strong in high-level play, where people have learned how to counter it, but you can bet that new players are going to be screaming for a nerf on the nearest forum. Why the disparity? It’s because countering the strategy, while possible, is much more difficult and requires advanced knowledge of the game. At its very worst, it requires knowledge of a very obscure part of the game, something that otherwise rarely comes up, but whose existence completely invalidates the strategy from the start, if only players knew about it. This is how we get “pub stompers” – characters that consistently ruin new players in low-skill games. The hallmark of a pub stomper is simple: the character is easier to play than he is to counter.

Probably the most common example of a pub stomper is the stealth character. Fighting against an opponent you can’t see is absurdly difficult for new players. The game up until this point has been training them to identify threats using visible information on the screen. If it’s worth knowing, it has some visual cue to signal to the player that something is up. So new players make the extremely reasonable assumption that if there’s nothing there in a spot they can see, there really is nothing there. And then – SQUISH. The game suddenly and violently proves them wrong. And since they don’t see any reason why they should be squished at one moment and not another, the squishings will continue until rage-quits ensue.

Meanwhile, from the stealth player’s perspective, everything is perfectly normal. There’s nothing out of the ordinary other than the fact that his opponents are making laughably bad plays they would never make if only they knew where the stealther was. His hapless victims have no context at all for what a stealth character does, not just because they can’t see him, but because they don’t know how a stealth character ought to be played in the first place. If you can’t see him, he could be anywhere for all you know.

That’s not actually true, of course – there are limitations on where he can be and more importantly incentives for where he wants to be. But a new player has no time for the advanced calculations of where he thinks a stealther is – he’s still struggling with his own play. And on and on he struggles as he gets caught again and again by the stealther, for no reason he can decipher.

To the new player, it seems like it takes more effort and skill to play against a stealth character than to actually be the stealth character, and that feels unfair. This becomes especially pronounced when a frustrated player actually tries the stealth character. Chances are, he’ll suddenly do spectacularly well by catching other players at his skill level with all the same traps he fell into – traps that are completely intuitive for him, but every bit as mystifying to his opponents. To him, it seems conclusive – this stealth character is way too powerful. Yet often there’s more to the story.

Take for example Riki from Dota 2. Riki is permanently invisible. He becomes visible only when attacking and returns to invisibility in seconds. As an agility carry, he scales in damage very quickly. He can silence opponents in an area and even shut down their basic attacks, meaning that opponents generally have no way to fight back and thus have few choices other than to flee. And on top of all of that, he deals bonus damage from behind, so fleeing is also a poor choice. In short, he’s devastating against new players. An inexperienced player never knows where Riki is, he will push too far by himself in an otherwise open lane, he’ll get hit by unexpected burst damage and be unable to hit back, and he’ll die even faster trying to run away. All of the basic heuristics that players learn first, Riki seems to directly counter just by being who he is.

But it’s not like there’s nothing you can do against Riki. You can track his location using map awareness. You can stick close to teammates when you’re unsure of his whereabouts. You can carry a Gem of True Sight to get the jump on him when he approaches. You can buy Monkey King Bar to stop missing on attacks and actually fight back. You can even reveal him directly using Dust of Appearance. Counter-play versus Riki definitely exists, it’s just that it’s much more obscure, and in some ways counter-intuitive to first-order heuristics. As a result, players at early levels of skill have a perception that Riki is overpowered. And here’s the twist: those people are actually right!

The Meta Spectrum

Wait, what? How can they be right when I just got done outlining all this counter-play to Riki? Don’t those players just need to git gud? To some degree, yes. But the fact that they consider Riki to be overpowered is significant. Their thoughts represent an understanding of the game that exists. It may seem “incorrect” to players who have a more advanced and complete picture of the strategic meta for Dota 2, but that’s a common misconception – the idea that there is only one strategic meta.

When players refer to “the meta” of a given game, they’re usually talking about the one way that the best players are playing at any given moment in time. However, for any sufficiently deep game, there are in fact many metas that exist in parallel. Each one refers to a certain level of understanding of the game. Each level carries with it a set of heuristics, a series of good and bad moves, that make up its individual definition of skill. And every single player of any skill level has one such meta in his brain.

Even single-player games have these meta levels – after all, a meta level simply refers to a successful method of play. A single-player meta might be whether or not it’s a good idea to rob shops in Spelunky, which Binding of Isaac items are the best, or the best time to spend an ultimate spell in Auro. But strategic metas really get put through the wringer in multi-player games, where players constantly compare their metas against each other.

This causes players to refine their metas very quickly and often reach the same conclusions as each other. An experienced player will show a rookie that his meta just won’t cut it in higher level games, and a master will do the same to that experienced player. But just because the master seems to have the most advanced meta doesn’t make his meta the “one true meta,” nor does it invalidate the existence of all of the others. After all, the master had to start somewhere, and what he considers the strategic meta to be now won’t necessarily be his strategic meta forever. When a game is brand new, what may look like the cutting edge meta will one day be the rookie level.

