Agouti — the four bays

What it does

Crossing two of them

Its gene carrot

Gene key

horsegenetics.agouti

Alleles

A a

Outcomes

wild, blood-bay, bay, liver-bay, seal-brown

Shows when

A/a, A/A — and only on a horse that makes black

Reads

horsegenetics.shade, via coatDependsOn()

Default allele

a

Wild frequency

50/50 per allele

Founder draws

1 × nextFloat()

Deterministic

no when expressed

Epigenetic draws

5 × nextFloat

Generator

coat.pattern.BayCoat, genetics.BayShade

What the shade score paints

BayShade.spread() rescales the score to 0 (black held to the points, reddest body) to 1 (black over nearly all of it). BayCoat reads that one number and moves four things.

The body ramp’s middle is the horse that was already right

The old bay had one body-black constant, 0.32, and it was verified in-game. BODY_CURVE is chosen so that bodyBlack(0.5) comes back to exactly that — the middle of the new range is the coat that was already known to be good, and only the two ends are new. BayShadeGeneTest asserts it, so the constant cannot drift out from under the claim.

The soft points — what makes a seal a seal

• Muzzle. Strongest at the nose, tapering to MUZZLE_BACK where it meets the head, so the mealiness runs out instead of stopping at the box edge. It runs after the face ramp, so it takes the muzzle back off it — an ordinary bay’s muzzle is black and a seal’s is tan.
• Over the eye. The forehead-centred falloff dun’s face mask already uses, at SOFT_EYE_REACH.
• Elbow and stifle. Two broad, low, overlapping fields on the barrel — deliberately broad: drawn small they read as a pair of orange freckles rather than as a light underside.
• The inner limb, added with CoatRegions.medial. The outside of a seal brown’s legs is black; it is the inner surface of the thigh and forearm that keeps the tan. It went undrawn for a while because “the inside of a leg” is a face of a part rather than a part, which nothing here could address.

The five draws

double spread = BayShade.spread(ctx.genotype(), epi);   // 1 - the expression roll
double leg    = legHeight(spread);

for (int i = 0; i < legs.length; i++) {                 // 4 - one jitter per leg
    legs[i] = leg * (1.0 - LEG_JITTER + epi.nextFloat() * LEG_JITTER * 2.0);
}

The order is a contract: the golden coats and every “adding a gene shifts the stream by this much” claim lean on it, and BayShadeGeneTest pins the count.

The fade, and the hoof floor

The bottom SOLID_PORTION of each leg band is solid black; the rest is a smoothstep fade to nothing. Smoothstep has a flat slope at both ends, so the black dissolves into the body colour with no visible cut-off line. (Splash’s socks do have a visible cut-off line, and that is the open issue there.)

Bay sets pigment absolutely — and why that matters

BayCoat paints its points absolutely: setBlack(1.0), setRed(0.0), via CoatRegions.blackenPart. Agouti runs before cream, pearl and champagne in Genes.naturalOrder(), so the dilutions do run after and do scale the points.

What broke, once, was the gradient, not the ordering. With red = 0 the gradient’s zero-red column stays visually jet black all the way down to black ≈ 0.4, so scaling black alone moved the sample without changing the colour: single cream’s black *= 0.7 landed on #111111, double pearl on #272727. Every “diluted” bay still had jet points.

The fix was PigmentField.dilute, which feeds a fraction of the removed eumelanin back in as pheomelanin and walks the sample sideways off that column into the warm browns. The ordering and the absolute bay point both stayed as they were.

The soft points are not points in that sense — they end with red ≈ 1, so the black × (1 - red) test a dilution uses to spare a point reads them as body coat, and dilutes them. That is correct: a seal brown’s mealy muzzle is body colour, not a black point, and a silver or dun seal should dilute it.

Interactions

• Invisible on a chestnut. No black to restrict — a chestnut can carry A, and a whole shade genotype with it, silently for generations.
• The dilutions decide the colour of the points; agouti still decides where they are — and shade decides how much body is left for the dilution to work on, so a cream over a liver bay and a cream over a blood bay are different buckskins.
• Grey reads bay’s output like any other pigment, and its point-retention knob keeps mane, tail and lower legs darker on the least-greyed horses — which reads as a young grey that used to be a bay.
• Seal brown and black are meant to be confusable. They are in real life; the muzzle and the elbow are what tell them apart, in the game as in a field.

Knobs

Values live in BayCoat and BayShade and are deliberately not repeated here — a number written into prose goes stale the next time the code moves. What each one means:

Why four outcomes and not one

Blood bay, bay, liver bay and seal brown are phenotype descriptions, not loci. Every one of them tests as the same foundation — one functional MC1R copy and one functional ASIP copy, E_A_ — and no validated equine allele separates them. What separates them in a real horse is a regulatory region, a couple of dosage effects and a polygenic tail, and what separates them here is the same list, summed into one shade score.

So agouti declares four bay Expressions and picks between them in expressionIn, by reading the rest of the genotype. Its pair-only expressionOf answers bay for any A horse and admits it is coarse — the same position the leopard complex is in, and the reason both genes list their combinations in the preview window rather than their outcomes.