Metagame Balance For eSports & Fighting Games

[@matthewcecil8552]

Your brother playing 5e D&D over at Critical Roll told me to watch this video. Well worth the time. This analysis technique probably has applications far outside the gaming world into that of gambling and investing. Very cool discussion.

[@SunfireGTX25]

I'm here from watching Critical Role and hearing Taliesin mention this video! 2 years late but hey, new views, yay! :P

[@GreekPackDragon]

LOL. I just watched Episode 5 of Campaign 2 live. But I also am listening to the 1st season episode 4 and that is also why I am here.

[@xwingvmanx]

Glad to see I am not alone lol

[@PurpleHonee]

Ditto.

[@portlandpope]

Me too!

[@johngaete2413]

same

[@le_nub6711]

for the question on around minute 24: actually there is, but its very subtle: the sackboy row and column are the key: while not peaking as much cooper and ratchet, its the only row where the lowest number is 5, meaning he will always have an at least neutral matchup and therefore no counter while otherwise still having an edge. and while cooper and ratchet have more more higher number, they both do have also one more thing in common: a single value below 5, and that is vs sackboy. this means he becomes the counter to both characters that are otherwise better than him. this and the lack of counter will fore the meta around him.

[@nomad2659]

Keeper of Arcane Lore Taliesin sent me here.

[@grudley]

The thing to note about slackboy is that he has no losing matchups. I think that in any such situation with one character with no losing matchups, the computation will give 50% as a lower bound on frequency?

[@nodthenbow]

The computation shouldn't give a lower bound of 50 on everyone without a losing matchup. An example is a game where every matchup is 50/50, the computation should give a lower and upper bound of 0 and 100 respectively for every character. They all have no losing matchups still though. As a note, I haven't checked in the actual algorithm he provided, but if it doesn't work then it's due to an error or lack of completeness in the implementation.

[@matejlieskovsky9625]

No losing matchups means that there is no upper bound.

[@MaesterofEvolutions]

14:42 Why would I not play Terran ever in such a chart?? I don't know how your linear program works but I am sure that just picking Terran is much better than picking them 36% of the time and protoss 34% of the time and Zerg 30% of the time. Am I missing something? The only reason I would pick someone else is if Terran had a less than 50% winrate against somebody and I want to counter that pick. Well, it seems to just be an example but the fact that he picked such a bad one makes me wonder...

[@archmagusofevil]

You are right. He used a bad example and if his program is accurate, then it would show Terran at 100% pick rate. The reason they are dominate is not because they have 50+ in every match up, though. What makes them the hands-down best strategy is they have the highest % in every column. That means, no matter what your opponent picked, the faction that does the best against them is Terran.

[@AndrewMabon]

This is very interesting. Looking at "expected population size" for a particular hero/choice/champion reminds me of analysing different ecosystems and seeing which sorts of predators or animals one might see occupying certain niches, and the relative sizes thereof.

[@KusogeMan]

this is way better than i expected

[@Mr_Potatobread]

Fantastic! I'm writing a paper on simulating populations of competitors using different characters for a proxy fighting game for a Grad-School Game Theory course and I was happy to see alot of what I am working on reflected here! Slight note to anyone reading this: The Nash equilibria calculated for the Starcraft example is wrong - you can see from the table that the best response for all 3 of your opponent's options is always to play terran, so the Nash Eq. is where both players always pick Terran.

[@Kakerate2]

13:40 anyone have a link for this talk?

[@avok23]

The tier list made up by adding match-up scores was surprisingly accurate for SF4

[@MajkaSrajka]

If you have game with most of the characters being viable across the board this will happen, but if you have crappy character that no one will play, having 10 or having 0 in this category doesn't matter. This rarely happens "in wild" because most of the times crappy characters are simply crappy, and having crappy character good against one of the meta picks is hard to create using most of the game mechanic systems.

[@transformerssurprise7563]

Great talk. I’m designing a game right now and this gives me some great insight on how to move forward. Thanks.

[@timdoesleveldesign]

I wonder how this is applicable in a game like CS:GO where the choices the players make (gun purchases) are offset by a price. So there's obviously overpowered "characters" but this is affected by an overlaid additional metagame that impacts the options for selecting other "characters" in the next rounds.. hmm..

