The problem is in 48C8. If you reverse the string (from 48C6 to 43DD9) you will see that it is not picking up the the 48C8 grain.
The problem is caused by flipping board 48. If you set up the map with 48 on top and 43 below and flipped, you will have (a) a correct map configuration and (b) no problem with the 43DD9 to 48C6 LOS. Of course, you are likely to find similar los errors on 43 now that it has been flipped.
When a map is flipped the pixel-by-pixel losdata is flipped as well and so is the hex grid alignment. Since the geometry of VASL board .gifs is a bit wonky this can cause minor issues at the margin of hexes (a pixel may be calculated to be in the wrong hex), and I suspect that flipping increases the chances of this taking place.
Clearly, a bug in the machine, I will note it as such. I suspect the search for a solution will take a bit of time.