Car materials and shaders

Car materials use the same material parameter format as the world, with MTPR v2.0 (a u8 parameter count) followed by a 12-byte trailer (u32, u32, u32 0) on all 1,859 files of Materials.zip. Parent chains end at game:\Media\cars\_library\shaders\<family>\<family>.shaderbin. Observed Each family’s bundle (Cars\_library\shaders\<family>.zip) holds <family>.shaderbin plus DXIL per variant (_Standard\, _DXRSimpleHit_Base\, or at the root when the family has one) and scenario (CarLightScenario = the forward main view, CarShadowDepthLightScenario, CarRTBufferLightScenario, …). Unlike the world families, car shaders use classic bound registers (Shader containers). Input element ids follow the signature’s interpolation table (car_standard: 0 to 3 TEXCOORD0..3, 4 WORLDPOS, 5 WORLDNORMAL, 8 to 10 WORLDTANGENT0..2, 11 COLOR, 12 DAMAGEIMPACT, 13 DAMAGESCRAPE). Observed

Each family’s CarLightScenario pixel shader was disassembled and its data flow printed back to material parameters, textures and interpolants; the lighting inputs were found by their signature in the shared lighting code (F0 = m(albedo − 0.04) + 0.04, the roughness floor max(r, 0.01), the derivative-based normal variance on the final normal). Where shaders were run, a DXIL interpreter reproduced the formulas: the tyre family over 234 runs on 14 tyre models agrees to 2.6e-6 in albedo, roughness, F0, normal, AO and DO. Observed The counts come from a recorded census of the install’s 682 cars; the install is no longer available. Observed

Run-time layers that are the identity on a clean car

On a clean, dry, undamaged, stationary, exterior-view car with lamps off these layers do nothing: damage (scratches use the DAMAGEIMPACT and DAMAGESCRAPE interpolants, zero when undamaged), SurfaceFX dirt, mud, snow and wetness (both blocks run only when a run-time amount in cb4 exceeds 0.05), rain streaks, and lamp emission (Headlight_Color × brightness × per-part light masks × run-time exposure). Observed Every lamp emission term is multiplied by a run-time light-group value cb4[Emissive_Trigger >> 2][Emissive_Trigger & 3] (0 for triggers above 79). Observed The quality switches cb4[43].x, .y and .z (coated orange-peel coat normal; carbon coat normal and fibre anisotropy; the back normal of detailed glass; cb4[43].w gates a dirt term) are run-time values. Unknown

car_standard

190 of the 240SX’s 306 materials (car_standardCarLightScenario.pcdxil.pso):

uv     = UVChoice_OnCh1_OffCh2 ? TEXCOORD0 : TEXCOORD1           (the tangent frame follows: WORLDTANGENT0 / 1)
uv     = rotate(uv, UV_Orientation degrees): u' = cos a·u − sin a·v, v' = sin a·u + cos a·v   (D3D uv; the tangent frame is rotated about N by the same angle)
uvX    = uv · (Override<X>TilingOnOff ? <X>TilingOverride : (U_Tiling, V_Tiling))   per texture X
base   = BaseColorAlpha(uvB)                       (SampleBias −2)
rma    = RoughMetalAO(uvR)                          r = roughness, g = metalness, b = AO
tint   = BaseColor_Tint · BaseColor_TintMultiplier
t      = sat(tint · base)
w      = m + (1 − 2m)·BaseColor_TintMode.y          m = rma.g
col    = t + BaseColor_TintMode.x·(1 − w)·(base − t)
albedo = min(max(col, PBRValidation_Min), PBRValidation_Dielectric_Max + (1 − Dielectric_Max)·m)
rough  = s == 0 ? r : s > 0 ? 1 − sat(1 − s)(1 − r) : sat(1 + s)·r          s = Roughness_Shift
n      : x = 2N.r − 1, y = 2N.g − 1, z = sqrt(sat(1 − x² − y²)) (from the unscaled x, y),
         n = normalize(I·x·T − I·y·B + z·N), I = Normal_Intensity, B = cross(N, T)·T.w
AO     = (1 + sat(I)(rma.b − 1))·AoScale        (with the instance AO bake on TEXCOORD2: occlusion terms)
alpha  = Alpha.r(uvA)·base.a; the ShadowDepth pass discards alpha < 0.5 when AlphaTransparency
lighting: diffuse = albedo(1 − m); F0 = lerp(0.04, albedo, m); GGX with alpha = rough²
          (sun lobe: alpha = sat(1.8(1 − (1 − 0.5556·rough²)(1 − v))), v = the screen-space normal variance, 0 on smooth normals)

