Grey — the coat that goes white

Crossing two of them

Where it is common

Breed changes the frequency of the alleles in the founder pool and nothing about what grey does once present — a G3 Andalusian and a G3 Quarter Horse follow the same rule. Breeds holds the tables; the wild pool outside a breed uses Hardy–Weinberg over the three alleles.

G3 is the broadly prevalent one and dominates the grey breeds: the Lipizzan and the Camargue (fixed grey), the Percheron, the Andalusian and Lusitano, the Arabian. G2 is real but breed-clustered, so only the populations the duplication was actually reported in carry it: the Andalusian, Connemara Pony, American Miniature, Mustang, Quarter Horse, Tennessee Walking Horse and Welsh Pony. (The reference also names the Mangalarga Marchador, which this mod’s catalogue does not carry.)

Its gene carrot

Gene key

horsegenetics.grey

Alleles

G3 G2 N

Outcomes

wild, grey-slow, grey, grey-fast, grey-rapid

Shows when

GreyGene.dosage ≥ 1 — painted on adults only

Default allele

N

Wild frequency

1 in GreyGene.WILD_G3_ONE_IN per G3, 1 in WILD_G2_ONE_IN per G2; Hardy–Weinberg over the three

Founder draws

1 × nextFloat()

Deterministic

no when expressed

Epi draws

1 long + 8 × nextFloat

Health

an informational melanoma-risk condition, in three tiers

Generator

coat.pattern.GreyCoat

Three alleles, because the locus is a copy number

Grey is a copy-number variation in and around intron 6 of STX17 (syntaxin 17), with regulatory effects reaching neighbouring genes such as NR4A3. What varies between horses is how many copies of that stretch they carry, and a commercial copy-number test reports exactly that.

Either copy-bearing allele makes a horse grey, so in ordinary inheritance terms grey is autosomal dominant — not recessive, not codominant. But dominant does not mean the genotypes look alike. They differ in rate, and rate is dosage: G3 counts two, G2 counts one, and the sum runs 0 to 4.

Inheritance is then just Mendel on three alleles. A horse carrying two grey alleles — G2/G2, G2/G3 or G3/G3 — passes one to every foal, so every foal is grey; a heterozygote passes it half the time. What the pedigree carries is not grey yes/no but which copy number, which is why the gene stores an allele pair and not a flag: a G2 line greys slowly and carries the lower risk, a G3 line greys fast and carries the higher one, and choosing between them is a real breeding decision.

A rate, without a clock

The dosage picks a window and the horse’s own epigenetics pick where inside it the horse sits. The windows overlap on purpose: individual variation inside a real genotype class is large enough that a slow grey and a fast one genuinely can meet in the middle. What is reliably true is the ordering of the means, and that is what GreyCoat.WINDOW encodes.

Even the least-greyed adult keeps well short of its birth colour (KEEP_YOUNG = 0.46). With no age input, every grey adult has to read as a grey; a horse indistinguishable from a plain bay would just be a bay with a hidden flag on it.

Grey is a remap, and that is the whole point

Greying replaces pigmented hairs with white ones, and a mix of white and dark hairs reads neutral. So a grey has to land on the gradient’s zero-red column, the only place the LUT is actually grey.

So GreyCoat instead:

1. works out how dark each texel currently reads (0.55×red + 0.95×black);
2. writes that darkness back as black pigment, scaled by how far this horse’s greying has gone;
3. keeps only a fading trace of the red.

What was underneath still shows: a greying chestnut ends lighter than a greying black, and a barely-greyed horse keeps a rose or steel cast. The stored base colour is untouched — grey is a display layer over a bay or a chestnut that remains a bay or a chestnut in its genotype, its code and its foals.

Regions do not grey together

A uniform mask over the whole horse is the thing that makes a rendered grey look like a recoloured decal. Real greying is regional, and the regions are correlated rather than independent — they all move with the same progression, but each has its own offset.

Why the mane greys from the root

Greying is a change to hairs as they are produced, so the new hair at the root is whiter than the old hair further down. On a body texel that has nowhere to show; on a mane or a tail it has the whole length of the hair. The ramp runs along HairPattern.axesBySpan(skin, part)[1] — the same root-to-tip axis the midtstol is centred in.

The nine draws

Off the expressing grey copy’s epigenetics, in this order: nextLong() for the dapple field’s seed, then eight nextFloat()s.

DAPPLE_DEPTH is generous on purpose: most of a composed texel’s variation is the white template’s own shading, so the dapple field has to push harder than it looks like it should. Sample bakes put the field itself at a 21–29% lightness modulation on the mid greys, and about 5% on a nearly-white one.

The dapples

Dapples come from BodyNoise.cellDistance — distance to the nearest point of a jittered lattice, sampled in body space so the rings cross part seams without a join, and warped by a low-frequency BodyNoise.value so the lattice flows instead of gridding up. Near a lattice point is a dapple centre (lighter); out in the gaps is the web (darker).

// Warp the sample so the dapple lattice flows instead of gridding up.
double n  = BodyNoise.value(noiseSeed ^ 0x51L, x * warpScale, y * warpScale, z * warpScale);
double m  = BodyNoise.value(noiseSeed ^ 0x52L, z * warpScale, x * warpScale, y * warpScale);
double wx = x + (n - 0.5) * warp;
double wy = y + (m - 0.5) * warp;
double wz = z + (n - m)   * warp;

double d   = BodyNoise.cellDistance(noiseSeed, wx * scale, wy * scale, wz * scale);
float  web = (float) smoothstep(0.35, 0.78, d);   // 0 at a dapple centre -> 1 in the web
float  keep = lerp(keepDapple, keepWeb, web);

Chubari spots and the bloody shoulder

Two bounded variations on a fixed grey. Both are visual modifiers only — neither alters the genotype, the risk tier or what the horse passes on. They sit after the main grey pass in the same painter.

The selector for the chubari spots is deliberately sampled at a much lower frequency than the cells themselves. Sample it at the cells’ own frequency and it varies inside each cell and cuts it in half — which is exactly how manchado’s islands first came out as slivers.

Melanoma: reported, never simulated

The same STX17 change that removes pigment from hair drives melanocyte proliferation, which is why the two travel together. Grey horses carry a markedly raised lifetime risk of dermal melanoma — UC Davis reports over 70% of greys past fifteen, and published reviews commonly cite around 80%. In life the tail dock and perineum, the sheath or udder, the lips and the eyelids are where it shows. G3 carries more risk than G2, and a grey that is genetically non-agouti — black-based, where extension permits black pigment — carries more again.

GreyGene implements HealthContribution and emits one Condition.informational, with a tier from the copy number raised one step on a black-based coat:

Foals

restrict no-ops when !ctx.isAdult(), so a grey foal is born its base colour and greys once grown. One wrinkle: isDeterministic cannot see the foal flag (only restrict gets the CoatBuildContext), so a grey foal is also treated as per-horse. It bakes its own texture, and that texture is identical to a non-grey foal’s. Harmless — just a few extra cache entries.

What grey is not

What was in the reference and is not here

The source note proposed a five-stage health pipeline — genotype, then appearance, then susceptibility, then an age-related chance of lesion onset, then lesion burden and severity with health effects. Stages 4 and 5 need an age clock and a per-horse disease-progression system, neither of which exists, so they are not built and are not stubbed. The Condition is the whole of it.

It also suggested vitiligo-like depigmentation round the eyes and mouth, and flea-bitten grey. Both are parked with aging: see the settled list.