Number of instructions affected by a particular optimization can't negative. It must be due to overflow. I think it is a low priority bug, but its good to be fixed.
Proguard output window
Thanks for the report. There are a few cases where ProGuard has to add one or more instructions to make sure the code is valid. In some rare cases, it is too conservative, and it is adding more than it is removing; hence a negative number (you can see the difference of both numbers being computed near the end of Optimizer.java)
Now, 1386 added instructions really seems like a lot. Can you tell if there is anything special about the code? Do you happen to have a smaller test case?