[@denstore1337]

+Tim Spaninks excpected kills in one round/cost, would be my first solution

[@timdoesleveldesign]

+Saltysausage underscore Mm, but there's not really such a simplistic "rock paper scissors" model, as the strength versus another weapon depends on the location and situation of the encounter. But I suppose you could make a simplistic model using kill/death ratio (rather than simply kills) per weapon and offset it by price. But that still doesn't really cut it. If you'd have a weapon that's, let's say, 3x as expensive, but 2x as good, this would put it below the power curve with this formula, even though it would be heavily overpowered, as the price vs power is not a linear thing to consider balance. I think this would require some pretty serious calculations :'D

[@MyAce8]

theres to factors that he didn't address like the fact that some characters are op but have low win rates because they are hard to play, but everyone is playing them regardless of how capable they are at using that character. also there are some characters that are popular but not necessarily top tier like cpt. falcon in melee

[@Hyperlooper]

+Ace shinigami Most of the source data he used defined the player skill level, so I think those assumptions are sort of 'baked in'.

[@zorro727]

If they are hard to play, they aren't Op, they are balanced...

[@joaomarcoscosta4647]

@@zorro727If they are hard to play, but have a 70%+ win rate among top players and completely dominate tournaments, they are OP. Those characters might not be a big issue for average players, but they could ruin a game's viability as an E-Sport if not banned. And make online matches for high ranked players pretty boring. Look at LoL's Akali example, at 8:00 in this GMTK video: kzbin.info/www/bejne/jYm0q5eOhbmVaKM

[@joaomarcoscosta4647]

@@Hyperlooper Not entirely. For one thing, those assumptions would certainly not be "baked in" the average player charts, only the high-ranking ones. But also, even among high-ranking players, learning a new character takes time. If people find out a new tech or start undertanding why a certain character is overpowered, the influx of new players means that a higher percentage of people playing a certain character will not be proficient with that specific character. At least during the first couple of weeks. On the other hand, high-ranking players who stick to a character considered low-tier are more likely to be overperformers. What I'm saying is... Matchup charts and win-rates work under the assumption that both players are equally skilled, but that assumption is somewhat dubious. (Since a player who got "carried" into a high-ranking by picking a character that is both strong and easy to use might have, overall, a similar win-rate as an over-achiever who sticks to a "Low-tier" character due to personal preference.) So... Yeah.

[@CameronOwen101]

I wonder if this can be a useful way to balance party roles in team PvP or even PvE games by treating the team composition as the "fighter" and searching for states where the team with the most variety reaches some required minimum but doesnt become totally dominant.

[@tyblazitar]

this video was so beautiful it made me cry

[@DylanJonesGame]

one of the best talks of all time

[@bobsmith-ov3kn]

4:04 He completely misses the point of the discrepancy between reading a character's win rates versus player perception of strength. YES, it's because those characters are PLAYED so much more, but he seems to think that is the entire answer and doesn't reason any further. The answer is of course that those characters are played so much more BECAUSE THEY ARE KNOWN TO BE OVERPOWERED, so a huge chunk of players are going to use them, and more likely than not the majority of the type of players who gravitate to the overpowered characters AREN'T VERY GOOD so they will HUGELY dilute the win count of someone at the company looking at metadata of millions of games.

[@scozzbaggs9224]

yea exactly what is this guy talking aboutttt

[@leonardoraele]

A solution for that is to look at the data of highly skilled players instead of all players.

[@GhostOfSparta305]

@@leonardoraele It's a good thing, then, that this is exactly what he did. If you download his slides and view his notes, he very clearly states that his All-Stars data used for his charts were restricted to medium-high skilled players, including the win-rate chart.

[@zorro727]

No, because the hivemind of "communities", id they pick one character, almost everyone in this "community" gonna do the same, not because they like or because is strong, because are told to them some opinion as it was the true

[@101graffhead]

@@leonardoraele that's how respected and standard tierlists are made usually in fighting games in japan ratio tournaments happen in which grand masters of each character are invited and than stats are computed. I don't know about latest fighting games but this thing did happen in super turbo

[@Rofel01]

Amazing Talk! Thanks for sharing Knwoledge that is still viable nowdays!

[@onione9138]

Thank you for posting this is really good.

