it's in my experience if you dump +12V into 4164s you won't have a rug pattern at all lol so perhaps you have the voltage gimmick correct. I always did the scoresaves method by cutting the +12V trace to the ram section and bridging over old +5V to the old +12V spot (where the new +5V has to go). there are instances on boards I fixed in the past where you have to nerf the old +5V (which is the new A7 line) and ground it instead, but that's a whole separate elaborate mess that probably has nothing to do with your issue.
I've only seen multiplexers fail in the output stage at pins 7 and 9, I've wondered for a long time if the inputs could get jacked up too, but the kind of fast method for diagnosing those is piggybacking new 74153 chips onto the ones soldered into the board. if the game boots up fine then, power down, remove one multiplexer, and power up again until it goes wonky. then you mark the bad chip with sharpie or remember where it was and socket and replace accordingly. (sockets provide convenience but also provide a tell that there was a bad multiplexer previously also)
most likely it's bad multiplexer though that's causing your 1-3-1 line issue. it's a pretty common fault that causes 1-3-1. in the words of Ken Graham years ago this hardware had TTL that was pushed pretty much to their limits, so it's very much a thing to pop 7474s and 74153s.
NINJA EDIT: one of the 74153 chips only uses half the inputs/outputs so it's normal for pin 9 to be tied low. in fact any unused gates on TTL should be tied low, which is what inspired me to do ground mods for 4164s on the A7 line. also in my experience the 41256 rams are way better (more reliable, less chance of artifacts and glitches) but you have to do an additional mod to remove -5V because that is A8 and feeding -5V into an address line will KILL those rams. the Arcadeshop 4164 ram adapters actually nerf the -5V supply so they're compatible with 41256.
the only issue with using the larger capacity rams I've found is that when you do get a real ram error, the 1-3-x code that's reported is actually wrong. it'll be about 4 rams off, so like a 1-2-2 may actually be a bad ram 26. I haven't had to muck with that in a long time, so a) I'm just throwing it out there and b) don't sue me if I'm a ram off.
