Starcraft: Brood War — A BGH Analysis — Part 1

Introduction

In the world of team Starcraft, there are 3 broad communities. Fastest Map, with its stacked minerals and oodles of geysers, low money, mostly played on 4 player 1v1 maps (i’ll throw Hunters in here as well), and in between these 2 extremes lies BGH. BGH stands for Big Game Hunters, and it’s essentially just Hunters but with more resources. BGH is often regarded by the ignorant as a Protoss playground, where 60 APM guys in the chat go to mindlessly mass zealots, cannons and carriers, bad Terrans go to build walls with holes in them, and bad Zergs go to make creep colonies that will never grow up. Anybody whose played a few public BGH games knows this is perfectly accurate, but when decent players play BGH there’s actually quite a bit of strategy and skill required to consistently win. Even teams of very strong 1v1 players can lose a lot to experienced BGH teams if they haven’t previously been exposed to all of the depraved tactics experienced BGH players use to win.

This will (hopefully) be a multi-part series, where I will use data from a curated dataset of roughly 11,500 games between relatively skilled players to answer a few questions about BGH. Since 3v3 is the most popular (and my favorite) form of BGH team play, the remainder of this post will be dedicated to analyzing 3v3. In part 1, we’re going to try to use data to answer the oft-debated question:

What is the most powerful composition in 3v3 BGH?

In this context, “composition” refers to the races played by a particular team, usually referred to by the first letter of each race. For example, 2 Protosses and a Terran would be referred to as PPT.

The rest of this article assumed a novice level understanding of both Brood War and BGH. All data is taken from the wonderful repmastered app.

DISCLAIMER
For each of the claims I’m going to present, there is an ocean of caveats and yes buts. In particular, spawn location plays a huge role in determining the outcome of any individual game. To keep this article from turning into a novel, I’m going to ignore spawns, and just look at how each composition performs in general, averaged over all spawns in the dataset. Spawn-specific analysis will come in a future article.

The Map

Before we dive into the data, it’s worth pointing out a few important features of the map. If you’ve played a lot of BGH before, you can probably skip this part.

The Center

BGH largely follows a hub-and-spoke model, where every main base opens into a natural expansion, and then outlets to the center of the map. With some notable exceptions (more on that later), it’s impossible for many of the spawns to interact with each other via land without passing through the center. This makes controlling the center extremely important, for one simple reason: the team that controls the center can combine their armies, while simultaneously preventing the other team from doing so. This means that even if the other team has a significantly higher supply, the team that holds mid can reliably force 1v2, 1v3, or 2v3 fights and trade very efficiently as a result. Many high level BGH games end after one battle for the center in the early game, with the losing team preferring to just GG rather than try to come back.

Non-Center Paths

In addition to the shared chokes noted above, 1/3 have an extremely short rush distance, and multiple paths to reach the center. In fact, all of the top spawns, plus 3 oclock, can reach each other by land without passing through the center.

Shared Chokes

There are 2 pairs of spawns that share a single entrance to the center of the map. These are 11/12, and 6/7. Furthermore, there is a path between these spawns that does not pass through the center, and the player at 11 or 7 must go past 12 or 6 respectively to reach the center. Combine that with extremely short rush distances between these bases, and you get a very interesting dynamic. If the shared players are on opposite teams, they always end up fighting each other, while if they’re on the same team, they can (relatively) easily combine their armies and block their choke without having to pass through the center. A huge skill gap between 2 shared players on opposite teams, or 2 strong allies sharing a choke, can often decide a game.

No Ramps, No Highground, No Mercy

Unlike many low money maps, BGH bases are not on highground, and have a very wide entrance. Combine this with short rush distances and the threat of double/triple teams, and it can often be very tricky to secure your choke. This is particularly a problem for Terran, who simultaneously have the weakest early game unit and the most incentive to wall off their choke. Additionally, the plus-sized mineral line and wide open terrain means that players who don’t try to wall have a huge amount of area they need to protect in the event of attack.

