Ellis, fascinating idea. Charlie Munger frequently echos 19th century mathematician Carl Jacobi as saying when you are faced with a difficult problem, "Invert, always invert."

Along similar lines, there may be another way (I think) to determine the robustness of the EV algorithm: to calculate EV based on different sample intervals (30 seconds, 10 seconds, ...), or even just by shifting the minute window by a fraction of a minute. If EV is reasonably robust, and if the large players are not hiding their activity from those looking at minute boundaries, then the results of EV calculations over different time periods should be more or less consistent with each other, at least when accumulated over a full day or more.

I've done some analysis (very, very limited at present) of collecting historical one-second data for a few stocks, aggregating the one-second data into various combinations of EV intervals as described above, then calculating LEV/SEV on those modified intervals. I do see some differences that, to me at least, look different enough that the system might give different signals. Keep in mind I've only done this with a small number of securities and it is quite possible that the averaging effect Pascal does over 1000+ securities may well smooth these differences out and wind up with roughly the same signals. IOW, these per-stock differences may not be meaningful in the full context of the EV method. That would be much harder to test.

-Mike