There are several variables that combine in skintone; color separation, saturation model, skin color, light color spectrum, white balance, lens coloration.
Color Separation: Good cameras have color separation. Less ideal color science will have 'accordioned' hues, compressing and expanding arbitrarily throughout the hue circle. For skintone - under quality light, the Alexa will have great separation between the pink aspects of the face (lips, blush) and skintone, which rides the line between too green and too magenta under flat spectrum light
Saturation Model: Photometric saturation models (like Alexa LogC+R709 or EC LogC+R709) have less saturation (the video model). Saturation amplifies hue difference, and photometric models can make skin tones seem more alike. The EC variations emphasize saturation (film backend), which foregrounds skin hue
Skin Color: The skin of the talent may be slightly more olive (dark and yellow in hue) or pale with pink as the primary hue accent (all vary towards magenta with emotion). Here is a recent still from the film I've been shooting, with two people in the same frame that had divergent skin hues:
Light Color Spectrum: On the Alexa (and cameras generally), filming under light that has a spikey color spectum (Leds, Flos, Sodium Vapor, etc.) can cast skintone distinctly to green or magenta, even with correct -
White Balance: A fully neutral WB is vital for EC, not only in skintone, but the whole of color space accuracy. Skintone is one of the reliable indicators that WB might be off on the green/magenta axis when under reference light without a white card reference (should anchor on the skintone line of the Vectorscope under sunlight and halogen). A common occurrence is that WB will be set for direct sun on the GH5 (possibly with the WB sun preset), and then the shoot will move into the shade, which will shift the WB towards blue/magenta (needs a new WB, or correction in post ahead of the conversion)
Lens Coloration: Interrelated with WB is lens coloration, which is like a WB that variably shifts throughout the grayscale [For EC: it is optimal to WB around middle gray, or a little brighter for the most salient part of the image]. This can vary from minimal coloration (Sigma 18-35) to heavy coloration (for example, Xeens). For NDs - the reference EC ND is the Firecrest FSND, which evenly cuts the full spectrum (visible/invisible). Non FSNDs (even IRNDs) will have a varying ratio between IR/UV and visible, which can unpredictably shift sensor response to a given light (interacting with the camera sensor's native cut). Variable IR ratio particularly acts on the green/magenta axis
Notably, EC Tungsten does have a slight magenta bias in the skintone region towards magenta compared to the Alexa. I've figured out what in the data was causing this problem (IR leak every third exposure slice), and this will be solved for the next release.