Thanks! For whatever reason, I'm not getting as good of a color difference in the ROM bits after removing the top metal layer. Before, the interior of the bits was always much brighter than the background, so I used a 2nd algorithm to ID the bits, then compared the two results to fix errors. But recently, the interior of the bits is nearly exactly the same as the background, so I have been doing more manual checking. I started converting the top metal pic to B&W and drawing the bits in red to help them stand out better. But checking 22,528 bits per game makes my eyes tired, so I'm happy you only found one bad bit.
I've made a couple of changes in my process that might have affected the bits. I started using sulfuric acid instead of nitric acid to clean the epoxy off the dies. This is much quicker; before I'd have to put the die in nitric acid for a while then fish it out and clean it to remove some of the epoxy, then put it back in nitric acid. Sometimes I'd have to repeat that 3-4 times to get it clean, and sometimes I couldn't completely clean it. Sulfuric acid acts faster and completely removes all epoxy. The only problem is that the acid gets dark brown and I can't see the die to remove it without pouring some into another beaker.
The other change is that I'm now agitating the test tube continuously when I use Whink to remove the top metal layer. Before I was agitating for a bit, then letting it sit. Agitating continuously takes about 1/2 the time, and I figured it would help remove the metal more evenly without "erasing" bits.
I need to experiment some and see if I can get back to running two algorithms against each other.