Surface formulas and texture roles

Material templates define the meaning of texture channels. A texture labelled Normal can pack metalness and roughness, and a diffuse alpha can mean height rather than opacity. Executing the original pixel programs under Direct3D 11 WARP verifies the formulas below for the stated permutations. Proven · executed families

Let sat(x)=clamp(x,0,1) and S(x)=x*x*(3-2*x). Material textures arrive in the shader’s sampling colour space; constants and mask channels follow the particular program’s arithmetic. Proven · executed programs

Normal frame

For the Fake, Tex1 and Tex2 families, decode x=2R-1, y=2G-1, z=sqrt(max(0,1-x*x-y*y)). The world normal is normalize(x*T + y*cross(N,T)*w + z*N). Blue is not reconstructed Z: it supplies metalness in these packed maps. Random-frame execution agrees within 1-cos(angle) = 1e-5. Proven · Fake, Tex1 and Tex2

Fake and Tex1

Input	MAD_Fake	MAD_Tex1	Confidence
Texture bindings	t75 albedo, t76 normal/surface	t75 normal, t76 diffuse, t77 mask	Proven
Base colour	albedo RGB	HSV-tinted diffuse, tint weight mask R; multiplied by 1+VCIntensity*(COLOR0.rgb-1)	Proven
Roughness	normal A	S(sat(normal.A+RoughnessOffset))	Proven
Metalness	normal B	normal.B*Metalness	Proven
AO	1	1+OnOffAO*(mask.B-1)	Proven
Alpha test	albedo A below cb2[18].y discards	diffuse A below cb2[18].y discards in the G-buffer path	Proven

The Tex1 formula matches 40 random colour/surface cases to at most 5.1e-7; Fake’s channel assignment and seven alpha-test cases match exactly. The Fake program negates the geometric normal on back faces and consequently the bitangent, without flipping w. Proven · executed cases

HSV tint adds Hue, Saturation and Luminosity to HSV coordinates, saturating saturation and value. The hue fraction retains a negative sign: h' = sign(h+H)*frac(abs(h+H)), rather than wrapping negative results into [0,1). Ordinary wrapped HSV differs by up to 0.06 in the random checks. Proven · Tex1 execution

Tex1_Wind and Decals1 use different binding tables from Tex1. Tex1_Wind binds normal t75, diffuse t76 and mask t77, with Hue/Saturation/Luminosity in cb3[14–16], AO at 17, translucency at 18 and UV rows at 19–20. Its roughness and metalness are raw Normal.A/B, with no Tex1 offset curve, metalness scalar or vertex-colour multiplier. Forty WARP cases with random decoy values in wind constants cb3[1–13] match colour and surface formulas within 5.8e-7; those earlier constants do not affect the pixel outputs. Proven · executed Tex1_Wind

Decals1 binds normal t75 and diffuse t76 at the mesh UV directly. It applies HSV tint without a mask, uses S(sat(Normal.A+RoughnessOffset)) and Normal.B*Metalness, and AO 1. Its largest material buffer ends at cb3[5]: it has no Tex1 UV rows, AO switch or vertex-colour factor. Forty WARP cases match within 5.2e-7. Proven · executed Decals1

Tex1_Emissive uses an emissive mask and scale, and an Alpha scalar. The traced emissive output is mask red times scale divided by t23. Observed · template bindings and pixel data flow

Tex2 height blending

MAD_Tex2 binds A as diffuse t78, normal t75, mask t80, and B as diffuse t76, normal t77, mask t79. Each layer has its own UV transform and HSV tint. Proven · executed opaque permutation

The blend is defined by the following equations. Diffuse alpha is height, not opacity. Fifty random cases match colour, roughness, metalness and AO within 5.5e-7, and the decoded normal within 1-cos(angle)=6.6e-6. Proven

w  = sat((COLOR0.a-.5)*AlphaContrast + .5 + AlphaContrast*AlphaOffset)
m  = max(w*heightB, (1-w)*heightA)
wB = max(w*heightB - (m-(1-HeightContrast)), 0)
wA = max((1-w)*heightA - (m-(1-HeightContrast)), 0)
qA = wA/(wA+wB); qB = wB/(wA+wB)
albedo = qA*tintedA + qB*tintedB
albedo *= 1+VCIntensity*(COLOR0.rgb-1)
roughness = qA*S(sat((normalA.a+offsetA-.1)*1.25))
          + qB*S(sat((normalB.a+offsetB-.1)*1.25))
metalness = qA*normalA.b + qB*normalB.b
normal = normalize(qA*decode(normalA) + qB*decode(normalB))
AO = qA*(1+OnOffAO*(maskA.b-1)) + qB*(1+OnOffAO*(maskB.b-1))

MAD_Tex2_Graffiti_Mode differs: roughness is the weighted raw normal-map alpha, there is no RGB vertex-colour multiplier, and a runtime state selects palette replacements through mask A’s red channel. Execution covers states 0–7 and 9 over 2,304 pixels. MAD_Tex2_TintBFromShaderConstants uses runtime Neutral constants for B’s tint. Proven · executed placed permutations

Character surfaces

Family	Established channel or formula	Confidence
Classic	BaseColor RGB → colour; Normal RG → frame, B → metalness, A → roughness; Mask B → emission; decal alpha blends a second colour layer	Observed · 96 traced four-output permutations
Classic VWIND Details	Detail and decal composition is executed with the template’s own constants and maps	Proven · executed permutation
5Zones	Mask1 R chooses zones at .25/.75; G overrides to zones 4/5 at .25/.75; zone constants supply colour and metalness; Mask1 A controls variance; Mask2 RG gives normals and A roughness	Proven · executed zone families
5Zones Motifs / Details	Motifs and details add layers to zone colours; detail variants include the mud layer	Proven · executed permutations
Classic SkiSnowboard / Skateboard	Colour and normal layers, roughness, metalness and emission follow their own binding tables	Proven · executed permutations
Skin / Skin_CIN / Skin_LOD	Skin-specific colour and surface paths are traced; a complete execution-backed formula is not established here	Observed · shader trace
Glass_Dirt	Dirt/glass channel paths are traced; full lighting and transmission are not executed	Observed · shader trace
Hair_Anisotropic	Channel dependencies are traced; the complete anisotropic lighting response remains unresolved	Observed · shader trace

Emission and colour space

The material G-buffer’s emissive scalar is divided by the exposure texture t23 in the executed emissive families. The ambient lighting path multiplies it by albedo, so a scalar alone is not an absolute radiance or a directly portable luminous intensity. Proven · executed emissive paths Observed · lighting consumer disassembly

TextureMap’s transfer flag, not a role guessed from its label, selects colour-space treatment. Observed · runtime resource-format comparisons In the complete join of all 35,918 stored texture properties, some normal-labelled BC7 maps carry flag 1 and some albedo-labelled maps carry flag 0. Their exceptions are stored data, not evidence that every normal is linear or every diffuse is sRGB. Proven · all 35,918 texture-property joins

See specialized materials for world-space, ice, screen and procedural families, and G-buffer classes for the different meanings of the output channels.

Which runtime values complete the material response?

Exposure, team values, camera-dependent ice colour, null texture fallback and the runtime vertex inputs cannot all be recovered from a Material body alone. The formulas above establish program behaviour for explicit inputs, not a capture of a complete game draw. Unknown