Here is a different view of the same structure. This is from a different specimen on which the two halves of the proboscis had come partially "unzipped" as I attempted to uncurl it. The area of separation seemed like a place that might give some good insight, so I shot it.
The photography setup was basically identical to the previous one, with the exception that the Canon T1i camera was replaced by a Nikon D6000 and its matching AI-chipped M42 adapter.
Here are two views, one from slightly above the area of separation and the above right at it. The scale of these images is probably a little different, because I was careless about the cropping. But now that you know about it, I don't think it will be misleading.
I think it's starting to become clear what is the structure of the "zipper" mechanism here on the dorsal side of the proboscis (the outside of the curl). It appears to consist of a bunch of overlapping flattened scale-like projections that just interleave with each other. I say "just" interleave because I'm not seeing any evidence of hooks at this scale, although of course there could be at a smaller scale and it would make great sense if there were.
This image is oriented the same as the earlier one, so that "down" is toward the end of the proboscis. This means that the projections are pointing against the direction of fluid flow while feeding. While that seemed counter-intuitive when I first saw it, after more thought it makes sense to me in terms of the process the moth has to use to get the two halves joined. As hinted by LordV's photo above, joining starts at the head and proceeds outward. With the projections oriented this way, they will be inclined to nestle together smoothly. If they were oriented the other way, pointing toward the head, then the joining process would be working against their springiness and I imagine they would tend to get tangled up.
According to what I've read, the ventral side of the proboscis (the inside of the curl) is more complicated. I suppose my curiosity won't stop here.
As a technical point about the photography, it's interesting to note that the first image comes from a 101 frame stack, while the second is 210 frames because it comes from a more slanting part of the curl. In these stereo renderings, they appear to be about the same total depth, but that's because in both cases I used +-3% shifts. To give a more accurate comparison, I could have adjusted the shift amounts to correspond to the number of frames, but quite frankly that didn't occur to me until I had already finalized the images. It's something to think about for next time, though.
--Rik
Edit: to add second view, above the separation.
