It’s very very easy to build all these nodes yourself, and I do on a basis of necessity, it’s more the principle of the thing.
A bunch of these also do exist just in the form of the Color Mix Node
Thanks though
I saw this problem with MacOs vs Nvidia (i guess MacOS Cpu vs Gpu) with the powerfunction of the DCTL.I have not testet this specific problem if its occurs with Blender math nodes.However there is a workaround if this problem occurs.I have builded a small nodegroup for this in my newest node build.
Look the video at 10:35 min
my little nodegroup of powf.since we dont have copysign function this should work the same.
would be nice to have a graph overlay to show the math output like curves as in resolve or desmos etc
edit,
i have added a simple UV math plotter to the compositing
default power function
powf function
In Blender this negative value problem with default power seems to occur as well
I have three more ideas I need to try. Currently I can’t because I’m not at home so I’m writing this in part as a reminder for myself.
The ideas stem from the recognition that the sRGB “gamma curve” is itself a shaper function meant to map values to something closer to “percetual” (twice the value roughly twice as bright to us  at least for greys) as well as to compress values so more of the limited 8 bit worth of information captures differences we can actually perceive.
Therefore here are my three ideas:

The simplest one is just to mimic what sRGB does and to adjust my sigmoid such that a grey in of ~0.18 gets mapped to ~0.5 and then, crucially, do NOT convert from linear to sRGB. The sigmoid curve itself effectively already did that job! At the end just do the hue constant gamut compression (lift negative values to a minimum of 0 in the target gamut, restore the intensity estimate to what it was before that lift, then, for everything that exeeds the gamut, divide by the maximum to fit without a clip)

Alternatively, use the sRGB gamma curve as preshaper and set up the sigmoid such that it maps 0.5 to 0.5 — once again don’t go from linear to sRGB. That part already happened. Just compress to fit again)

The most complicated variant involves going through the HDR PQ curve, first shaping with it, then sigmoid (0.5 → 0.5), then undo PQ, and finally go through sRGB gamma and compress to fit.
If anybody wants to try any of these, go ahead. I’ll get around to it eventually if not
Why so complicated?The question is where is middle grey.The gamma curve is device dependent needed for your display.
Even if you match your sigmoid to the gamma curve,unless you want a different result as with the linear to gamma sRGB, i dont see the advantage.
Thus is technically less complicated for the first and equally complicated for the second approach, only the third is actually more complicated
I also found a different sigmoid I’d like to try although the differences are probably marginal at best: use the Fréchet distribution
It’s probably not a super meaningful difference but I like it for its theoretical properties concerning the maximum.
Just gotta reparametrize it in terms of middle grey behavior and slope, which I think is very doable