Somewhere in the background, while all this is going on, a minor horde of high-ranking programmers and prediction market experts are frantically scurrying around, trying to get out ahead of testing the new untested prediction-market trading backend, turning up bugs before they explode - now with 70% less statistical learning, but still enough statistical learning that advice propagates in three seconds instead of ten seconds.
Among the acerbic comments being traded on backend, where nobody in the operating theatre can hear, are comments to the effect that statistical learning is sort of like summoning a deranged fictional nonreductionist alien-spirit into your system: it will serve you faithfully until it stabs you in the back, and maybe the entire invention of statistical learning was a bad idea with bad brain-incentives, because it often initially seems to work great and your brain is happy about that. They're sad about how miserable their fates as programmers are, that they have to get involved with these statistical learning methods that are not always perfectly legible and sometimes produce bugs that are hard to understand even when you're looking directly at their causal antecedents. They, too, lack context.
Actually in this case, three-quarters of the people on this team have context; but everyone who has context is trying to act exactly like someone who doesn't.