[@lunar_lotus4841]

Alex, your brother Tal. sent me here via CR Campaign 1 episode 4. 👍😃❤

[@MaesterofEvolutions]

Has anybody tried to use the code he mentions at 25:18? I am not a game designer but I want to analyze a game right now and I do not know how to use it. I would appreciate it if someone could help me out here. ^^

[@ianmcginn3094]

If you're still wanting to analyze that game, the code is a little painful to set up if you're unfamiliar with it, but it does work. The only issue is that when you put too many characters in, the names start to overlap and it becomes hard to read.

[@MaesterofEvolutions]

@@ianmcginn3094 Thx. I had a friend who is more familiar with this than me do it. It was indeed as you say, the names overlapped as I took a huge table of contents xD

[@luqman_azeem]

The concept of "metagame bounds" is something that Economy designers are already used to. They may be worth tagging along with, to get some help on this type of multiplayer problem.

[@paladinIV]

In the computation of the Bayes Nash equilibrium, I believe the optimal strategy (for both players) is always to play Terran (given the values). You should not randomize.

[@paladinIV]

Correction: it's Nash equilibrium not Bayes Nash. But the rest of the comment stands.

[@Blackwindzero]

Anyone know the hour long game theory talk he's suggesting we watch?

[@Hagb3rd]

I think it's this guy, dunno which video: kzbin.info/www/bejne/f3esi2iCjtmEl6s

[@madichelp0]

Interesting talk, but I question it's usefulness. If we assume that new players on average play the character that wins them most games, that still doesn't have any meaning on the mid or high level play. Because the character that wins low level play can have tools that are completely countered by good players. In SF4, Vega has many special moves that are difficult for new players to counter, but are trivial for a pro to deal with rendering him low tier. Seth on the other hand is the opposite, lots of difficult combos and low health, but once you master them becomes incredibly strong. I think if there is an algorithm that can predict character use then it must incorporate player perception somehow.

[@igotdembombs]

+madichelp0 A lot of the time he was specifically talking about high skill players. When he referred to new players, he was talking about new players to the community. You can sort of assume these players are trying to take the game seriously. What's more is that you can consider a top player changing his character in practically the same way as he was treating new players to the community. But yes you are right, tiers work in different ways in different skill sets. But I imagine you can analyze them in the same way except constraining for your skill set. I.e. only calculate the win rate for players between 1,000 and 5,000 points.

[@nerusskyhigh]

The githib link doesn't work anymore. Do somebody know where I can find something similar?

[@matejlieskovsky9625]

The link works for me github.com/Blinkity/metagame

[@nerusskyhigh]

@@matejlieskovsky9625 I didn't manage to access it. Maybe it was a typo? Anyway, thank you.

[@codyglenn7231]

I was told to come here by Taliesin Jaffe, so voila, I'm here.

[@firstnamelastname6699]

shouldnt that SC example mean always play Terran?

[@joeschmoe4376]

>samus at potentially 4th most played character el fuego is w o k e

[@moncala7787]

Joe Schmoe that’s not what the data says. It says if a community of players doesn’t play samus at least X% of the time then a Samus player can exploit that community’s vulnerability to Samus and have an over inflated win rate. Samus essentially it points to Samus being a unique enough character that players who have no experience fighting her will struggle with. Which is supported by the data with players like Leffen having a Samus sized achilles heel of sorts. The reason being his region/play group doesn’t meet the minimum number of Samuses

[@alphacore4332]

tfw metaknight is the meta

[@rapid3384]

Can't for the life of me get Jaffe's code to work, and I'm basically illiterate when it comes to coding in Python. Having some real difficulties making use of the metagame bounds code Has anyone else tried using it and had any problems?

[@danieldosso2455]

Players give great feedback, but poor solutions to their feedback

[@TheIronicRaven]

If you are interested in a slightly longer talk that goes into game balance more, check out these guys: kzbin.info/www/bejne/hImnaYyHbM-Sh80 They really go into depth on these subjects

[@Freefork]

1:10 What's Smash?

[@electrolyteblend]

I see this is before asma won a major tournament

[@Megalobytes]

Tal sent me here. I approve.

[@joao20able]

Seems cool, let's see this.

