Everyone who plays board games has the friend who already knows the rules and teaches you while you play. Some of them are dull about it. Some of them layer the systems in one at a time, and you never notice you're being taught. We wanted the video game version of Catan to be the second friend.
Two goals. Learn it solo, because it's a social game but you can't count on people to help you. And make trading fun, because trade is the whole social engine and every competitor's trade screen was miserable.
The problem with trade is a sentence. I say brick for wheat. Am I offering brick, or asking for it? At a table you resolve that in half a second with your face. In a UI it is a genuine puzzle. We tried arrows. We tried the words Give and Want. We tried an animated marquee. Each round of usability we iterated the screen and each round it helped a little less.
So we stopped fixing the screen and fixed the consequence. If a new player made the reverse trade by mistake, the AI was friendly enough to trade it right back. You lost nothing, and you learned the direction in one shot. We also loaded the dice early and used Klaus Teuber's balanced starting map so a bad first placement couldn't lock you out.
Brian Reynolds sat in the lab with us. He'd watch someone struggle and code the fix while the next person played. The rule in that room was simple. You can't interrupt. You have to watch the pain. It transfers to you, and then the next person has a little less of it.
The thing nobody expected: in later tests, people were sure the AI players were ganging up on them. They weren't. It just felt that way when three opponents all refused your trade. That one is still on my list of problems I haven't fully solved.