Once you understand how to set up an actual "context" for understanding the market and can do so repeatedly, some portion of the market's movements become predictable with a certain level of probabilistic certainty, at least with regard to the binary question "Will I be able to make a profit on a trade if I enter here?". Think of the way a scientist tries to ensure that conditions are as similar as possible when running a set of experiments. Even something as simple as the humidity in a lab can throw off an experiment, so each variable needs to be tightly controlled and its impact understood. Only then can the scientist truly test the replicability of results. That's where you need to get with your trading. Once you have that, you have a trading set-up that you can go to over and over again and profit. Once you have that, the remainder of the market's movements can remain a mystery and, so long as you have the patience to wait for your set-up, are irrelevant except as part of your set-up. For example, sometimes a market will take 90 minutes to give me a set-up and sometimes it will take days. What happens during that 90 minutes or during those days is irrelevant except as part of the set-up. To continue the science analogy, it's just like a scientist doesn't care if it's raining outside or sunny, so long as the lab conditions are the same as the previous experiment. There may or may not be additional set-ups you can discover in the remaining market movements, but if you get even one that works and it works across markets, two is overkill.
I think complexity implies too much attention to detail. In the context of probability based trading (i.e. systemic trading), the more rules (more complexity), the greater the likelihood of overkill. I'd even speculate that its an exponential function: degree of overkill = concept^rules. So your concept better be darn simple or you'd better have very few rules. Fuzzy logic theory delves a bit into this: one can create a system of independent rules, each with significant individual merit. When those rules are combined into a system, a performance measure is taken. Then individual rules are randomly removed and a performance measure is again taken. The resulting system with removed rules will sometimes display significantly better performance and stability over the same input set.