First I collected all matchup data on 58 different Standard archetypes - everything from Hand Druid to Turret Warrior. For the most part I used HSReplay.net, but when applicable I used Vicious Syndicate Legend-only or Rank 4-Legend data. Where HSReplay didn't have matchup data, I didn't count that matchup during the simulation.
Then I simulated 1000 meta cycles. In each meta cycle, first all archetypes are recalculated using matchup data and the number of players playing each deck. Then, players leave decks with winrates below 50% (the further below 50%, the greater the percentage who leave) and pick up decks with winrates above 50% (the further above 50%, the greater the percentage of those players gained). These new population figures are carried forward to the next cycle.
THE TIMELINE
First comes what you'd expect if you've viewed HSReplay or VSyndicate recently - the (further) rise of Even Paladin, as well as Cubelock. Even Paladin is the best deck you can be playing as I write this. Even Paladin saturation causes Quest Rogue numbers to decline.
Second comes the rise of Control Priest, a deck with the unique characteristics of preying upon both Even Paladin and Cubelock. As Control Priest edges out Even Paladin, Quest Rogue reappears to punish both Control Priest and Cubelock.
Third comes the meteoric rise of Odd Rogue and a resurgence of Spiteful Druid, as a result of Even Paladin getting suppressed and Quest Rogue gaining popularity. While the meta eventually corrects Spiteful Druid back to reasonable numbers, Odd Rogue defies counters and joins Cubelock in Tier One.
Fourth we have a stabilization phase. Warrior gains some popularity as it preys upon Odd Rogue, and Keleseth Tempo Rogue gains a viability it didn't previously have.
TIER DEFINITIONS
Tier One: Meta-defining. Played in obnoxious numbers; defies counters. Possible/worthy nerf targets. Craft at least one of these decks first.
Tier Two: Meta-defining, but not consistently played in obnoxious numbers. Can be the best deck at a particular time, but a meta correction will drop population down to reasonable numbers sooner or later. Maybe worth crafting if the moment is ripe, but Tier 1 is more important generally. Not deserving of nerfs but sometimes get them anyway.
Tier Three: Viable but lacking any capacity to dominate the meta, even temporarily. Not worth crafting, but playable if you happen to have the cards anyway. Won't be (directly) nerfed.
Tier Four: These decks aren't viable and shouldn't be played for rank-climbing purposes. However, they're close enough to viable that they're decent candidates for experimentation; a tweaked version might be viable. Only dust if desperate.
Tier Five: Garbage. Dust these cards.
STAGNATED META BREAKDOWN
TIER ONE
1. Cubelock. I know we all love a good underdog story, but in this tale Cubelock never gets what it deserves; if anything, it becomes more dominant. I guess the more things change, the more some things stay the same.
2. Odd Rogue. Like a proper opportunist, Odd Rogue will find its way to the top by taking advantage of the chaos caused by metagame struggles it is not directly involved in. Overall, Rogue will be played more than Warlock.
TIER TWO
3. Quest Rogue. The definitive anti-control combo deck, Quest Rogue is a check on the meta; when there's too much control it's a winner, when there's too much aggro it's a loser. Because that balance will always be somewhat close in the upcoming meta, Quest Rogue will always maintain relevance.
4. Spiteful Druid. Well, look on the bright side: at least the least skill-intensive highroll deck in Standard isn't going to be Tier 1.
5. Odd Warrior (no Quest). Perhaps the biggest surprise will be the resurgence of Warrior in order to contain the rise of Odd Rogue. Of the various Warrior archetypes, Odd without Quest is in the best position to make this happen.
6. Control Priest. This deck's ability to supress Even Paladin while holding its own against the rest of the meta causes it to play a key role in the timeline. Although it will be the best deck at one point in time, it won't hold it too long and will return to mundane viability.
7. Even Paladin. The current best deck, but not for too much longer. Eventually Warrior and Paladin will have about the same metagame presence.