In a sense, players don’t really “evolve” the strategic meta so much as they discover more of it – the many layers of strategy are pre-existing and emergent of the rules. A player of any skill level has the heuristics that he does because he learned it to be true based on his own wins and losses. What works at a certain level or in certain circumstances won’t work in others, but all strategic metas work in some capacity or else the player wouldn’t play that way. Additionally, all players tend to grow and revise their meta understanding over time, and hopefully there’s room at the highest discovered meta for even more growth. So from that perspective, all of the strategic metas at all skill levels are all true, at least for the individual people who hold them. They’re all members of one giant, navigable meta-spectrum.

The reason this is relevant is that the play experience at each meta level is important. Perceived imbalances, even if only perceived, can affect the play experience. A designer’s instinct may be to design solely around the highest level meta, but doing so can, at times, be detrimental. It’s important that high-level play not break down, but it’s important that lower meta levels not break down either, or people might write the game off as degenerate before they see the rest of the meta spectrum. This can happen at any skill level, high or low, and to prevent it, a designer should try to avoid designing at the expense of low-level play, which is just as important as high-level play.

This might be hard for some designers to accept and even harder for players, whose feelings on the meta could be delicately referred to as “intense.” However, if you ask any two of them what needs a buff or a nerf, you’re likely to get very different answers, and in a way, that reinforces my point. Every single player has some set of balance changes in mind based on his meta level, things that from his perspective will improve the game. All of those perspectives deserve at least a little consideration. The high-level meta perspective deserves attention, too, but be careful of over-privileging it – after all, the current strategic meta might be made obsolete by the discovery of a higher layer anyway.

The bad news is that this makes balancing games about a hundred times harder. In a very real and practical sense, you’re taking on the task of balancing many games in parallel. Each meta level of understanding the game is kind of its own game all by itself, and to make matters worse, you have to balance them all using the exact same rulebook. Every problem you try to fix is going to affect other parts of the system – often many other parts. Trying to balance all the meta levels makes the equation so complicated, it’s practically impossible to get the balance 100% right.

The good news is that you don’t have to get the balance 100% right. A deep game system, by its very nature, is going to have micro-imbalances. That’s what allows any decision-making in your game at all – if everything were equally good, there would be no reason to choose one move or strategy over another. So it’s actually not expected that the game be 100% balanced – this gives players room to grow and strategies to learn.

It does, however, require a change in perspective. Balance isn’t as direct as playing whack-a-mole with the buff-stick and the nerf-bat. It’s more nuanced than that – it involves making tradeoffs between all the metas. It means trading a little bit of imbalance at one level for a huge improvement on another. And yes, this does mean that the high-level meta sometimes takes a back seat. Remember: ideally the high-level meta will shift all on its own over time as players discover more depth. It takes courage to shake the assumption that the high-level meta won’t shift without your intervention, but as long as your confidence in your game’s depth is well-founded, your faith will often be rewarded.

Still, it might be hard to imagine what the act of balancing for all metas actually looks like in practice, so for a practical example, let’s return to Riki. If you get players of many skill levels into the same room, Riki becomes a controversial character. New players just get completely crushed by Riki, and they’re likely to flood forums with calls for his nerf. But these cries are laughed off by high- and even medium-skill players. Take a look at this graph from Joe Kelley’s blog. It details the combined pick and ban rate of all Dota 2 champions in a two-year period – in other words, it acts as a general barometer of how desirable these champions actually are in competitive play. Now see if you can find Riki.

See, what I haven’t told you yet is that not only does Riki have counter-play, but that counter-play is so strong that past a certain skill threshold, Riki is hardly played at all. The fact is, Riki thrives on ignorance for his parlor tricks to work, but if people don’t fall for them anymore, he’s got nothing left. He can sort of compete, but if you want an agility carry, there are just better choices than Riki. And that situation has gone on for years.

So this is our situation: Riki is a dominant choice at the entry level, but at the highest levels he evaporates almost entirely. On top of that, in the games where Riki does well, he’s strong to the point of being comical. So in a way, Riki is a trap for new players. He’s not actually that strong, but early on it seems like he might be. Low-skill games are grossly imbalanced in his favor, but past that skill level he quickly becomes basically irrelevant. He adds practically nothing to the meat of the game at the levels where most people play, and in exchange he more or less ruins games at the low level.

This is a bad trade. In fact, it’d be a better situation if Riki weren’t in the game at all. Low-level play would feel more balanced and in line with high-level play, and high-level play wouldn’t really miss him at all. Outside of a few hardcore Riki fans out there, we’ve made Dota 2 a better place overall. You can’t please everyone, and sometimes making the right choice involves making a hard choice.

