Banisher combo critique

Banishers and panthers do not form particularly good combos, partly because banishers will teamkill your panthers if enemy units run up on them and partly because panthers are way faster than banishers thus preventing squads from staying together well. You might still want to have both at the same time, but in that case it'll be more like 4-6 panthers per banisher and the banishers will be defensive.

2:1 reaper/banisher is pretty standard, but 3:1 is generally better in the late game since banishers die easily and are only there to stop your reapers from getting swarmed.

1:1:1 panther koda banisher is dumb. You need lots of panthers to stun stuff and kodachis do not pair with anything because they teamkill horribly with fire. Kodas and panthers are particularly bad because the panthers that get teamkilled will then additionally stun your units with their death explosion.

Golis don't pair with banishers particularly more than reapers do. You're better off building one or two golis and then 3:1 reaper-banisher spam to back them up. You could even do straight 1:3:1 goli-reaper-banisher if you can afford it.

quote: The right number of banishers? As many as you need.

I'd have said "as few as you can get away with".

quote: I'd have said "as few as you can get away with".

This.

quote: Unit ratios are bad in the long term, you should be reacting to the information your receive about your opponent's composition.

I don't really agree with this. If this was a good strategy then circuit would be programmed to use unit counters with a higher weight than it does.

Basic unit composition/ratios are good for building an army that's strong regardless of what your enemy does in general, and even if the enemy does something that demands a counter you'd still respond with a unit composition with more or less fixed ratios, just a different unit composition than if they hadn't.

Sometimes, and especially with banishers, if you need to build them as a counter you can just alt insert them into your regular queue and then return to business as usual once you have sufficient numbers.

Banisher should support heavy assault units. Mixing it with other raiders can be great mistake because banisher aoe is very large. Banisher well works with reapers and as goliath support. Also banisher can pretty good work vs not only raiders but also sometimes ravager balls.
Two reaper plus banisher is good. Golly plus three banisher seems too much. Golly have its own slow gun and its now buffed. So its enough with one banisher. But if you are suicidal player then you need even more.

There are definitely units you build because you want to and units you build because you feel your opponent has forced you to. I'm unsure whether this is good design or just a symptom of slight imbalance. Banishers are perhaps good early but later in the game I think it is important to win at attrition. Consider some non-linear build 'ratios'. For example you could make a Banisher whenever you have built a square number of Reapers. 1,4,9,16 etc...

I don't think about my compositions in terms of "ratios", but it is important to maintain a well-rounded composition which can deal with many threats. Specifically, you want to deal with the range of threats which you believe your opponent might deploy, which is where reaction to the opposition comes in.

That being said, building generically OK compositions with few critical weaknesses (or at least, weaknesses which are difficult to exploit at that time in the game) like Bandit/Racketeer, Rocko/Warrior, Reaper/Banisher, Grizzly with Duck/Archer/Scallop escort, etc. and then outskilling your opponent in other dimensions like macro, micro or strategy is a viable strategy (and is I expect how KGB wins if it doesn't react much to opposing compositions).

Deciding what units to build (and what factories to switch to) to counter an enemy composition (and what you expect your opponent to have built to supplement their composition by the time your new units get to the front) is a thoroughly nontrivial proposition and it would impress and surprise me to see an AI demonstrate skill in that area.

Just because you're reacting to what your opponent has built doesn't mean you don't have a proactive strategy. You do want to be making an army which will be effective against the opponent's army when you attack.

quote: Also, circuit is more like "build random units based on a probability distribution", whereas ZKGBAI is more like the "maintain a ratio of 2x to 1y and repeat" that you describe, yet beats circuit's fancy unit counters system just with that (although it does build some counters reactively, but that's only arty and AA).

I dont think KGB beating Circuit has anything to do with production choices. I think its production choices are pretty good. It comes down entirely to a combination of


-> KGB concentrates force far more intelligently than circuit

-> KGB uses AA a tonne better

-> Circuit often fails to make the early defences it is supposed to make at the moment, leading to early raids wiping it out

-> Circuit often makes far too much energy and expands very slowly, failing to compete on map control


Circuit can sporadically beat KGB on comet - on the rare instances that it survives the first 10 minutes without completely losing map control. I don't think the improvements needed to Circuit to make it competitive again will be too difficult to implement.

quote: Banisher

One per reaper ball - any more is metal wasted that could be spent on MORE REAPERS!