TIER THREE
8. Murloc Paladin. Despite the hype in Vicious Syndicate Data Reaper #88 that say Murlocadin is the winningest deck at Legend last week, Murloc decks are always at least partially countered by meta shifts - no Crabs needed.
9. Quest Warrior (no Baku). Because Saronite Chain Gang is very good in Quest Warrior, or something.
10. Odd Paladin. Overhyped? Yes. Still viable? Also yes.
11. Keleseth Tempo Rogue. Technically, this deck is not currently viable. However, it will be in the not-too-distant future.
TIER FOUR
12. Dude Paladin (no Baku, no Genn). Not much is known about this rarely-played archetype; we know a few of its strengths but not a lot about its weaknesses. That's why when I first ran the simulation and Dude Paladin dominated the meta, I deleted Dude Paladin from the roster then ran the simulation again. Still, if further investigation reveals few weaknesses... This is where experimentation should be focused.
13. Taunt Druid. During 16% of simulation cycles, Hadronox enjoyed a positive winrate. Problem was the other 84% of the time. A little tweaking could possibly make it viable.
14. Tempo Mage. Kripparrian can thank the meta, because unless something changes it will eventually render Tempo Mage non-viable.
15. Secret Paladin. Yeah, I don't get it either, but that's what the simulation tells me.
16. Even Shaman. The only Shaman deck to enjoy any cycles of a positive winrate during the simulation.
Big spell mage, Combo priest and Control lock tier 5? Ahem, suuure you have numbers and all, but these decks will beat up Odd rogue and most of tier 2 (while being wake to Cubelock, yeah) soo maybe your method is incorrect in some way, i dunno
Big spell mage, Combo priest and Control lock tier 5? Ahem, suuure you have numbers and all, but these decks will beat up Odd rogue and most of tier 2 (while being wake to Cubelock, yeah) soo maybe your method is incorrect in some way, i dunno
Big Spell Mage is favored 56.88% against Odd Rogue but 40.92% against Cubelock; I wouldn't even need to list more decks, more Cubelocks than Odd Rogues. However, Big Spell Mage has a pathetic 17. 52% against Quest Rogue and a fairly even but disfavored 47.4% against Spiteful Druid. Not only will it not be viable, it isn't now; winrate currently below 50%.
Control Warlock is 53.4% against Odd Rogue; again, slighly favored. However, it's 42.13% against Cubelock; again, the gain against Odd Rogue is more than lost against Cube alone. Again, the deck you name is stomped by Quest Rogue (24.81%) and is slightly disfavored against Spiteful Druid (45.31%). Again, it doesn't even have a positive winrate currently.
Combo Priest is 32.6% against Odd Rogue and 46.3% against Cubelock. Lul.
All those decks are bad in the future meta because they're bad against the top 4 decks of the future meta.
I cant argue with numbers. (But i will try anyway, lul)
Exactly which decks are we talking about? Most popular by the stats or with highest winrate? Control Warlock, for example have around 20 cards which are "core" to the deck, which leaves many cards for tech slots for current meta. Assuming your simulation is right, and talking about tier 1-2 meta, we have -
"instaloss" against Q rogue. This is extremly polarising deck, and in meta with too many of these fuckers we dont play Controllock, so i am assuming we have enough aggro to prey on, which brings us to -
Odd rogue, Pally and Spiteful druid as aggro, which are easily targetable by control deck.
40-60 with Cubelock is managable, unfavoured, but can be outskilled, i ll take it
Odd warrior and Control Priest lose to me in fatigue (also Rin) and slow enough not to kill me before that.
So unless meta is overflowed with Q rogue this archetype should have pretty decent winrate. I mean, if you have 10% winrate against Q rogue and about 55% against everything else and Q rogue is about 5% of the playerbase you would still win more than you lose.
