VNP
Found another one in the meanwhile 
ok... 
I think the internal reflections may be a contributing factor, it eerily resembles the flat's and reflections' position 
I think you could go back from the assembled disk: take the final disk apply a large radius median, like 100 pixel (for the disk of 3000px diameter) subtract from...
In the meanwhile I got into some even more weird lines. Maybe I need to refine my google skills, but I guess I went where no amateur has gone before....
You said these are auto-generated binaries. Is there a way to know the difference between those using the same version number?
Perhaps a build number-somthing, the commit's hash included into the processing log file?
The log files of the fail and then the success look basically the same 
It's scanlines being skipped or something, perhaps a failed malloc/new 
found another one, quite randomly, in Helium 