I’ve exhausted all ideas, I’m sorry… I hope one of the developers stops by your post. I guess I’d try saving the original product you read to BEAM-DIMAP and then reading this product to subset to see if you get the same bug. Another idea, although it’s a longshot : is the image flipped ? Are the features in the subset mirrored compared to the original ? I only mention this because you say your target area is on the far left but the result is on the right.