I am also especially curious about combo Priest, as most popular list on hsreplay is garbage and wants to convince me control decks are not a thing, and highest winrate one rocking such great cards like Chief Inspector, Nightscale Matriarch and Wyrmguard, but has earlygame of a potato, which is an interesting decision, but not very smart winrate-wise. (I am not even talking about how Shadow Madness NOT in every list for some amusing reason)
Big spell Mage not viable? That proves data isn’t always right. I got legend today for the first time using only Big spell mage. The beauty of hearthstone are tech cards imo. That’s why my version IS viable. At least for me.
On the other I think hand bis spell mage is not easy to play. That’s why people do a lot of mistakes.. and doing mistakes with a deck like big spell mage isn’t forgiving at all. It makes you lose... might cause some bad winrates...
your data might be right but you shouldn’t judge only by data.
If you're MacGuyver you can disarm a time bomb with a paper clip. If you're Jiren you can block an attack by staring at it. If you're Kibler you can win games with Pickpocket Rogue. Skill is quite literally the ability to make stuff that is useless to others and make it useful, but in no way does this make such things any less useless to others; to them, trash is still trash. I'm writing for them.
Congrats on Legend. Just keep in mind that statistics show other players' Big Spell Mage experience differs significantly from yours.
This season I've played (according to my deck tracker) 166 games (score of 93-73). Seing as this is an expansion season, I did play more. I also played with terrible decks throughout the season just to try all the cards I got from my packs (which is why my winrate is a little worse than usual). That being said, I'm curious: according to your data, how will the Elemental Shaman, Rush Warrior and Big Beast Hunter (which I assume is Recruit Hunter in your research) archetypes fare throughout your study?
I'm asking because those are the three decks I played the most and had most success in my climb to my highest rank ever of rank 2 (then again, that might be because of the ladder changes last month, starting at rank 17 or at rank 9 is a lot different) and they ended up being tier 5 according to your data although I can guess at least how Rush Warrior ends up that way (Odd Pally falling off and Cubelock not falling off). And what archetypes would see a surprising amount of light according to your data?
First time post for me on this forum, because this post raised my scientific curiosity.
First of all, congrats to the OP for such a tedious work! From a scientific perspective your paper would have a hard time going threw the review process though, mostly because of issues of reproducibility (methods are not explained clearly enough to be understood by peers). It would have been much better to separate input data (obtained from whatever external source) from the actual code, which itself should ideally been published in whatever language you like (R, python, whatever).
That said, my guess is that you modeled the meta as a typical Rock-Paper-Scissors species competition problem, like biologists are doing for species diversity studies (example from PNAS ). In this paper you will see very similar behavior as your long term meta, which you call "stalled" (see your own figure attached): a periodic behavior, i.e "stalled" but still with regular variability. A typical Rock-Paper-Scissor behavior.
This is very nice (and an indirect "test" that your model does something meaningful), but this is the problem of your model: as others pointed out, it doesn't allow deck evolution (tech cards) and doesn't incorporate the facts that we (humans) do not always react like bacteria: certain decks are (thankfully) still played because they are fun, not because they are best. Or put otherwise: people will never go "all aggro" or "all control", because it corresponds to different play styles, game duration, etc.
Another indication that the meta doesn't work exactly like a Rock-Paper-Scissor system is that empirical evidence never showed such periodic behavior. See the long-term prevalence plot from Vicious Syndicate:
Even after long times without a patch it is hard to find such a periodic behavior.
Altogether, a very nice study but still open to subjective interpretations, like most commentators already mentioned ;-)
(another point: maybe your post would have a less provocative touch without the "Here's the future" subtitle, which implies that you think that your model is perfect. However, we know that all models are wrong, and only some are useful)
First time post for me on this forum, because this post raised my scientific curiosity.
First of all, congrats to the OP for such a tedious work! From a scientific perspective your paper would have a hard time going threw the review process though, mostly because of issues of reproducibility (methods are not explained clearly enough to be understood by peers). It would have been much better to separate input data (obtained from whatever external source) from the actual code, which itself should ideally been published in whatever language you like (R, python, whatever).