All Observed. The cut-out threshold 0.5 is read from the shadow pass; that the main view uses the same cut-out is not traced. Unknown

Relatives

• car_standard_emissive: car_standard albedo, roughness and normal, with the dielectric maximum fixed at 0.871367 (hi = 0.871367 + 0.128633·m); no alpha, no SurfaceFX; emission (EmissiveCH1 desaturated by EmissiveColor_Desat × EmissiveColor_Tint × Emissive_Intensity, anisotropic falloff) times the run-time light group. Observed
• car_standard_fabric: tint' = TopStitchSwitch ? lerp(tint, StitchColor_Tint, sat(20·rma.g)) : tint; col = base·tint'; rim = lum(col)^0.2 · RimColorAddAmount · sat(sat((N·V − 0.7)·−1.42857)^(4 − 3·FabricFuzziness)), lum weights (0.2126, 0.7152, 0.0722); albedo = clamp(col + rim, PBRValidation_Min, 0.871367 + 0.128633·rma.g); metal = TopStitchSwitch ? 0 : rma.g. Observed
• car_label: albedo = clamp(BaseColorOverrideSwitch ? BaseColor_Override : base, PBRValidation_Min, 0.871367 + 0.128633·Metalness), metal sat(Metalness), roughness sat(Roughness); coverage A = (UseAlphaBlend ? a : 1)(UseAlphaTest ? a : 1) written as dual-source output (lit·A, 1 − A). Observed
• car_livery: albedo = clamp(BaseColorTextureSwitch ? BaseColorAlpha(uv0) : BaseColor, PBRValidation_Min, 0.871367 + 0.128633·m); normal OrangePeelNormal_Texture(uv1·32) with intensity OrangePeelNormal_HeightScale on the UV1 tangent frame. Observed
• car_brakerotor: uv = (UV_Scaler(u − 0.5)(UV_Flip ? −1 : 1) + 0.5, UV_Scaler(v − 0.5) + 0.5) on TEXCOORD0; normal on the same uv but with the WORLDTANGENT1 frame; rotor glow (RotorGlowSwitch × run-time brake temperature) is run time. Observed

car_automotive_paint

Stock path (UserColorOverride and UniqueLiverySwitch off), _Standard:

