RFD 1151 details: the allocation, the floor, and the cards
What the survey measures, and what it does not
Q2-1, avatar species, n=1,012: humanoid 50, semi-humanoid 38, robot or cyborg 6, animal 2, plant 2, other 1, monster 0 per cent. Measured directly.
Q1-1, residence: Japan 73.3, North America 14.1, Europe 6.0, Asia except Japan 5.1, Oceania 0.7, South America 0.3, Africa 0.2, Middle East 0.1, Russia 0.1 per cent. The slide states a sampling error of 4 to 14 per cent at maximum, and Africa is n=2 while the Middle East and Russia are n=1 each.
The deck asks about skin tone nowhere.
Why tone stays clear of that weighting
Run through a region-to-tone table built for the purpose, the survey gives Monk 10 about 0.19 per cent, which is two frames of 900. That is the tone where the flat multiplier is weakest, dE 4.83 against 17.28, so population weighting hands the worst case the least screen time.
That table came from nowhere citable. The Monk scale publishes colour rather than demography, and the one global-looking distribution in the literature is a study of skin-tone EMOJI on Twitter in five bins, measuring selection behaviour on one platform.
The survey also finds 66 per cent prefer an avatar unlike their physical selves, so avatar appearance is chosen rather than inherited, which leaves a user demographic describing the wrong population.
The floor, and the minimum honest length
Every category gets 90 frames, 1.5 seconds, the same hold the tone ladder uses because that is the duration shown to read. A category measured at zero still gets it: monster is 0 per cent of 1,012 and a category rounded off the screen cannot be looked at to see whether it is handled.
Seven personas at that floor is 10.5 seconds before any weighting applies:
| clip | floor share | humanoid : monster |
|---|---|---|
| 15 s | 70% | 2.5 : 1 |
| 22 s | 47% | 5.0 : 1 |
| 30 s | 35% | 7.6 : 1 |
At fifteen seconds a 50 per cent share and a 0 per cent share differ by 2.5x, so the clip would look survey-weighted while being mostly floor. The floor therefore sets a minimum length below which the weighting is a claim the picture does not support.
The cards
Six, one per axis with no data: persona species with no avatar assets, skin tone by population with no citable distribution, the test split at 0 records, identities at 1 of 23,000, global illumination unimplemented, and subsurface and outline assets unused.
They are derived from one GAPS tuple, and a control asserts every gap appears in the sequence. A gap added to CORPUS_DESIGN.md without a card, or a card with no gap behind it, fails rather than passing quietly.
Left unmeasured
Whether the persona weighting is right, because it cannot be rendered: the axis needs six species of avatar asset and this repository has one human body and 1,254 CC0 objects.