That said, my guess is that you modeled the meta as a typical Rock-Paper-Scissors species competition problem, like biologists are doing for species diversity studies (example from PNAS ). In this paper you will see very similar behavior as your long term meta, which you call "stalled" (see your own figure attached): a periodic behavior, i.e "stalled" but still with regular variability. A typical Rock-Paper-Scissor behavior.
This is valid criticism. I updated the spreadsheet with a few extra labels, and I'll explain the formulas a bit here. In short, yes, I was going for the species competition thing.
The initial table at the top of the main sheet (Sheet1) is matchup data. Below that table on the same sheet is the actual simulation.
Each line of the simulation section is one cycle. Each line has five sections:
1. Current archetype population. Total population equals 1.
2. Current winrate. This is the sum of the products of matchup winrate and population of the enemy archetype in question. The INDIRECT function is used often so I wouldn't need to write a new formula for every column; the row to draw matchup data from is in the top row.
3. Population loss. For decks below 50% overall winrate only. Equal to (.5-winrate)*currentpopulation.
4. Population gain "shares." For decks above 50% winrate only. Equal to (winrate-.5).
5. Net population change for next cycle. For decks below 50% winrate this Section3*-1; for above 50%, it's ThisDeck'sShares/TotalShares*TotalLosses.
Sheet3 is a values-only paste of only the Tier 3 or better archetypes, simply for charting purposes. Sheet2 is a working version of Sheet3 from before I deleted Dude Paladin.
This is very nice (and an indirect "test" that your model does something meaningful), but this is the problem of your model: as others pointed out, it doesn't allow deck evolution (tech cards) and doesn't incorporate the facts that we (humans) do not always react like bacteria: certain decks are (thankfully) still played because they are fun, not because they are best. Or put otherwise: people will never go "all aggro" or "all control", because it corresponds to different play styles, game duration, etc.
I think tech cards are exaggerated. How many top tier decks have gone through established decklist changes that are responses to ephemeral meta conditions instead of generic refinement? I don't think it's a total nonfactor, but I think people tend to exaggerate the value of their tweaks - and that such tweaks lose more games than they win in the vast majority of cases.
However, as far as the "fun" point, you're absolutely right. Not all participants choose their deck solely on the basis of winrate; if anything those players are a minority. Furthermore, even if the intent is there, they might choose the wrong archetype to achieve that goal. Because of this I won't be too surprised if it takes the meta a long time to advance through my timeline, perhaps to include the timeline never being completed.
That said, I still think this predicts accurately the order in which events will unfold, even if things stop before getting to the end. If a site like TempoStorm pushes a deck as Tier 1 without the matchups to support it, I doubt it'll take hard enough that the natural competition between "species" won't be able to revert the meta back to its previous state, given sufficient time; furthermore, suvh errors embarass them, so they have incentive to avoid repeating their mistakes.
(another point: maybe your post would have a less provocative touch without the "Here's the future" subtitle, which implies that you think that your model is perfect. However, we know that all models are wrong, and only some are useful)
Admittedly I was deliberately going for clickbait there. One doesn't get noticed without trying to draw attention.
I think you are underestimating the effect prevalence has on win rate. Cube lock is probably the strongest deck. However, because it is the most played deck, most other decks on the ladder target it and as a result the stats show other decks end up more successful.
This repeats most metas with only a few notable exceptions where the strongest deck proves to be non-counterable (post kara midrange shaman as an example). Cube lock appears to be counterable so far.
FiAs said before, the "fun factor" is important in the shaping of the meta, but very difficult to model.
But there is also the "budget factor". F2P players will tend to stick to their deck even if the winrate decreases (unless it goes REALLY low), and also tend to prefer crafting "cheaper" decks when crafting a new one (or a deck for which they have most cards, but this is much harder to model). The first part could be modeled by having an "inertia" factor that diminishes each generation loss/gain population by some factor. The second part would be harder but possible, but adding data about "dust cost" for each archetype, and having a growth factor for winning decks that is slower if decks cost more dust. Yes, this would be a lot of work... Just a few ideas if you want to increase accuracy.