f      = smoothstep(sat(Ngeo·V))^GlancingFlopPower          (Ngeo = interpolated WORLDNORMAL)
paint  = ColorFlopSwitch ? lerp(GlancingFlopColor, PaintColor, f) : PaintColor    (flake: the same with FlakeColor)
cov    = FlakeCoverage; cc = ClearCoatCoverage; rc = ClearCoatRoughness
diffuse lobe  : Lambert, paint·(1 − cov)
flake lobe    : GGX, F0 = cov·flake, alpha = max(FlakeRoughness², rc²)
                normal n_f: X = −(I·Nx + k·Gx), Y = I·Ny + k·Gy, z = sqrt(sat(1 − Nx² − Ny²)); n = X·T + Y·B + z·N
                (Normal(uv0), I = Normal_Intensity, frame = WORLDTANGENT1 with B = cross(N, T)·T.w)
                glitter G = lerp(N00.xy, N0.xy, N00.w)(2v − 1), N00 = NormalMap00Texture(uvg·UVTiling@0x9A7DB1FA),
                N0 = NormalMap0Texture(uvg·UVTiling@0xFD6BB566), uvg = uv1 rotated about (1, 0.97):
                u' = (u − 1)·0.594079 + 1 − (v − 0.97)·0.804407, v' = (u − 1)·0.804407 + 0.97 + (v − 0.97)·0.594079
                k = sat((FlakeRoughness − 0.75)·−4)·(0.0125·smoothstep(sat(Ngeo·V)) + 0.0125)·GlitterIntensity
base lobes × (1 − F(Nc·V)·cc) × ((1 − cc) + cc·ClearCoatTint·T), T ~ (1 − F(V·H)) (energy term)
coat lobe     : GGX, F = 0.04 + 0.96(1 − V·H)^5, alpha = rc², weight cc, normal n_c:
                X = −(I·Nx + s·Ox), Y = I·Ny + s·Oy (O = OrangePeelNormal(uv1·32), 2v − 1),
                s = lerp(GlancingAngleStrength, OrangePeelStrength, sat((sat(Ngeo·V) − 0.4)·2.85714))
                (the coat normal is lerp(n_c, Ngeo, wetness), wetness 0 when dry)
the base lobes see L and V bent towards the normal by (1 − 0.37·cc) (coat refraction)

The glitter and orange-peel perturbations are added only when the run-time quality switch cb4[43].y is non-zero (DXIL: both are zeroed otherwise). Observed With UserColorOverride the paint and flake colours are both UserColor0; with UniqueLiverySwitch the livery texture BaseColorAlpha replaces them (MaskedLivery: masked by sat((Alpha.r·base.a − 0.5)·100)); with UserLiverySwitch the run-time livery render target (premultiplied) is blended over by its alpha (Liveries). Observed

Glass group

car_glass, car_glass_detailed, car_window, car_window_limotint, car_lens, car_livery_transmissive, car_tinthack. All write two targets: SV_Target0 the lit colour and SV_Target1 a transmittance (rgb) the background is multiplied by, i.e. dual-source blending dst = src0 + dst·src1 (the blend state itself is run time). Common terms: Observed

NoV   = sat(dot(N, V))
a'    = 1 − max(1e-4, sat(1.8(1 − (1 − 0.555556·r²)·v)))       v = screen-space normal variance; v = 1 → 1 − max(1e-4, sat(r²))
F_k   = (0.05 − k·a' + 0.95·a'^6)(1 − NoV)^5
Tif_k = 1 − k·a' − F_k                                          one interface's transmission weight

