Tuesday, December 25, 2012

metagame is hard

This installment is dedicated to all things 'metagame'. It is not about 'the current metagame' but rather I will focus on metagame as a concept and how we ought to interact with and make use of it. I intend to begin by debunking various common misconceptions about what metagame is and what sort of things are relevant to say about it - this will show what metagame isn't. In the second section I explore a more positive notion of metagame, making use of analysis of the metagames of other non-DotA games as a reference point. Finally, I will discuss in more detail the relationship between playing a game [particularly DotA] and playing its metagame and offer my own personal advice on how to play a metagame and to what extent to prioritize it relative to its game.

What metagame isn't

I think a natural starting point is to distinguish between 'game' and 'metagame'. While it is trivially true that a game can never be identical to its metagame, this is a truth that is often under-appreciated, misunderstood or entirely ignored. In other words, some analyses of DotA are game analyses and others are metagame analyses. Analysis which is appropriate for the one will, although related, hardly ever be directly appropriate for the other. Unfortunately, within the DotA community, common usage involves an almost systematic conflation of the two terms. You see people saying that

"Lion's ultimate is good in the current meta." 

Under a generous interpretation, this statement might not be problematic, so I find myself enquiring further as to WHY it is that this is allegedly the case.

In response, the following kind of reason is usually given:

"Oh come on, it does such insane damage at level 1",

"40 second cooldown at level 16 - it's like a free dagon 5!"

Unfortunately, these are reasons which one might use to justify the claim that Lion's ultimate is good in the current GAME. Calling this 'metagame analysis' would be incorrect and misleading. The metagame is NOT the game. The metagame is, in a matter of speaking, the 'game of the game'. The problem with the above analysis being used as metagame analysis is not that it lacks context. It would not become metagame analysis if, for example, we added the claim that many heroes in the current version of DotA all have very low hitpoints. This part is a bit less clear because it's almost right. I've used this example deliberately though, to emphasize that for your analysis to be metagame analysis, it requires a level of abstraction which is beyond the level of the game itself. What one could say about a given metagame is that Lion's ultimate is very good in that metagame because it deals a lot of damage at level 1 and people tend to pick heroes with low hitpoints early on, leaving themselves prone to being picked off by big nukes. What's changed? Well, I added a premise which refers to the habits of players. And that, in essence, is where metagame and game come apart. Metagame refers to how the decisions others are likely to make affect the decisions that you plan to make. In what ways do your expectations about your opponents' picks affect your picks? That is metagame. Technically speaking, every single DotA game has its own metagame but zoomed in to this extent we will rarely be able to derive any useful or successfully predictive conclusions. Thus, when people talk about metagame, what is being referred to is normally a more general category of games. The 'current' metagame, then, refers to what decisions teams are expected to most commonly make, in a given context, at a given time. Before I move on I want to clarify one more thing. Many people might infer from the previous few lines that the metagame is something very vague and intangible and thus we can only say very vague and general things about it. While this is partially true, given the level of abstraction that a metagame exists at, it is a mistake to go from this premise to the conclusion that one can readily attach 'in the current metagame' to any piece of analysis about the game. And yet, time and time again, people do this. In fact, I think the reason people do it is because they believe that appending 'in the current meta' to a statement somehow gives the statement more authority - it really doesn't; usually it just makes you look like an idiot. Either you are talking about the metagame and we probably already know that or you're not talking about it and it's incoherent to assert that you are.

