[xapata]

I don't think there's necessarily a contradiction between a weak form of the efficient market hypothesis and the quite evident phenomena of reflexivity. Inefficiencies form, the agents seek them out and in so doing remove them. The question is how fast the inefficiencies are removed and how costly they are to discover. There's decent evidence that the inefficiencies can take quite some time to disappear. The tougher question is how costly they are to discover and exploit. Unfortunately, it's hard to gather evidence on that, because of the confounding variables (trade secrets and competition, selection bias).

Another way to put it: You might be the best poker player at the table, but you might not be enough better to beat the rake. Or, sure, you can count cards and win at blackjack, but the casino ensures through table limits that on average they'll still come out ahead. Card counters will spend too much money finding a table with a good count that they won't win enough back exploiting the count to make up for the cost of information. On average.