Rxquality is measured from drivetest or from statistics ?
it looks like a paradox indeed, but you should check the number of samples : maybe you have the same amount of samples within RXQUAL 0-4, and more samples with RXQUAL > 4 (compared to before). Therefore, the ratio is less favorable after AMR. Perhaps, before that, the samples > 4 were limited by the occurence of a drop. Thanks to AMR : no more drop, but the call is kept with a qual > 4.
It has to be analyzed, i’m not saying this is what is happening exactly. But you see how AMR can modify your average quality.