What metagame is

Having discussed what metagame isn't and how not to refer to it, it is now time to try to better understand what 
it does consist of. To start with we'll look at a very simple and familiar game: Rock Paper Scissors. In Rock Paper Scissors, there are two players. Each player has one choice to make and both players must make their choices simultaneously. There are three options to choose from, each capable of beating one other option. Is there an ultimate strategy for a single game of Rock Paper Scissors? Is there an argument that can be made for why one ought to pick Paper as opposed to Rock or Scissors, for example? Clearly not. Given zero information about your opponent's choice, you have a 1/3 chance of winning, a 1/3 chance of drawing and a 1/3 chance of losing. What then, does the Rock Paper Scissors metagame look like? Well, lets ask the question - how might our decisions be affected by the decisions we expect our opponents to make? Naturally, if you expect your opponent to choose Rock, you will choose Paper, if you expect her to choose Paper, you will choose Scissors - and so on. But what reason might you have for expecting your opponent to choose one option over the others? As discussed above, there is no reason that the game itself advances in favour of either of them. Perhaps, however, you are familiar with your opponent, and are aware that she has a strong preference for, say, Paper. Well, if that is the case, you have a good reason to expect that Scissors will have a higher chance of winning than usual against this opponent. Of course if your opponent is wise and knows you have information about her preference, she might use her knowledge of your knowledge to affect her decision and actually choose Rock instead to pre-empt your Scissors. This kind of psychological analysis is recursive in this case and if you are playing Rock Paper Scissors you will simply have to choose an arbitrary line to draw at which you stop expecting your opponent to modify their choice based on what they expect your choice to be [This is part of why Pros normally have the power to redefine metagames at will, given the right kind of subsequent success]. So the Metagame of Rock Paper Scissors is a simple case of finding out what your opponents like and trying to counter that - because both players choose at the same time and because both players using this analysis will lead to the winner being decided on an arbitrary basis, this is not a very exciting or useful metagame to think about. Why then, did I bother using this example? Simple - we need to contrast between two ways in which metagame can change. The first way that metagame can change is if the game itself changes, thereby changing which choices are best to make. The second way in which metagame can change is by players changing their preferences. Note, of course, that even in the first case, players will have to change their preferences for actual metagame change to occur - and because of this, the first type of change is actually a subset of the second type. Many would contest that this second type of change is true metagame change, wanting to distinguish between a choice being popular and a choice being expected in a given metagame. However, once you see that the first type is a subset of the second type, it is difficult to understand how we could say that the role of one's raw preferences in influencing one's choices should be excluded from metagame analysis, given that this role is a necessary one in the case of any and all metagame changes - all metagame changes could be accurately framed as changes in preferences.  So, while I would deny the claims that 'whimsical' preference changes aren't relevant to metagame, I will come back to this distinction in the next section as it is a very important distinction [just the wrong type of distinction here] to understand in determining how one interacts with a given metagame.

Click here for a more detailed introduction
 to the concept of metagame in MTG.
Before moving on, I'd like to briefly talk about the metagame of a game whose metagame and analysis thereof is very well established: Magic: The Gathering. MTG is a card-based strategy game. There are different formats of MTG and each develops it's own metagame, naturally. However, the kinds whose metagames are talked about most are those that fall into what are known as 'Constructed' formats. These are formats where players construct their decks in advance of a tournament and then enter with that deck and only have access to the cards in it for the duration of the given tournament. The metagame for MTG, then, is a question of figuring out which decks are likely to be played at a given tournament and then choosing and tweaking your deck accordingly. Similar to DotA, because there are so many different ways for a strategy to be good, the metagame is used to narrow down which types of good are more relevant to a given context. Unlike Rock Paper Scissors, though, MTG's metagame is probably only negligibly affected by baseless or arbitrary preferences. MTG metagame involves things such as 'skews', where a given community might have a regularly 'skewed' or distorted metagame in one or another direction, preferring certain general types of strategies in certain types of competitions. But aside from this, it is a metagame which is ultimately grounded very closely in the game. New sets of cards are released every 3 months and with each new set there is usually some degree of metagame change. Of course the metagame also changes without the game changing, but this is normally because players take a while to fully understand the game and because metagame changes almost inevitably lead to further metagame changes, except in formats where one type of strategy is overpowered and thus unchallengeable [these are normally called mistakes]. What this is meant to emphasize is the relation between game and metagame - in a well designed, well balanced, competitive game, superior understanding of the game will just about guarantee a superior understanding of its metagame.

