All of us have underestimated the potential for aerodynamic misbehavior in this design, although I think it will be fairly easy to repair by adding a baffle and controlling detachment lines.
Here is a preliminary shot around 35 MPH:
That's right, I've got air moving up the rear glass. If I'm not mistaken, the low pressure at the base of the windshield and trunk is actually pulling air up from the underside of the car.
It's unexpected, but something I knew was coming, since at 65 MPH last week I saw rain water fly up from behind the trunk and land in the middle of the rear glass. (And thought 'd&^#, what a spectacular goof'.) I even knew right from the first day I drove it to work that something was amiss, since I coast down a hill from my house and I couldn't tell any difference in coasting or P&G rhythm. Certainly the MPG numbers from our 2 long-hauls with the box in place have been lack-luster. I don't have good baselines to compare to, but I think they registered somewhere around 0% or -5% (loss).
(So, a lot of good I did myself meeting my fab deadline without any testing.)
Not shown, because the image quality was terrible after it got too dark, I also got a brief tuft test with a quick-and-dirty cardboard-and-duct-tape half Kammback over the rear windshield (the version that doesn't cover the trunk at all). Even after rewatching it twenty times to tease out the info, I have to call it preliminary. The tufts that are directly in the stream (top of the Kamm and top of the Aero Box) seem to look reasonably smooth, and I think the tufts on the rear windshield may indicate a reduction in the violence of the flow, but I'm still getting some flow from the bottom to top. The Kammback did not improve the turbulence along the sides that I noticed.
So I need to improve my camera boom and test in broad daylight, and add some more detachment garnishes on the sides, and maybe a baffle to stop the upward flow at the bumper, and see how far that gets me.