UC Berkeley StarCraft Class, Week 8

This week's StarCraft class was about the techniques of scouting and anti-scouting.

Replay 1

In an amusing replay of a Protoss vs. Terran match, the Protoss player pulls off quite an anti-scouting maneuver. Apparently he's a Korean player who is known for ridiculous strategies and he likes to suddenly put on some music and start singing during the critical moments of his plans.

In this game, he uses two probes to block the ramp leading to...an empty base. The enemy's SCV scouts, tries to get past these probes, but the probes kill the SCV. Later, he replaces these probes with two zealots to block the ramp, and he even builds a pylon outside the ramp. The other player is thinking "this is obviously his main base." The zealots fend off another SCV scout.

The Terran then sends the full force of his army at this ramp, and finally gets up it. The zealots run around a bit, trying to waste time as the Terran player realizes this is a completely empty starting zone with no Protoss base at all. At this time (cue the music!) the Korean player does two things. Remember that pylon just down the ramp to this fake base? Next to it he suddenly builds a whole set of cannons to lock the enemy units inside the empty base! Some nuance of strategy was lost on me here, but apparently the Terran used a fairly standard Terran vs. Protoss build here and that meant his tanks didn't have siege mode yet, so the cannons were especially devastating. Second, the Protoss player had been building mass dragoons this entire time at his actual base and sent them to the Terran's (undefended) main base. The Terran realized he was in trouble before the dragoons even arrived, as the Terran had already floated away all his builds to be annoying. Interestingly, the Protoss player forfeited at this point. Apparently this match was broadcast live and it was too boring to have to tech to air units and go hunt down the floating buildings, so the Protoss player just quit, ha.

Replay 2

In this match of Zerg vs. Terran, the Terran player scouts with an SCV to find a Lair in the process of building and a hydralisk den already in place. He expects lurkers, we assume, so he builds up a ground force capable of responding. He also starts building two starports, which the match commentators take to be an interesting and unusual move in this situation. The Zerg player does show up with lurkers, and this really contains the Terran player. He's stuck with only comsat as a way to reveal them, but it does look like he has the forces to deal with these lurkers.

Then the Zerg does a surprise tech switch. Remember, the Zerg only need to build one spire in order for ALL of their existing production facilities to make mutalisks. He shows up with a pack of mutalisks at the Terran base, and the Terran is totally unprepared for this. The mutas start threatening the Terran mineral line. The Terran player manages to get 4 wraiths together, and attacks right away, unfortunately not having cloak yet. Some bad micromanagement eventually leads to him losing one of those wraiths, and only barely holding off the mutas. He then builds science vessels and uses irradiate to stop a fair number of mutalisks, but by this time the Zerg player has build a huge number of hydralisks and lurkers (switching back his tech!) and simply overwhelms the Terran.

Theory

Professor Feng reminded us that the enemy's resources consist of 1) the stuff you know about and 2) the stuff you don't know about. Scouting is most important if there's something you MIGHT discover that you can actually respond to.

He also showed us a very simple decision tree. First, ask if you can win the game RIGHT NOW. If you can, then you should attack and build more units to secure the win. If you cannot win right now, you then ask yourself if you are going to lose right now. If you are going to lose, you should build more units anyway, in hopes that the enemy's travel time or bad micromanagement skills or something might give you a lucky chance. But if you are not going to lose right now, you should invest in economy so you can be as strong as possible later.

Another way of thinking about it is that you are often in a position where you would ideally invest in all economy, but you must cut into that to make enough units to avoid losing immediately. Scouting tells you how far you need to cut into economy to avoid losing immediately. And if you cut into your economy more than you need to (build too many units and not enough economy) then you are just defeating yourself because you'll be too weak later.

Finally, he reminded us that scouting is most beneficial when it involves a discrete choice. Many things you could reveal are on a spectrum, and you might find the opponent is about in the middle or a little to one side or the other of the middle. But in a discrete choice, they are either all in one way or all in the other, so that is very useful to know. If you see a spire building very early in the game, they simply don't have the resources to support a lurker rush if they spent the minerals and drone on that spire. And if their tech involves dark templars and you reveal that, you know you must make detector units.

Protoss vs. Protoss Idealized

Another example Feng covered was the number of gateways each side might build in a Protoss vs. Protoss game, assuming a fairly open map without ramps. Each side might choose to build 0,1,2,3, or 4 gateways in th early game. If the opponent chooses to build 0 and you build more than 0, you can probably rush them down and prevent whatever trickery they tried. Likewise, if they build 1 and you build 2, you can probably rush them down and win. If they build 1, they probably also built a forge and some cannons, but even then, Feng said that is really your advantage if you are the player who built 2 gateways. You can move all your forces around the map, but the cannons are stationary. The 2 gateway player might still win on rushing, and even if he can't, it's easier for him to expand with his more numerous and mobile army.

The more-gateways-is-better concept breaks down at 4 though. If you make 4, the opponent can actually out produce you with 3. It's just too hard to have enough resources to keep your 4 gateways constantly pumping out units in the early game, especially when you had to pay the price of one more building than the opponent with 3 gateways.