Putting it all together, we can start to get a sense of the most powerful way to play on BGH. Teamwork, mobility, and aggression are the (un?)holy trinity of high level BGH. Let’s see how these ideas impact the 3 races.

Race Summary

First, let’s compile some summary metrics for each race individually. We’ll start by looking at each race in isolation, then we’ll look at pairs of races (for example PPZ/PPT but NOT PPP), triplets (PPP, TTT, ZZZ), and then we’ll put it all together to see how each composition performs in totality. For each of these parts of the analysis, we will exclude mirror matchups, since mirrors by definition skew win rates towards 50%. First, let’s get a feel for how each race performs in isolation.

Protoss is by far the most played, at 45.2% of plays, followed by Zerg at 30.2%, and Terran at 24.6%. However, contrary to popular belief that BGH is a Protoss map, it is in fact Zerg that wins the most often in 3v3, with a 56.6% win rate, compared to Protoss at 50.9%. Terran is a distant 3rd at 40.2%. That’s quite a difference! What’s going on here? Basically, it comes down to each race’s early game unit.

Zerg’s basic unit is the zergling, which becomes available very early, is very cheap, and has a speed upgrade that can reliably be researched extremely early. In the hands of a skilled player, speed lings are very powerful in the early game. Their extreme mobility essentially guarantees zerg-heavy teams early game map control, providing many opportunities for ganks. In high-level games, it is very common for early game zergling aggression to decide a game in the first few minutes.

Protoss also has a powerful early game on BGH, though less so than Zerg. The zealot is tanky and relatively cheap, but its also slow. Speed lings run circles around them, but the many narrow corridors on BGH means that once Protoss can get 5 or more zealots out, they can sometimes force decent trades even vs large numbers of lings. Doing so, however, requires the Protoss to be verrryyy careful, lest they get surrounded and wiped out. Protoss can also quickly get photon cannons, which can be used both offensively and defensively, or in some situations, dragoons. Zealots also combine extremely well with zerglings and marines, soaking up lots of damage while the lings and marines deal free damage.

Terran, by contrast, is by far the weakest early on. The marine is cheap, and the only basic unit with a ranged attack, but is also slow, easily killed and has a low DPS. In a straight up fight with zealots and speed lings, unupgraded marines typically get massacred, especially in low numbers. Combine that with the short rush distances, and terrans often find themselves facing very strong zealot/ling attacks before they even have time to build more than a few marines. Many a BGH game just ends on the spot with a Terran getting his wall busted and dying 4 minutes into the game. Even if Terran survives the early game, it often gets blocked in by offensive cannons or bunkers, forcing their team to play 2v3 until the Terran can get tanks, or enough bio to break out gorilla style. During that time, one of the Terran’s unfortunate teammates will often be brutally gang murdered, forcing the Terran and his remaining ally to play 2v3 the rest of the game.

For a bit more detail on how this plays out in practice, we can look at a duration chart, which show’s each race’s win percentage as a function of game duration, indexed to a 50% win rate. This means a value of 0 would indicate exactly a 50% win rate, and anything above or below zero indicates a positive or negative win rate respectively.

Protoss has by far the most consistent graph, never deviating more than 5% from 50% for any game length. This seems to indicate that Protoss is a very good all-arounder in BGH, being able to contribute meaningfully in the early game with mass zealots and cannons, in the mid game with dragoons speed zealots, dark templar and robo units, and in the late game with a strong economy and powerful spell casters.

Terran by contrast suffers brutally in the early game, losing huge numbers of short games, and not climbing above a 45% win rate until 10 minutes! Note that even though Terran has arguably the strongest BGH mid game, they still never reach a 50% win rate until 26 minutes. This is because even Terrans who can get cleanly into the mid game often still have to play from behind, since their team has had to play 2v3 for several minutes and has usually taken a lot of damage.

The Zerg chart is basically the mirror opposite of Terran, winning a hilarious 70% of 3 minute games, and staying well above 50% until the 10 minute mark, when their lack of economy starts to become a serious problem. However, even in these 10–15 minute games, Zerg win rates are still positive, as they’ve often gained enough of an advantage in the early game to survive even on a low economy.