[@WanderingNasi]

Major flaw in having a perfectly balanced game; if the goal is for every character to have a 50% win rate, what is stopping players from playing the easiest characters? A character with a high skill cap and one with a low skill cap will theoretically have the same win rate. If you know both characters have the same potential, but one is easier to play, why wouldn't you pick the easier one? The players who choose to play the harder character are going to feel cheated since they have to work harder for the same results. While the players who pick the easier characters will be mocked for doing so. And in terms of esports, from a spectator's perspective, the easier character is almost always more boring to watch. Easy characters typically have a one dimensional strategy with little to no outplay potential, whereas the hard character are typically the inverse. Players/fans are quick to realize this and it is considered "cheap" in many gaming communities. Look at Bayonetta in smash 4. People who mained Bayo and won tournaments, were often belittled by the community and even the shoutcasters for being "Bayo-abusers." The problem is that gamers want to win and if the optimal strategy is to play an easy or over powered character, more players will play that character. Win rate should be a reflection of skill. A difficult to master character should have a higher potential. Player skill should be rewarded. Not to say that a high skill cap character should always win, but the outcome should be in their favor of the the player utilizes the character to its full potential. Not based off of their character selection. This is the most commonly neglected factor in most esports. Players should win or lose depending on the level of their execution of character/game mechanics. We aren't robots. Characters will rarely be played at their absolute potential, but the winner should reflect the better player. I often think of the what is probably the most iconic moment in modern esports. Faker and Ryu's Zed 1v1. The most hype, insane, showcase of player skill. I have personally watched and played around ten thousand hours of League of Legends and not a single moment has even come close to comparing to this moment. Go back and watch the play. It was a thing of beauty. The APM in that moment was astonishing. That moment alone, made me want to play Zed. Those are the moments that esports were made for. A play that's akin to watching Michael Jordan dunk from the free-throw line for the first time. It leaves your jaw hanging on the floor. Faker made a name for himself in literally one play. Bringing life to a new era. The era of the "Demon King." If we want esports to continue growing, especially in the MOBA genre, developers need to bring back those moments. Allow all-star players to shine and make these insane plays. Macro play is interesting in a sense, but there's a reason why basketball is more popular than chess.

[@archmagusofevil]

You would pick the harder one because a 50% average win rate across all matches doesn't mean a 50% win rate in every match. Think about it in terms of rock paper scissors. The game is perfectly balanced and every option has a 50% average win rate across all games. If I choose to only pick paper because i find that to be an easier gesture, I'm not going to win 50% of all of my games because people will notice that I'm not good with the "harder" choices and my personal play style can be hard countered with someone really good at making a scissors gesture.

[@monstereugene]

Yoshi has now won a major so there’s no way to do this correctly because if you buffed him he’d probably be as dominate as fox 💀

[@awxangel6781]

"There's nothing obviously pointing to Sackboy being too good" except being the only character with only neutral or positive matchups.

[@crustman5982]

Nobody can beat my crumpet dash in shrek super slam

[@ThePharphis]

No items, Fox only, Final Destination

[@yokokuramaful]

3:00 , this is really worrying fallacious logic he's using. Winrates, especially when collected in gross from online play don't really demonstrate character power, they demonstrate ease of accessing power, a medium-power character that anyone can play optimally will have a waaaayyyyy higher winrate than a strong character with a lot of skill and complexity to them. I'm not necessarily arguing IGN was right, just that the reasoning he uses to dismiss them casts doubt on his authority to speak to game balance at all.

[@silinsar5804]

I really felt the urge to clap in front of my screen after watching this.

[@fcsart6616]

>character balance *shows PlayStation Allstars*

[@joeschmoe4376]

21:59 Zain has entered the chat

[@Pan_Z]

This is stupid. If players are enjoying characters not because they're overpowered, but because they find them fun to play as, nerfing them is only going to piss the player base off. The goal of patches should be to make a more fun game for everyone, not perfectly balance the pickrates of every character. The 2 usually go hand-in-hand, but in this case they don't. People wanted to play as Kratos because they liked God of War, tier status be damned.

[@SpaceMcDarky]

Well done. Critical Role sent me here.

