>>981117
>wouldn't it be better always to try to get it over ATK o DMG
Over ATK, yes.
Sheer scales linearly with both HP and ATK, and given a limited number of sub-stats, HP% is always better, no alternatives.
The same idea as how Ben technically scales with DEF%, but both ATK% and DEF% sub-stats contribute to the same pool and ATK% sub-stats simply do it better than DEF% ones.
DMG% is a separate multiplier. Whether or not running it is better or worse than HP% relies entirely on the saturation.
Yixuan saturates DMG% herself a bit but her BiS supports generally don't scale it
too much (still scale it quite a bit), so DMG% #5 for her tends to be very good exactly because of that. But if you saturate DMG% one way or another, HP% might start outperforming it.
HP% on Yixuan suffers from the same exact problem ATK% does on everyone else (and arguably more so, because Yixuan also has a non-zero amount of ATK contributing to Sheer Force as well) - HP% sub-stats do not multiply everything. 1% HP increase does not equal to 1% damage output increase, because there's always flat HP coming from the disc 1 and that is not being multiplied.
1% DMG increase is a 1% damage increase.
The "HP% is better than ATK% by 70%" applies to HP% for Yixuan compared to ATK% for Yixuan.
When you're looking at #5 for Yixuan, you're generally working with 30% HP from her W-Engine, then 30% more from #6 and then you're considering either getting 30% more from #5 or 30% DMG instead.
She usually has 60% from her Core + 30% on specifically Cloud Shaper (but not her CA or Ultimate which are are huge part of her total damage output) + 16% from her W-Engine.
If you're running Pan, you add 20% DMG but also a huge chunk of flat Sheer Force (which devalues HP% in a similar way flat ATK devalues ATK%).
Astra would add 20% DMG, 40% if you're running her signature, while also devaluing HP% a bit with her flat ATK buff.
Astral is another 24% DMG. If you're running Fufu, you'll be getting 20% DMG on CAs and 40% DMG on Ultimates.
So, depending on a combination, you might have your DMG% saturated more or less than usual. And that would result in #5 being better or worse when compared to HP%.
Because DMG% is inherently better than HP%, it will take it longer to saturate to the point where it becomes percent-per-percent worse than HP%, but there can be that point.
HP%, meanwhile is not limited to your W-Engine and #6 either - you can saturate it with your sub-stats. If you have very few of those, there's a good chance HP% #5 can be BiS for you. But usually DMG% performs better.