Double Race Summary

The double race numbers follow a similar trend to the singles, but more extreme. Double protoss wins a shade under 52% of the time, but double zerg goes from 55.8% to an even more muscular 59.5% win rate. Meanwhile, double Terran wins a miserable 27.8% of the time. It’s hard out here for a Terran! Somewhere in the world, Artosis is smashing a keyboard in vindication. Double Zerg’s dominance is easy enough to explain. Having 2 Zergs going 9 pool speed essentially guarantees map control, unless the other team also has double Zerg. Skilled Zergs can turn BGH into a post-apocalyptic hell scape, with massive packs of rabid lings prowling the map and devouring any army foolish enough to move out too early. Meanwhile, the Terran/Protoss on their team is free to tech to mech or goons respectively, which gives some much needed beef to the mass zergling army. Double ling all ins with goon/vulture/tank support are a contender for deadliest build in all of Starcraft.

Looking at the duration charts, we again see more extreme versions of the singles trends. Double Protoss breaks even for the first 7 or so minutes, then has a nice spike from 7–15 minutes, and a mixed bag after the 15 minute mark. Double Terran loses most of its games in the first 10 minutes and doesn’t really manage to reach equality even in longer games. Double Zerg is the mirror opposite, winning most of its games in the first 10 minutes, breaking even from 10–15 minutes as their opponents start to outscale them, and then doing quite well in 15+ minute games, when things like high economy mass muta and hive tech start to come into play.

Triple Race Summary

Triple race teams are quite rare in high level BGH, occurring in only 7% of games, with PPP making up the vast majority of that. Given how strong double zerg is, you might think triple zerg would be unstoppable, but this is not the case. While ZZP/ZZT combine for over a 59% win rate, ZZZ has below a 49% winrate, only about 1% higher than PPP. This highlights a shortcoming of zerg on BGH that I’ve avoided mentioning until now. Because Zerg must choose between making workers and units, and because map control is so important on BGH, Zergs usually spend all of their larvae on units rather than drones early on. If the Zergs can’t get a quick kill, their speed lings and low economy scale very poorly into the mid game. This is why having a Protoss/Terran on the team is so important. Their large economy and beefier units provide the extra muscle needed to break heavily fortified enemy bases, or if that’s not feasible, they can use their higher tech army to buy time for the zergs to drone up and play a longer game. As we would expect, TTT gets mercilessly barbecued, winning only 20.4% of games. Ouch.

And the Best Composition is…

Let’s put it all together, and see if we can find the most powerful race composition on BGH. Double zerg seems overpowered, but triple zerg is just OK. That means ZZP and ZZT must be the best right? Let’s take a look! We’ll start by simply calculating the aggregate non-mirror win rate for each composition.

Ummm… what? You just wrote a mini-essay about how ZZX goes to the moon on BGH, only to rug pull and tell me PPZ wins the most often?!? Yes indeed. Never trust anything you read on the internet. To figure out what’s going on here, there’s one last piece of the puzzle we need. The time has finally come to look at individual matchups. First things first, how many possible matchups are there? There are 10 unique compositions possible for each team in 3v3. Since there are 2 teams, that makes 10 x 10 = 100 possible matchups. However, 10 of these are mirrors (PTZ vs PTZ for example), and half of the remaining 90 are duplicates. PPZvTZZ is the same as TZZvPPZ after all! That leaves 45 unique non mirrors, plus the 10 mirrors, for a total of 55 possible matchups. To help us visualize such a large number of matchups as easily as possible, I’m going to display the data in the form of a matrix. Each cell in the matrix will contain 2 numbers: the matchup win rate, and the number of games played for that matchup. Any matchup with fewer than 30 total games played will be greyed out, as will mirrors. Any matchups without at least 1 game played will be blank. Each matchup will be shown twice in the matrix, once for each side of the matchup. Winning records are highlighted in green, and losing records in red.