How to play the DotA 2 metagame

Before offering any kind of prescriptive advice about how to approach DotA metagames, it is important to return to our earlier distinction. We saw that metagames can change either where game changes result in preference changes or when something else results in preference changes. I will call the first type hard metagame changes and the second type soft metagame changes. My normative evaluation, then, is that hard changes should be taken more seriously than soft ones, though soft changes must not be denied their status in the metagame. The latter part of my position has already been explained above but I will briefly motivate the former part here. Metagames cannot be won. Only games can be won. Playing a metagame is always a part of playing a game. Thus, if there are facts about a metagame which are necessary for being successful in the game, those are important to know. However, there are plenty of metagame facts in any given metagame that a very successful team could be entirely ignorant of. A drafter doesn't need to know that his opponent likes to pick X if he is able to deal with the pick in the given draft. A drafter also does not need to know what the most popular picks are if, in fact, the most popular picks are not really the best ones. Soft metagame facts are often of this type - statements of popularity or preference. While soft metagame facts certainly can help a drafter to outpick her opponent, they will often not be necessary for that purpose. A drafter who is ignorant of hard metagame facts, though, will not have to wait long before being punished for it. And what's more, knowing the game well and thus having a good understanding of the 'hard metagame' will allow you to exploit areas of picking which are still very open to definition. If you can find reasons in the game why one pick ought to work better than some other pick which is popularly utilized, you should be testing your idea out - and either finding out what your mistake was, or turning your insight into a competitive edge. My position is actually quite straightforward if you realize that knowing hard metagame facts is mostly reducible to understanding the game. If you don't know your game well, you're not going to pick well. If you don't know what reverse polarity does, or that skewer has been buffed [or now, nerfed], you might misjudge the value of Mag as a pick, for example. Whereas, if you do know your game well and can apply the appropriate analytical tools, a good grasp of hard metagame will automatically follow. Yet, because all metagame changes must be filtered through our subjective preferences, it is very easy and exceptionally common for people to conflate these two types of metagame facts. And, as a result, it is usually only the very best teams who are confident enough to be creative with their picks, recognizing that a given metagame might be partially fixed in only the soft sense.  Let me illustrate the distinction one last time by means of a recent example. 

A few nights ago I'm sitting at my pc watching EternalEnvy co-cast Na'Vi vs Fnatic with Sheever for Starladder. Nearing the midgame, I hear EternalEnvy literally facepalming at the unfortunate reality that Trixi has bought vanguard on Antimage.

"That's not how you play Antimage these days," he remarks, "You buy battlefury to split push and annoy the hell out of your enemies, you aren't meant to go anywhere near them." Now EE has captured very succinctly the way in which AM play evolved in recent years. However, imagine you were Trixi and you heard EE say that the way you are playing the hero is simply "not how you play it". You might be a bit confused and reply, "It is how I play it - can't you see!". Of course, the comment here really means that it's not how one should play Antimage these days. Now, as such, the point probably holds easily for the game that was being casted at the time, given that the Vanguard did look a lot like a desperation buy and not at all the original plan. However, what EE actually did was make a comment about the general case. And yet, we aren't given a reason why people SHOULD play AM in a certain way at the moment. Of course this was in the middle of a cast and it would be unreasonable to expect every comment casters make to include exhaustive and sound justifications. Still, it is important to reflect on the way casters, and players, tend to talk about these things. In this case, are left assuming that the justification is merely a reference to the way in which people popularly do play the hero or have played it recently. But if we want to answer the question of how people should play the hero, we need to do more analysis than this. And in fact, as it happens, people hardly play AM at all at the moment. I would hazard a guess that the main reason it has fallen out of favour is precisely because it evolved into a battlefury-rushing hero and battlefury-rushing, itself, has become a difficult thing to do of late. So picking a hero that needs to rush battlefury is a very risky thing to do in a climate of aggression and stressful lane phases. Ironically, I would say, this might suggest that if you are going to run an AM at the moment, it is probably best that you plan to run a vanguard AM . Now I can't say for sure here whether there is a realisitc place in current picking trends to fit this - both due to lack of space, and because this sort of thing likely needs to be tested -  but there certainly isn't any prima facie reason to expect there not to be. In fact, a carry that can be robust in the early game seems like quite a good thing at the moment. Meanwhile, the battlefury-rushing AM appears to be a very difficult strategy to pull off right now, given the pressures teams tend to face in the early-to-mid game. The point in this example is not to try undermine EE - a player who i have tremendous respect for, especially in terms of the depth of his understanding of the game - but merely to show that even our common ways of talking about metagame-related things are unclear and ambiguous. There is a big difference between facts about metagame that derive directly from the game and those that exist more indirectly, as a product of potentially whimsical preferences - this is essentially the distinction I have made between 'hard' and 'soft' metagame changes. What's more, metagame facts can easily change between the two types. And indeed, in this case, the way the game [and metagame] was at the time that AM was most popular in DotA 2 made it a fact that "AM is currently played by rushing battlefury". However, that was then and this is now. It is difficult to find a game-based reason now for why this would be the case. [Meanwhile, there seems to be a very powerful metagame reason which motivates to the contrary.] And so, if the AM-BFURY relation is still a metagame fact, it's of the soft type. But if this is the case, its entire status as a fact is more dubious and if one can find evidence in the game to argue against it, one can replace or update it with a more suitable metagame fact. Maybe it is still the best way to play AM and AM is just unplayable but, either way, there is no real certainty to be had here. And it is that lack of certainty which makes soft metagame facts less significant and less reliable than hard ones. The distinction is not one of types, but one of intensities. As such, I must reiterate here how foolish it would be to make the distinction, only to wholly dismiss soft metagame facts - it won't remove your chance to win but it will severely reduce it. After all, at the end of the day, most of the time you spend thinking about metagame will probably be in order to prepare counter-picks and if this is your sole focus, it doesn't seem to matter much why your opponents pick the way they do - just that they do. 