• car_glass (lamp bulbs): tint = GlassColor, or with IridescenceSwitch lerp(GC, (1 + (1 − sat(d/1500))(|cos(d·k)| − 1))·GC, GlassIridescenceStrength), d = GlassThickness/dot(N, V), k = (0.00448799, 0.00615999, 0.00713998); Lambert albedo tint weighted by Opacity; specular roughness Roughness; normal geometric; T = Tif_0.06² (1 − Opacity)·sat(tint). Observed
• car_window, car_window_limotint (identical except limotint lacks the interior tint step): colour = GlassColor (× max(UserColorMultiply, 0.1022417) with PlayerColorSwitch; the interior camera cb11[3].w ≠ 0 applies lerp(colour, 1, InteriorTintReduction)); opacity o = 0.005 (exterior, dry: the shader’s constant); roughness 0.2·sat((d − 0.02)·1.020408), d = max(0.03·x, 0.005), x = WipersEnabled_Part2 ? ShatterMap.w·0.5 : sat(ShatterMap.w·0.5 + ShatterMap.y) (ShatterMap at TEXCOORD1·0.6); normal geometric; T = Tif_0.06²(1 − o)·sat(colour). Observed
• car_glass_detailed (lamp lenses with a modelled back face): front interface on the geometric normal with roughness rF = RoughnessMask ? RoughMetalAO.r(TEXCOORD0) : Roughness; back interface on the BackNormal map (uv UVChoice_OnCh1_OffCh2 ? TEXCOORD0 : rotate(TEXCOORD1, UV_Orientation) × (U_Tiling, V_Tiling), intensity BackNormal_Intensity, used when the run-time switch cb4[43].z ≠ 0) with rB; Tf = Tif_0.04(rF, N)·Tif_0.04(rB, Nback); o = Opacity, or with DualOpacityOnOff lerp(Opacity, OpacityB, sat((sat(max(|bn.x·S|, |bn.y·S|)·I) − Min)/(Max − Min))) with S = BentAngle.x + .y + .z; coverage A = AlphaTransparency ? Alpha.r(TEXCOORD0) : 1; T = 1 + (Tf(1 − o)·sat(tint) − 1)·A. Observed
• car_lens: SV_Target1 = 1 (1 − cb4[28].w when fading): the lens only adds light. Specular roughness 0 (alpha = sat(1.8(1 − v))), F0 = 0.00608 + 0.281536·r, diffuse albedo 0.0160632·r, r = BaseColor_A_Texture.r at the parallax-offset uv; emission EmissiveCH1 × EmissiveMaskCH1 × EmissiveColor_Tint × Emissive_Intensity × the run-time light group. WarpingFX parallax (one step): V = normalize(−WORLDPOS), N = normalize(WORLDNORMAL), T = normalize(WORLDTANGENT0.xyz), B = cross(N, T)·WORLDTANGENT0.w, s = sat(WarpingFX_intensity·dot(V, N))·3.074 (0x4008978D40000000), uv' = TEXCOORD0 + s·(dot(V, T), dot(V, B)). An evaluator of the DXIL instructions (float32) on the inputs at 33 samples (3 points × 3 view directions on each of the Supra’s two parallax lens materials, both headlights, offsets up to 0.43 in u) gives a maximum difference of 2.2e-7. Observed
• car_livery_transmissive: premultiplied alpha a = Opacity·sat((Alpha.r·BaseColorAlpha.a − 0.5)·100), SV_Target0 = lit × a, SV_Target1 = 1 − a; albedo min(max(BaseColorTextureSwitch ? BaseColorAlpha : BaseColor, PBRValidation_Min), 0.871367), roughness sat(Roughness), no metal. Observed
• car_tinthack: a tint film: c = ColorListSwitch ? ColorList : Color, coverage = alpha test, diffuse albedo 0.3·c, roughness 1, `T = 1 − a
  • a·0.7(1 − 0.05(1 − NoV)^5)²·sat(c)`. Observed

Clean-state values of the glass group (all run-time): unbroken (b = sat((sat((cb4[42].y·cb8[0].x − 0.5)·1000)(BreakOut ? BreakMap.r − 1 : −1) + 0.639)·2.94985) = 1, so the broken-glass roughness 0.063, opacity 0.211 and bent normal drop out), no fog (the cb3[53..54] and cb2[111..112] height-fog factor 0), no livery layer, exterior view, fully faded in (cb4[28].w = 1), lamps off. Unknown

Coat and special group

car_standard_coated, car_reflector, car_carbonfiber, retro_licenseplate_base, retro_licenseplate_atlas, car_metal_colorshift, car_mirror, car_blackhole, car_wheelblur, car_digitalscreen, car_nodraw, diffonly, car_interiorlod1, car_rotatingfan. Shared terms: Observed

