For me it's the same, but with one additional step.

Test the RAM in a known stable machine. If it passes an overnight memtest, then replace the flaky machine's PSU.

RAM has always been my "go to" "I bet this is it" problem, but I've debugged my own machines and other people's machines and I haven't seen bad RAM in ages.

I avoid XMP, however, and typically manually specify the voltage that is listed in the RAM's spec sheet and run at the speed that the CPU manufacturer recommends. (So the RAM might be sold as 3200 but Intel says 2400 is the max, so I run it at 2400.)

I get your desire to be conservative here, but this is not good practice with modern CPUs. For example AMD desktop CPUs are designed with DDR4 speeds well north of 3000MHz in mind[1]. Problem is that JDEC profiles simply don't go that hight for some reason, so you you have to run with XMP profiles if you want the chip to perform as the manufacturer intended.

[1] https://www.anandtech.com/show/14525/amd-zen-2-microarchitec...

I look at Intel ARK and use what they say the CPU is rated for. I am a couple generations behind, so not at the bleeding edge. (Haven't used an AMD CPU, but I would find their equivalent if I did.)

Keep this in mind for when you do: The first boot will be in a "failsafe" 2133MT configuration, but the fabric clock is intended to ramp all the way up to 1800MHz.

You need 3600MT to match that while keeping 1:1 sync, which means XMP unfortunately.