Another point that makes your study less accurate is that metas and winrates vary accross different ranks. This can be seen in ViciousSyndicate's analysis, since they display winrates and playrates at various ranks, and we can see that "the meta" is not the same at rank 10 than at legend. Due to lack of data (sadly), you mix winrates data "overall" with "legend-only", which leads to an hybrid meta that might not be accurate. This difference between legend and rank 10 winrates are probably related to differences in skill, and differences in winrates lead to differences in playrates, which means different meta.
The problem is that more data is not (yet) available.
Finally, on the "teching" point. I'd say that, more than tech options, learning to understand a "new" opposing deck will lead to an increase in winrate. This means that if a new underplayed deck appear with a very high winrate, it will become more popular. But then, its winrate will decrease for 2 reasons : - Counter-decks will raise in popularity (like your model predicts) - Its winrate in each matchup will slightly decrease as people learn to play against it, with or without tech options. Yes, this phenomena will not change winrates drastically, but it still changes. Some factor might also be that the deck start being popularised by good/pro players, and get then played by players with less skill and thus lower winrate. This 2nd point is much harder to take into account, since matchup winrates are your input data, and it's pretty difficult to predict how those will evolve with the increased playrate of new decks.
there is also the "budget factor". F2P players will tend to stick to their deck even if the winrate decreases (unless it goes REALLY low), and also tend to prefer crafting "cheaper" decks when crafting a new one (or a deck for which they have most cards, but this is much harder to model). The first part could be modeled by having an "inertia" factor that diminishes each generation loss/gain population by some factor. The second part would be harder but possible, but adding data about "dust cost" for each archetype, and having a growth factor for winning decks that is slower if decks cost more dust.
That's actually a really good suggestion, and it wouldn't be very much work at all: just one extra stat per archetype and changing the gain-shares formula to (winrate-.5)/dustcost. If I do this type of thing again, I'll incorporate dust cost into the modeling.
METHODOLOGY
First I collected all matchup data on 58 different Standard archetypes - everything from Hand Druid to Turret Warrior. For the most part I used HSReplay.net, but when applicable I used Vicious Syndicate Legend-only or Rank 4-Legend data. Where HSReplay didn't have matchup data, I didn't count that matchup during the simulation.
Then I simulated 1000 meta cycles. In each meta cycle, first all archetypes are recalculated using matchup data and the number of players playing each deck. Then, players leave decks with winrates below 50% (the further below 50%, the greater the percentage who leave) and pick up decks with winrates above 50% (the further above 50%, the greater the percentage of those players gained). These new population figures are carried forward to the next cycle.
THE TIMELINE
First comes what you'd expect if you've viewed HSReplay or VSyndicate recently - the (further) rise of Even Paladin, as well as Cubelock. Even Paladin is the best deck you can be playing as I write this. Even Paladin saturation causes Quest Rogue numbers to decline.
Second comes the rise of Control Priest, a deck with the unique characteristics of preying upon both Even Paladin and Cubelock. As Control Priest edges out Even Paladin, Quest Rogue reappears to punish both Control Priest and Cubelock.
Third comes the meteoric rise of Odd Rogue and a resurgence of Spiteful Druid, as a result of Even Paladin getting suppressed and Quest Rogue gaining popularity. While the meta eventually corrects Spiteful Druid back to reasonable numbers, Odd Rogue defies counters and joins Cubelock in Tier One.
Fourth we have a stabilization phase. Warrior gains some popularity as it preys upon Odd Rogue, and Keleseth Tempo Rogue gains a viability it didn't previously have.
TIER DEFINITIONS
Tier One: Meta-defining. Played in obnoxious numbers; defies counters. Possible/worthy nerf targets. Craft at least one of these decks first.