[@geoffreymiller4838]

Banging my head on the keyboard about using win rate to determine strength. When everyone sees that Kratos is overpowered, they play him expecting free wins without understanding why he's so strong, and first order optimal strategies from characters that may well be weak at the highest level of play will still dominate at that level.

[@jamdeluxe7456]

smash bros? more like shrek super slam knock-off

[@crustman5982]

Jam Deluxe I bet you main red riding hood

[@woenilyks]

Obviously win rates can be used to draw conclusions about game balance, and your chart is interesting to look at, but your basis is flawed in the context of e-sports. The goal of a bracketed tournament is not to achieve some high average win rate, it is have a near 100% win rate (dock some for the existence of losers bracket.) Take a hypothetical fighting game, for example, where one character, A, has a 99% win rate against the entire cast, except for against character B, whom character A has an impossible match up with a 0% win rate. Assuming character B is a solid character with competitive win rates against the rest of the cast, you'd still find that character A would have some staggeringly high win rate. Clearly this character is grossly overpowered. In reality, however, character A would be trash and never played in tournaments because, even though you'll win against most characters, you are guaranteed not to advance if you ever encounter character B, which is likely to occur at some point over the course of the tournament. If you were to plug that into your chart, however, it would display a result showing that character A should be played a lot in a "healthy" community, defined by people wanting to achieve high average win rates. Is character A broken? Clearly. Is he overpowered in the context of e-sports? No, he's weak. This is why strategies in matchmaking systems like you find in Starcraft, League, and other game can sometimes be deemed overpowered in that setting while being very weak in tournament settings. A relevant example would be cannon rushing in Starcraft, which has a very high win-rate against unwary players but falls to a near 0% win rate against players who expect it. Doing this strategy every game is very effective in matchmaking, where the goal is the have a high average win rate (over 50% = increase in ELO score) but is terrible in a tournament setting. So while your chart may be relevant to anonymous matchmaking systems where the goal is to achieve average win rates, it would only have a very narrow range of application: only for certain types of games where a player's strategy is forced to be static and certainly not for e-sports, tournaments, or non-anonymous communities. After all, on average the Rocky Mountains is a flat stretch of land.

[@julianfuchs9253]

Tom Marzullo i think it would be interresting to make a game with an algo which sorts the chars.... if char a is played 90perc in a hour he gets a buff in the next hour and so on... fully automatical that means u will have a relatively balanced gameplay cause strong chars get played more often if other players think they are broken... I really wonder if this would work in a brawler Greetings

[@leonardoraele]

Tom Marzullo The graph actually accounts for that. That's why Sackboy shows as the better character in the game and not SkyCooper. If you loot the chart at 24:00, you can see your very example in practice (although with more realistic numbers). SkyCooper is a strong character with a very high win rate against EVERY character, except Sackboy. Sackboy, in the other hand, has more balanced win rates, but is pointed as the dominant character in the graph because he is the only counter to SkyCooper.

[@VideoGamesMania]

They see RANKED statistics, not championship statistics as winrate/pick rate.

[@zorro727]

9 Communities are cancer

[@CreepyCrow6]

Can you make a game that avoid metagame?

[@GrassSquid]

This is extremely easy to do, just make a game where the player can only choose one character so everybody is forced to pick that character. However, this would obviously also be a very boring game.

[@angryvibes8611]

theres nothing random about any of it at all

[@sergisubirats7791]

This is innacurate at so many levels. Converges at some good points, but this approach isn't the best one to get to them. Even the maths he uses are over complicating a simple enough statement: No need to crash a hero, it's better to just win two of them exhaustively. So two 6 are better than a 10 and a 0. I'm not saying there aren't good maths to pursue this conclusion through machine learning, I'm just saying that this model of him isn't the best one. Even intervals or individual Gaussians would give such information better than his approach. His conclusions are pretty much fair and good, but I would not use his methods.

[@Uncle_Sunpuppy]

And then Kat was released for Playstation All Stars lol

[@planetzr3]

Lmaoo, just looking at graphs based on shitty netcode online matches? Do they even play their own game? If u really want to balance a game just play it. take top player feedback seriously, not so much in the early lifespan of a game tho, and buff other characters more then nerfing

[@ganjaman59650]

all star battle royale was fun? i must have missed something playing it, because what i played was a shameless ripoff with worst mechanics of smash bros...