So then, what's my advice? The ideal DotA player, I think, would not need to spend any time thinking about metagame. If you understand the game perfectly, you should be able to make the best game decisions without needing to ever abstract your analysis to the level of metagame. What's more, because most of DotA's metagaming occurs during the drafting stage, and the drafting stage is turn based, you actually don't need to predict all that much about your opponents' strategies. You are given insight into their picks while you still have time to make yours. This not only greatly reduces the amount of guesswork involved, but it totally eliminates the possibility of recursive 'Rock Paper Scissors' analyses occurring. Of course, significantly, there is serious time pressure during a draft but the fact remains that the ideal drafter will be able to perfect 3-4 of their picks purely based on information given to them during the draft. But there is no ideal DotA player, no ideal drafter. And because of this it would be absurd to advise any real DotA players to ignore the metagame. Rather, I would encourage players to spend most of their reflection thinking about the game, and thus, the hard metagame. The soft metagame should be considered as well but should be analysed much more critically, with a very open mind. Innovation on this front is one of the more likely places for a team to gain an advantage over a stronger opponent.

Above, when I used the word 'most', I alluded to the fact that not all metagaming in DotA occurs during the drafting stage. This might actually be seen as an understatement if we go back to our original definition of 'metagame'. If metagame is the way in which decisions you expect your opponents to make affects the decisions you plan to make, surely DotA players make most of their decisions after the drafting phase? But this is a trap. Precisely what makes a pick a good pick is what you expect that pick to do in the game - which decisions you expect it to enable you to make and, thus, which you expect you will make. However, DotA is an extremely complex game and, as such, it is not always possible to accurately predict the shape that a game will progress into. Because of this, it is important to use metagaming as a shortcut in balancing out the strengths of your strategy. By this I mean simply that over and above looking at your picks themselves, you ought to be thinking about what sorts of things you might expect your opponents to do in the given game. Probably the hardest metagame fact about DotA 2 at the moment is that teams will make several attempts at shifting the balance of the game by using Smoke of Deceit. Softer metagame facts of this kind include prima facie rules about which types of heroes can solo the off-lane, how likely it is for teams to use dual lanes, etc. Because the game is very difficult and because there are no drafters who are even near to flawless, it is important to condense these sorts of metagame observations into something of a framework to evaluate your picks against. You never want to be thinking about the metagame so much that it reduces your ability to think about the game but inasmuch as they can be complimentary, this is a good thing.

