We're far afield from my known space
Same! It's hard to convey uncertainty in writing, so consider this a blanket statement for my position.
We agree on what chaos means, although we use different approaches. The econ paper I mentioned also uses "SDIC", sensitive dependence on initial conditions, as their definition of chaos. As an example, if you were to make the smallest possible change to the initial conditions, the effect of that change on the system must grow if time grows. So in a chaotic system, one begets the other: if something is SDIC, then it must also be unpredictably unpredictable.
I dug up that econ paper, which is from 1997. Their conclusion is that there's some interesting theories about chaos in markets, but that the indicators of chaos used up until then aren't good enough at separating chaos from regular noise and that the datasets are too small. The only interesting followup study I found looked at the Euro/Dollar market in 2012, and it rejects chaotic behaviour. Personally I think the problem with chaos is that it is too difficult to empirically prove, which is why it seems to be largely ignored by econometrics and economists.
With that out of the way, I think there are a bunch of interesting questions Ben doesn't answer satisfactorily (which was the point of my earlier post.) Assuming that all markets are to some degree chaotic, what do we do with that information? I mean, if infinitely small initial conditions can throw an entire system into a change, what does that mean for predictions? Historic analysis is then useless, but I don't know any other means of creating market behaviour theories. His narrative approach sounds interesting but I can't distinguish it from what I know about (market) sentiment. What I think he's aiming for is a non-human-centered approach. Maybe instead of looking at the activity of whales in crypto and following in their footstep by buying/selling alongside them, you could let a computer decide on market movements and whether buying or selling will lead to more or less money. Both examples sound like sentiment and like "narrative" to me.
Another thing I wonder is why Ben seems so sure that this isn't being done already. I mean, did you read Lewis' Flash Boys? Besides ML-vagueness, I wouldn't be surprised if high-frequency trading bots use minute-to-minute indicators to predict the next minute. They probably don't make on the fly assessments on whether Germany's bonds are good ones or bad ones, they probably make hundreds of estimates of whether this market will go further up or down if they have been going up or down. It's the inertia they're making money on, I think.
The nugget of Ben's writeup, at least to me, is that there's a good argument to make for throwing out all of our economics-related social constructs. Yield spread? Doesn't say shit. Quality? Nope. Death to the KPI, long live computations.