I agree. And there's even a more fundamental problem with the hypothetical argument: the random seed. Every team could build their own map but each one would have a different random seed and would behave differently.Originally Posted by DS
As an example of the other extreme, suppose in a GOTM we'd decided to reveal the entire map upfront, including locations of resources. Now, someone who had enough patience with worldbuilder could do:
4. Build a test game that happens to have exactly the same map in every detail as the real game.
I think everyone would agree that is completely unacceptable, since playing their test game then playing the GOTM is no different from playing the real GOTM twice through.
That's a very poor hypothetical argument, IMO. Obviously, the thing that's wrong with it is not the test play itself, but the fact that you're playing the same map repeatedly with perfect prior knowledge. In SGOTM practice, what we do is update test games as knowledge of the map is revealed in game. There is no prior knowledge being used as a basis for play decisions at any point. In fact, the practice of test mapping is the opposite of your example, in that it gives rapidly diminishing returns the further it is advanced. To take it to the extreme, it would be both insane and utterly pointless to model a perfect test game up to the turn before you launch a spaceship. The primary use is for optimizing the early game tactics. As many people have already pointed out: this is no different from using your brain, pen&paper or Excel. It's just much more fun, and (most often) quicker and easier, to do with a test map.
This highlights the danger of test games and the fundamental reason why stw's argument is fatally flawed. Test games tell you nothing for certain about what will happen in the real game with respect to AI behavior.