RFD 1151: Which axes the survey weights
placeholder_cards.py, CORPUS_DESIGN.md
Problem
A sweep across an axis has to allocate time or samples somehow. “Uniform” and “by population” make different claims, so choosing by taste makes a claim without saying so.
The Nem x Mila Social VR Lifestyle Survey, n=1,012, measures avatar species directly. It does not measure skin tone. It measures residence, and its sample is 73.3 per cent Japan.
Decision
Persona species is weighted by the survey, because the survey measures it: humanoid 50, semi-humanoid 38, robot 6, animal 2, plant 2, other 1, monster 0 per cent.
Skin tone stays uniform, as a decision rather than a default. No published mapping from residence to Monk exists, and running the survey through one built for the purpose put Monk 10 at 0.19 per cent, two frames of 900, which is where the defect is largest.
Every category carries a floor of 1.5 seconds, including one measured at zero, and axes with no data get a card naming them, so a clip cannot imply coverage it does not have.
References
DETAILS.mdcarries the allocation and the six cards.