Try this different way. You’re right, something seems to have changed in the Transform node with that old way.
Scaling it down and back up again usually works, but concatenation carries through the original pixel data. So, if you scale an image down 10%, then back up 10x, it NOT push through the scaled down pixels, it will recognize that you want it to be the original size again and simply pipe that image through as if there’s no change.
This way scales it down, then inserts a node that does nothing except break the concatenation (the passing through of the original image), then scales it back up to the original size.
The old way should work, as well, if you stick a filter node or a blur node between the scales and set it to 0. It simply breaks the concatenation.
ps - I spelled concatenation wrong on the image. Sorry.