Balance Isn’t Everything

When asking players for input about games, a huge proportion of that feedback inevitably ends up being about game balance. Every time one of the metas thinks they’ve found a gross imbalance and complains loudly about it, there’s an impulse to fix it, and with as many as can crop up, the temptation is to timidly adjust numbers until the people complaining go away.

Often, however, there are more important considerations than game balance, more important groundwork that needs to be laid – or uprooted – before balance becomes something to worry about. Games should be solution-resistant, but at the same time they need to be at least somewhat intuitive. The learning process should make it at least partially clear how to respond to strong tactics that emerge in the low-meta levels. A player should have a strong direction for improvement as much as possible. Ideally, a game lets players see the depth above their heads instead of perceiving a brick wall they can’t pass.

Even more important is that your game is as interesting to play as possible, and that includes all parties on all sides of any given strategy. That’s ultimately the big problem with Riki in low-level play: playing against Riki is a challenge, but playing as Riki is too rewarding for too little effort. Since you get to be permanently invisible, you can sit around and watch opponents in utter safety, waiting for them to screw up catastrophically and cackling to yourself as you effortlessly collect another kill.

But compare that to the Spy from Team Fortress 2. An invisible Spy doesn’t get to freely wander around the map – his cloak has a short duration. As a result, the decision to enter cloak is every bit as tense as the decision to leave it. In fact, it’s even more tense, since starting up your cloaking run sets the stage for a situation that drips with uncertainty. You don’t know how things will play out after you hit that cloak button, but once you do, you’re committed to it no matter what. The Spy kill is a complicated and nuanced art, but successfully landing one is one of the most gratifying moments TF2 has to offer.

Notice how Riki kills feel like you’re reveling in how unfair it seems, while Spy kills feel like you’re legitimately outplaying your opponent. A Riki player has nothing compelling him to leave stealth until the exact moment he’s ready, and it’s a comparatively automatic and sterile process from start to finish. Meanwhile a Spy player has this looming guillotine over his head, constantly threatening him with “Hurry up and leave stealth, or I’ll do it for you!” Situations are a lot more messy and unpredictable, but they also have a lot of room for creative play and a higher skill cap. The Spy is balanced around a much more dynamic role, where invisibility is a resource rather than a default state. As a result, playing a Spy is exciting and frenetic.

The Spy also feels more fair to the victim. A player who feels the Spy is too powerful can try playing as one, and will immediately realize how difficult it really is to actually set up a Spy kill. Instead of finding what he perceives to be a cheap trick, he gets a sense of the game’s depth of skill. He realizes that getting to the point of reliably making Spy kills is going to take a lot of growth on his part, and in the mean time he should really learn to use all the counter-Spy play that everyone else was using on him! He gets to witness the depth of play on both sides of the coin, and he sees a clear direction toward genuine improvement.

Now, let’s be honest here, there will be people calling for nerfs on all kinds of strategies, characters, or classes almost no matter what. It’s not wrong to want to file off the sharp edges that might cause players to stick to their perception of unfairness. What you might not have considered, however, is that it’s sometimes a matter of redesigning in a better direction instead of trying to make the old one fit. To that end, it’s critical to maximize the possibility space and skill headroom so that as the meta evolves, there’s room to grow instead of a strict paring down of strategy.

See, maybe it doesn’t feel quite right to get rid of Riki entirely. But if we want to keep him, maybe it’s time for something radical, and the ramifications of his permanent invisibility are having a degenerative effect on his play at all levels. So maybe it’s time to ditch the idea as a failed experiment and try something more dynamic and decision heavy, something more like the Spy.

One thing you may notice about that pick/ban chart is that many characters with invisibility see a lot more play than Riki does. The reason is that these other characters have an identity and role outside of just being invisible. Riki’s role is defined by his invisibility, and Dota 2’s dynamics surrounding invisibility cause him to run out of power very quickly. He doesn’t need higher numbers, which would only make the problem worse at low metas. He needs something more interesting, something that widens his possibility space and makes him less one-dimensional. We can always tweak the numbers afterward, but we need to build them on a better foundation.

You can spend all the time in the world trying to fix a pick/ban rate or trying to get perfect balance, but there’s a limit to how much you can improve the game that way. It’s a local maximum that implies polish, but doesn’t have a huge effect on game depth. A better route is to upgrade the depth of the game with better play dynamics. Do your best to increase the density of decisions over time, and try to get each decision to be as thick as you can make it. Make sure the decision tree leads to as much variety as you can possibly muster. Doing so will reward you with a much broader game than merely balancing the game you started with. Maybe this is just me, but between a game that’s perfectly balanced and one that’s more interesting, I’ll take interesting any day.