Tier Two: Meta-defining, but not consistently played in obnoxious numbers. Can be the best deck at a particular time, but a meta correction will drop population down to reasonable numbers sooner or later. Maybe worth crafting if the moment is ripe, but Tier 1 is more important generally. Not deserving of nerfs but sometimes get them anyway.
Tier Three: Viable but lacking any capacity to dominate the meta, even temporarily. Not worth crafting, but playable if you happen to have the cards anyway. Won't be (directly) nerfed.
Tier Four: These decks aren't viable and shouldn't be played for rank-climbing purposes. However, they're close enough to viable that they're decent candidates for experimentation; a tweaked version might be viable. Only dust if desperate.
Tier Five: Garbage. Dust these cards.
STAGNATED META BREAKDOWN
TIER ONE
1. Cubelock. I know we all love a good underdog story, but in this tale Cubelock never gets what it deserves; if anything, it becomes more dominant. I guess the more things change, the more some things stay the same.
2. Odd Rogue. Like a proper opportunist, Odd Rogue will find its way to the top by taking advantage of the chaos caused by metagame struggles it is not directly involved in. Overall, Rogue will be played more than Warlock.
TIER TWO
3. Quest Rogue. The definitive anti-control combo deck, Quest Rogue is a check on the meta; when there's too much control it's a winner, when there's too much aggro it's a loser. Because that balance will always be somewhat close in the upcoming meta, Quest Rogue will always maintain relevance.
4. Spiteful Druid. Well, look on the bright side: at least the least skill-intensive highroll deck in Standard isn't going to be Tier 1.
5. Odd Warrior (no Quest). Perhaps the biggest surprise will be the resurgence of Warrior in order to contain the rise of Odd Rogue. Of the various Warrior archetypes, Odd without Quest is in the best position to make this happen.
6. Control Priest. This deck's ability to supress Even Paladin while holding its own against the rest of the meta causes it to play a key role in the timeline. Although it will be the best deck at one point in time, it won't hold it too long and will return to mundane viability.
7. Even Paladin. The current best deck, but not for too much longer. Eventually Warrior and Paladin will have about the same metagame presence.
TIER THREE
8. Murloc Paladin. Despite the hype in Vicious Syndicate Data Reaper #88 that say Murlocadin is the winningest deck at Legend last week, Murloc decks are always at least partially countered by meta shifts - no Crabs needed.
9. Quest Warrior (no Baku). Because Saronite Chain Gang is very good in Quest Warrior, or something.
10. Odd Paladin. Overhyped? Yes. Still viable? Also yes.
11. Keleseth Tempo Rogue. Technically, this deck is not currently viable. However, it will be in the not-too-distant future.
TIER FOUR
12. Dude Paladin (no Baku, no Genn). Not much is known about this rarely-played archetype; we know a few of its strengths but not a lot about its weaknesses. That's why when I first ran the simulation and Dude Paladin dominated the meta, I deleted Dude Paladin from the roster then ran the simulation again. Still, if further investigation reveals few weaknesses... This is where experimentation should be focused.
13. Taunt Druid. During 16% of simulation cycles, Hadronox enjoyed a positive winrate. Problem was the other 84% of the time. A little tweaking could possibly make it viable.
14. Tempo Mage. Kripparrian can thank the meta, because unless something changes it will eventually render Tempo Mage non-viable.
15. Secret Paladin. Yeah, I don't get it either, but that's what the simulation tells me.
16. Even Shaman. The only Shaman deck to enjoy any cycles of a positive winrate during the simulation.
TIER FIVE
Everything else: Big Druid, Hand Druid, Odd Druid, Quest Druid, Recruit Druid, Token Druid, Dragon Hunter, Midrange Hunter, Odd/Face Hunter, Quest Hunter, Recruit Hunter, Spell Hunter, Big Spell Mage, Elemental Mage, Even Mage, Odd Mage, Quest Mage, Aggro Paladin, Dragon Paladin, Quest Paladin, Combo Priest, Deathrattle Priest, Odd Priest, Quest Priest, Resurrect Priest, Spiteful Priest, Kingsbane Rogue, Miracle Rogue, Pickpocket Rogue, Elemental Shaman, Murloc Shaman, Overload Shaman, Shudderwock, Control Warlock, Echolock, Evenlock, Zoolock, Odd Quest Warrior, Pirate Warrior, Rush Warrior, Spiteful Warrior and Turret Warrior.