The professor said that 2 gateways is probably the best choice, though it looks like 3 is reasonable as well. Apparently with 2 it's possible to live long enough to invest the extra resources in dragoon tech or dark templars or expanding, but scouting would be vital here to know how much of a danger you are in.

Scouting by Implication

This is when you find clues that point to what the truth of the situation must be. If you scout a starting position that's empty that can imply where the enemy's actual base is, even though you haven't been there yet. Or seeing an enemy worker scouting can imply to you where main base is, considering the direction and timing of the enemy worker. As mentioned above, scouting an early spire actually does imply that mutalisks are coming because it's too costly to fake that spire early in the game. Seeing a Terran build a full wall implies something about what they are up to also, especially considering that walling off a ramp means the Terran lost many mining cycles he could have gotten (lost minerals) if he had built the supply depot closer to his command center.

Scouting by Intention

Sometimes scouting reveals an immediate intention the enemy has. In one example on a map where there are bases in the four corners, but nothing else worthwhile on the entire map, one player scouted to find the enemy army halfway between two bases. This immediately revealed the enemy's intention to attack. If he were not going to attack, it would be foolish to have his army away from his own base. A different example is if you scout a drone leaving the zerg base, and then your own base is NOT scouted by drone soon after, you revealed the opponent's intention to expand. Surely that drone is building a hatchery somewhere right now.

Automated Scouting

Remember the patrol command with the "P" key. You can tell an unit to attack-move back and forth across several points that you specify. This is most useful with an invisible unit like a dark templar or observer because it lets you keep tabs on several bases without spending any APM later. If something happens to your automated unit, you know something is up. Similarly, a burrowed zergling at an empty base lets you know that base is not yet an enemy expansion.

Scouting by Denial

Terran walls, zealots blocking ramps and other blocking maneuvers can contain the enemy. The more you contain the enemy, the less your scouting burden is. In other words, the more you can slack off on scouting and invest your attention in other ways.

Scouting by Timing

If you see a building with a long build time that's in the process of building, or even if it's finished, that tells you something about what the opponent had when he made that choice. You might learn that the opponent had to cut very deeply into his own economy to have this building so close to completion at this moment. This is best for things with long build times like spires, hatcheries, fleet beacons, and factories.

A Crazy Trick

Professor Feng surprised many students with this strange trick. Build an assimilator (or your race's gas extractor building) on the enemy gas geyser. On the most straightforward level, this can act as a scout if you built it at their expansion before they do. If your building is destroyed, you know they expanded. But the trick is to assign a hotkey to this gas-building. Then AFTER it is destroyed, you can press that hotkey exactly ONE time to go back to the gas geyser. When you do this, you can see the amount of gas left, even though you don't have vision to the area anymore. If you kept track of how much gas was there before, you'll know if the enemy has started mining gas there, and how much exactly they have mined. If you keep your vision there for a moment, you can even see the RATE at which the gas is being collected, so you'll know if they have only one probe, or two, or three mining there. Remember that you only have one hotkey press to use here; after your building is destroyed, the hotkey becomes unassigned after that one press. Crazy!

Lore's Replay

Finally, Lore showed us a replay of one of his matches this week. It's called "3 hatch spire to 5 hatch hydra."

He started by building a spawning pool at the 9th drone, a fairly balanced approach that keeps good economy without being too open to attack. He then sends 6 zerglings out to kill the enemy's scouting, and succeeds. The enemy is mostly in the dark here, but he at least knows that lots of zerglings might be coming. The zerglings show up at the Protoss base and to the credit of the Protoss player, he just finished building enough cannons to avoid losing immediately. The zerglings managed to kill about 5 probes before being killed though, still a good advantage to the zerg here. Also, zerg has stopped any fast expansions the protoss might have tried.

Zerg has 3 hatcheries now and a spire completed, just as a corsair shows up. The corsair starts attacking an overlord, and it looks somewhat bad for a moment until Zerg builds a pair of scourge that chase the corsair all the way back this base. The important thing here is that the scourge were the best choice to stop the *scouting* that corsair might do. He didn't even need a mutalisk to stop the scouting.

Next, Zerg builds a pack of mutalisks and goes to harass whatever he can. Meanwhile, he gets up to a total of 5 hatcheries. It looks like the most important thing the mutas ended up doing was actually the response they forced from the Protoss player. He was building templars, but he morphed some of them into archons to protect his mineral line. This move mostly worked, but from the Zerg player's perspective, the most deadly thing around is psionic storm. Every pair of templars that lose their storm abilities by merging into an archon is kind of a win.

With 5 hatcheries, Zerg switched out of mutas and back to hydras, massing a huge number of them. The Protoss did decently with psionic storm here, but it wasn't nearly enough. The mass hydras were good enough to overwhelm and incidentally, they are pretty good against archons too.

That's it for this week. Next week is spring break so there's no class, but I'll be at the Game Developer Conference anyway.