A final caveat

Much has been said about the distinction between two types of metagame facts. But a perceptive reader might remark that both soft metagame facts and hard metagame facts can be ignored by anyone at anytime, rendering them ultimately equivalent in status. After all, game dominates metagame, doesn't it? The difference between the two, in light of this line of thinking, is that ignoring hard metagame facts will likely lead to poor results whereas ignoring soft metagame facts need not do so, assuming that your strategy in a given game is a sufficiently competent one. However, the final caveat that I'd like to add here is that it is often extremely difficult to tell these two types of facts apart. While in practice, many theory-crafters easily distinguish between what they call 'trends' and 'actual metagame', there are also plenty of hard cases to consider. Notice, for example, that Nyx and Kotl are both staple picks in the current DotA 2 metagame. Nyx has recently been described as the best pick there is at the moment by xiao8 while Kotl is picked or banned in almost every Chinese tournament game right now. Meanwhile, Visage has started to rear its head as a potentially lethal tri-laner. What do these three heroes have in common? These are the three that were added just before TI2. To what extent did these three heroes affect TI2? Negligibly. Kotl is the only one that was picked often and camp-stacking was not yet as popular as it is now, meaning the hero was doing a lot less than it could. Nyx had a small cameo in the grand finals, but was probably more of a cost than a benefit in that game. Have any significant changes been made to any of these heroes since TI2? Not really. We have seen minor tweaks here and there - largely buffs, but none I would say make the difference between "totally unpickable" and "first pickable"
. So how do we account for this? On the surface it would appear that their rise in popularity is owed mostly to soft metagame changes. But concluding this would be a mistake, I think. For example, Nyx's carapace is the reason the hero is so potent at the moment - several professional players have attested to this fact in recent weeks as well. But if game changes are not to blame, and preference changes are not to blame either, what is going on? Here we have a hybrid of the two, I think, and one which is very common in the DotA world. Icefrog brings new changes to DotA, and there is a delay before those changes are properly received by the players. There are several reasons for this to happen. For one, players sometimes need time to adapt their thinking and their picks to fit new changes. Meanwhile, some game changes only become relevant when combined with other, later, game changes. And then there are simply changes which players undervalue, or catch onto very slowly. Darkseer, for example, was an insanely strong pick in competitive DotA for several years now. However, it only became very popular when its Scepter was buffed to create images of allies. This buff was clearly over the top and was quick to be reversed. But when the buff got reversed, the hero kept being picked. Why? Because it was a very strong hero in the first place. We have subsequently several nerfs to Vacuum, which was likely the culprit in the first place, and the hero is still seeing plenty of popularity. Sometimes, it seems, a hero has to go over the top in order to be recognized for its already exceptional power.

What do we take out of this? Well, serious DotA players are continuously getting better at DotA. Sometimes it is difficult to notice because a lot of this improvement takes place in the form of masses of new content being internalized, re-arranged, and analysed. But we must not mistake metagame changes resulting from players' improvement for soft metagame changes. These are merely hard metagame changes whose game changes do not appear immediately before them. This does not affect the type of change they are - indeed, if the game is such that something works then the game is such that something works. The time at which it is discovered to work is entirely arbitrary. This should hopefully make us even more sensitive to the metagame type distinction, and thus even more careful in analyzing this complex and slippery beast.

So here's the TL:DR version of it:

1. Spending time thinking about the game is more important than spending time thinking about the metagame.
2. However, because we are merely lowly humans, to think only about the game would require too much of us. So thinking about the metagame is also important, albeit less important.
3. Metagame analysis should be especially sensitive to the distinction between 'soft' and 'hard' metagame changes and facts. Make sure not to confuse the two with one another - it can be very difficult at times.
4. Hard metagame facts, by their very nature, should not be ignored. They must form the foundation for your thinking when deciding how to strategize and play. Soft metagame facts are flexible and must be approached critically.
5. Only think about metagame inasmuch as it does not reduce your ability to think about the game itself.