Link to all number crunching (8.2MB Excel file): https://drive.google.com/file/d/1PTB_5pmRSaqSk6ZIqYHT9xIdEf2o14_m/view?usp=drivesdk
Edit 4/30: Added images, switched rankings of Quest Warrior and Murloc Paladin (initially ordered them wrong).
Is it joke?
Big spell mage, Combo priest and Control lock tier 5? Ahem, suuure you have numbers and all, but these decks will beat up Odd rogue and most of tier 2 (while being wake to Cubelock, yeah) soo maybe your method is incorrect in some way, i dunno
¯\_(ツ)_/¯
I cant argue with numbers. (But i will try anyway, lul)
Exactly which decks are we talking about? Most popular by the stats or with highest winrate? Control Warlock, for example have around 20 cards which are "core" to the deck, which leaves many cards for tech slots for current meta. Assuming your simulation is right, and talking about tier 1-2 meta, we have -
"instaloss" against Q rogue. This is extremly polarising deck, and in meta with too many of these fuckers we dont play Controllock, so i am assuming we have enough aggro to prey on, which brings us to -
Odd rogue, Pally and Spiteful druid as aggro, which are easily targetable by control deck.
40-60 with Cubelock is managable, unfavoured, but can be outskilled, i ll take it
Odd warrior and Control Priest lose to me in fatigue (also Rin) and slow enough not to kill me before that.
So unless meta is overflowed with Q rogue this archetype should have pretty decent winrate. I mean, if you have 10% winrate against Q rogue and about 55% against everything else and Q rogue is about 5% of the playerbase you would still win more than you lose.
I am also especially curious about combo Priest, as most popular list on hsreplay is garbage and wants to convince me control decks are not a thing, and highest winrate one rocking such great cards like Chief Inspector, Nightscale Matriarch and Wyrmguard, but has earlygame of a potato, which is an interesting decision, but not very smart winrate-wise. (I am not even talking about how Shadow Madness NOT in every list for some amusing reason)
P.S.(edit)
¯\_(ツ)_/¯
https://www.youtube.com/watch?v=HQey-2LP1E4
Source : Dude trust me. Kappa
This season I've played (according to my deck tracker) 166 games (score of 93-73). Seing as this is an expansion season, I did play more. I also played with terrible decks throughout the season just to try all the cards I got from my packs (which is why my winrate is a little worse than usual). That being said, I'm curious: according to your data, how will the Elemental Shaman, Rush Warrior and Big Beast Hunter (which I assume is Recruit Hunter in your research) archetypes fare throughout your study?
I'm asking because those are the three decks I played the most and had most success in my climb to my highest rank ever of rank 2 (then again, that might be because of the ladder changes last month, starting at rank 17 or at rank 9 is a lot different) and they ended up being tier 5 according to your data although I can guess at least how Rush Warrior ends up that way (Odd Pally falling off and Cubelock not falling off). And what archetypes would see a surprising amount of light according to your data?
Ranked game wins per class (26/Dec/2018): Druid - 457; Hunter - 482; Mage - 345; Paladin - 435; Priest - 295; Rogue - 234; Shaman - 430; Warlock - 419; Warrior - 367
I really should update my signature more often...
I had a 65% w/r with my Control Rinlock deck in my run to legend this season.
I find it very hard to believe it's a tier 5 deck lmao.
OP have you ever read The Foundation by Isaac Asimov? It's all about predicting the future based on statistics, seems up your alley.
