Ah aight
I've written it all in more detail on rentry but let it be
So ime Crunchy responds to RAM nicely as well so that's the third discipline I've suggested (apart from valid and PYPrime)
Valid isn't limited to TR4 as of now but could be ? I just see no point in the handicap, let it be whatever runs best. Maybe something will be discovered on TR4 that isn't known to work on AM4. Other limitations, I say let it be all the same deal as usual in the category. No one ime has complained about disabled cores there, for one thing.
The remaining stages, it was strongly advised to limit the core count for the TR4 stage to something like 8 (as per the 1900X) and I agree - otherwise someone with a 2990WX will just swoop the leaderboard. Whether SMT must be off, not sure (all TRs have it so no unfair advantage). This stage as of now will be Pi-10b Crunchy.
Whether AM4-only PYPrime Hynix (or any other "unusual" idea, say, only run GDM off, only run 4 DIMMs etc.) needs to have a core and/or SMT limit, not sure. There wasn't any at the last Challenger or Team Cup for all I remember ?♂️
Honestly I say minimal rules best score. There are enough effi comps happening on the bot but maybe there's more to the Zen1/+ memory controller than effi. So let it be all means necessary (that means all cooling) except those fooling the monitoring software. There are enough tricks circulating around that help RAM OC on early Ryzen (or on any platform in general) on a hardware level, not software, but they're not widely used for submissions in this particular scenario (no results on par with Rocket Lake let alone DDR5 = no points = no interest). This could be a neat showcase of those imo, and I personally wouldn't even consider them cheating, they're known stuff.
Cheating in the way of forging IC identification (for the Hynix stage if that gets a green light) is possible to prevent though, trust me on that. G.skill comp has already set a precedent for video proof ?
Also core clock WILL have to be limited for PYPrime and ycruncher (so anything except valid), agreed on that. Don't wanna see this turn into a core godbin contest instead of IMC godbin contest (or that+clever application of known stuff) like it was supposed to. There's a typo in the TR4 ycruncher stage, should be 4005 MHz like for AM4 PYPrime.