If you examine this table closely, you’ll see that there’s not a single composition that has a winning record in every single matchup! So rather than asking what the “best” 3v3 comp is, we’re going to have to be a little more precise with our wording. Given all we’ve explored so far, I think it’s fair to make the following statements:

• Given a game with 6 random players from this dataset who all pick random, the most desirable comp is PPZ, and the least desirable is TTT.
• For any comp, it is always possible for the other team to pick a comp that counters it.
• Having multiple Terran sucks.
• Not having a Zerg almost always sucks if the other team has one.

That’s not very satisfying! And it turns out this still isn’t the whole story. Some of you are probably wondering, what happens if we control for the skill level of each team? After all, Protoss dominates at the lower levels of 1v1, but gets flame broiled at the very top end of the ASL/SSL, while Terran and Zerg seem roughly even at the very top, but struggle more at lower levels. How do these numbers look if we include only games from the top players? To see that, we’re going to have to define how we decide who the “top” players are. While we can easily filter by ladder MMR, or ASL/SSL games to see high-level 1v1 games, theres no such equivalence for 3v3 BGH. The only 3v3 ladder currently running is the 3v3 BGH MMR bot on Europe, which while excellent, isn’t particularly useful for filtering for high level games. In practice, the top end of MMR bot is more a measure of how well you can carry weak teammates, rather than who can play well when all 6 players are very strong. Furthermore, many of the best BGH players don’t play the MMR bot games at all. So when it comes to player rating, we’re going to have to roll our own.

For simplicity, I’m going to use a system very similar to the Elo system used in chess, but with some tweaks specific to 3v3. First, I’m going to keep track of 3 ratings for each player instead of 1. Each player will get a separate rating each for their Terran, Zerg, and Protoss, starting at 1500 MMR. For each game, I’ll calculate the average MMR for each team, and use that to calculate how many points to add/subtract for each win/defeat. If you want more details on how Elo is calculated, see here. Games will be analyzed chronologically from the date they were played. Once we have these ratings, I’ll calculate the average MMR for each game across all 6 players, and take the top 20% rated games for each matchup. I’ll then show the exact same matchup charts as above, but filtered only for these “elite” games. For players with multiple aliases, I’ll assign all games to a single alias where possible. For fun, here’s what the MMR distribution for each race looks like. If you’ve never seen a box-whisker plot before, the box represents the 25–75th percentile of ratings. The line inside the box is the median, and the the whiskers at the end are the maximum and minimum ratings in the dataset (minimum 30 games played).

Zerg OP.

First let’s look at the updated win rate per composition table. I’ll put it side by side with the unfiltered one for comparison.
PPZ is still on top, but it’s now in a dead even heat with PZZ. Meanwhile, ZZZ gets a huge glow up, going from a 48% win rate to almost 59%! 3 of the top 4 spots are now occupied by compositions with at least 2 Zergs.

Now, let’s look at the matchup matrix.

This seems a little less ambiguous! PPZ absolutely crushes any comp with a terran in it, but now has substantially losing records against PZZ and TZZ. TZZ has a winning record in all matchups for which we have sufficient data, but struggles against PZZ over a small sample. PZZ also has a winning record in all matchups, with the only potential weakness being vs PPP, though we only have 12 games in the sample.

Conclusion

So, what is the strongest composition in 3v3 BGH?

It’s PZZ, and heres why: it doesn’t have any weaknesses.

If you know the other team has a Terran, PPZ is the way to go, since it simultaneously guarantees early game map control like PZZ, just with a stronger economy. This is why PPZ has the overall highest win percentage, it dominates any comp with a Terran in it. However, it’s hard to rank PPZ #1 when it has such a losing record in 2 different matchups. While PZZ has an overall lower win rate vs Terran comps, it still crushes them, plus it has a winning record in every single matchup for which we have sufficient data. The only possible weakness indicated in the data is vs PPP (33% in high level games), but only over a sample of 12 games. TZZ is a good 2nd option, since its favored vs everything except PZZ. Here are my final rankings:

1. PZZ
2. TZZ
3. PPZ
4. PTZ
5. ZZZ
6. PPP
7. PPT
8. TTZ
9. PTT
10. TTT

If you made it this far, thanks for reading, and I’ll see you in the next one!