Erm, stats seems fine but not perfect.. dunno what I would fix
First time post for me on this forum, because this post raised my scientific curiosity.
First of all, congrats to the OP for such a tedious work! From a scientific perspective your paper would have a hard time going threw the review process though, mostly because of issues of reproducibility (methods are not explained clearly enough to be understood by peers). It would have been much better to separate input data (obtained from whatever external source) from the actual code, which itself should ideally been published in whatever language you like (R, python, whatever).
That said, my guess is that you modeled the meta as a typical Rock-Paper-Scissors species competition problem, like biologists are doing for species diversity studies (example from PNAS ). In this paper you will see very similar behavior as your long term meta, which you call "stalled" (see your own figure attached): a periodic behavior, i.e "stalled" but still with regular variability. A typical Rock-Paper-Scissor behavior.
This is very nice (and an indirect "test" that your model does something meaningful), but this is the problem of your model: as others pointed out, it doesn't allow deck evolution (tech cards) and doesn't incorporate the facts that we (humans) do not always react like bacteria: certain decks are (thankfully) still played because they are fun, not because they are best. Or put otherwise: people will never go "all aggro" or "all control", because it corresponds to different play styles, game duration, etc.
Another indication that the meta doesn't work exactly like a Rock-Paper-Scissor system is that empirical evidence never showed such periodic behavior. See the long-term prevalence plot from Vicious Syndicate:
Even after long times without a patch it is hard to find such a periodic behavior.
Altogether, a very nice study but still open to subjective interpretations, like most commentators already mentioned ;-)
(another point: maybe your post would have a less provocative touch without the "Here's the future" subtitle, which implies that you think that your model is perfect. However, we know that all models are wrong, and only some are useful)
I think you are underestimating the effect prevalence has on win rate. Cube lock is probably the strongest deck. However, because it is the most played deck, most other decks on the ladder target it and as a result the stats show other decks end up more successful.
This repeats most metas with only a few notable exceptions where the strongest deck proves to be non-counterable (post kara midrange shaman as an example). Cube lock appears to be counterable so far.
FiAs said before, the "fun factor" is important in the shaping of the meta, but very difficult to model.
But there is also the "budget factor". F2P players will tend to stick to their deck even if the winrate decreases (unless it goes REALLY low), and also tend to prefer crafting "cheaper" decks when crafting a new one (or a deck for which they have most cards, but this is much harder to model). The first part could be modeled by having an "inertia" factor that diminishes each generation loss/gain population by some factor. The second part would be harder but possible, but adding data about "dust cost" for each archetype, and having a growth factor for winning decks that is slower if decks cost more dust. Yes, this would be a lot of work... Just a few ideas if you want to increase accuracy.
Another point that makes your study less accurate is that metas and winrates vary accross different ranks. This can be seen in ViciousSyndicate's analysis, since they display winrates and playrates at various ranks, and we can see that "the meta" is not the same at rank 10 than at legend. Due to lack of data (sadly), you mix winrates data "overall" with "legend-only", which leads to an hybrid meta that might not be accurate.
This difference between legend and rank 10 winrates are probably related to differences in skill, and differences in winrates lead to differences in playrates, which means different meta.
The problem is that more data is not (yet) available.
Finally, on the "teching" point. I'd say that, more than tech options, learning to understand a "new" opposing deck will lead to an increase in winrate. This means that if a new underplayed deck appear with a very high winrate, it will become more popular. But then, its winrate will decrease for 2 reasons :
- Counter-decks will raise in popularity (like your model predicts)
- Its winrate in each matchup will slightly decrease as people learn to play against it, with or without tech options. Yes, this phenomena will not change winrates drastically, but it still changes. Some factor might also be that the deck start being popularised by good/pro players, and get then played by players with less skill and thus lower winrate.
This 2nd point is much harder to take into account, since matchup winrates are your input data, and it's pretty difficult to predict how those will evolve with the increased playrate of new decks.