GGX alpha    a(r) = sat(1.8(1 − (1 − 0.555556·r²)·v))      v = screen-space normal variance; v = 1 → a = r²
normal       x = 2t.r − 1, y = 2t.g − 1, z = sqrt(sat(1 − x² − y²)) (unscaled x, y); n = normalize(I·x·T − I·y·B + z·N), B = cross(N, T)·w
shift(r, s)  s > 0: 1 − sat(1 − s)(1 − r);  s < 0: sat(1 + s)·r;  s = 0: r                (Roughness_Shift)
tint(b, m)   t = sat(tint·b); w = m + (1 − 2m)·TintMode.y; col = t + TintMode.x·(1 − w)(b − t);  tint = BaseColor_Tint·BaseColor_TintMultiplier
pbr(c, m)    min(max(c, PBRValidation_Min), dm + (1 − dm)·m), dm = 0.871367 where the family has no PBRValidation_Dielectric_Max parameter
F0           m(albedo − 0.04) + 0.04, diffuse albedo (1 − m)/π
std UV       uv = rot(UVChoice_OnCh1_OffCh2 ? TEXCOORD0 : TEXCOORD1, UV_Orientation)·(U_Tiling, V_Tiling), each texture's tiling replaced by its *TilingOverride

Coat lighting (one function, identical in car_standard_coated, car_reflector and both retro plates; car_carbonfiber has its own with the same coat terms): coat Fresnel 0.02 + 0.98(1 − VoH)^5 (F0 0.02, i.e. index 1.3294313), scaled by the mask, on the coat normal with GGX alpha ac = max(0.001, a(cr)·mask); the base specular lobe uses alpha max(ac, a(r)): the base is never smoother than the coat, and never below alpha 0.001 even with mask 0. The coat normal is geometric unless the quality switch supplies one. Observed

Family	What it writes	Confidence
car_standard_coated	car_standard inputs on the std UV; coat mask ClearCoatMask, coat roughness ClearCoatRoughness; with the quality switch the coat normal is the orange-peel map OrangePeel_Texture(TEXCOORD1·64), (r − 0.5, 0.5 − g, z) at strength 0.5 on WORLDTANGENT1; cut-out in ShadowDepth AlphaTransparency ? Alpha.r(uv_Alpha)·BaseColorAlpha.a : 1 < 0.5	Observed
car_reflector	lamp reflector bowls: albedo BaseColor, metal 1, roughness Roughness (+ CracksOnOff × run-time break mask), coat mask 0; normal Normal(rot(TEXCOORD1, UV_Orientation)·(U_Tiling, V_Tiling))	Observed
car_carbonfiber	uvW = rot(TEXCOORD1, UV_Orientation)·(U_Tiling, V_Tiling); weave = lerp(WeaveColorTintA, WeaveColorTintB, WeaveMask.x(uvW)); mask = sat((Alpha.r·BaseColorAlpha.a − 0.5)·100); L = UniqueLiverySwitch ? (MaskedLivery ? mask : 1) : 0; albedo lerp(weave, clamp(BaseColorAlpha.rgb, PBRValidation_Min0, PBRValidation_Dielectric_Max0), L); base normal blends the plain normal with WeaveNormal(uvW) by whiteout; coat weight ClearCoatCoverage·(1 − M·L), coat F0 0.02; with the quality switch the coat has a normal from ClearCoatNormal_Texture and the base lobe is anisotropic: alpha_T = max((1 + k)·a, 0.001) along t, alpha_B = max((1 − k)·a, 0.001), k = 1 − sqrt(1 − A²), `A = (1 − wet)(1 − L)·AnisoStrength·sat((	WORLDPOS
retro_licenseplate_base	RetroTextureSwitch off: albedo RetroDiffuse, F0c = RetroF0; on: t = CH1DiffuseTexture(TEXCOORD0), F0c = lerp(0.333·t.rgb, RetroF0, t.a); metal = luminance of F0c, roughness 0.15, coat 0; cut-out CH1OpacitySwitch ? sat((CH1NormalMap.a − 0.3)·2.5) : 1 < 0.5	Observed
retro_licenseplate_atlas	plate characters: uvA = TEXCOORD0 + (frac(i/8), trunc(i/8)/8), i = cb4[55 + ((c >> 2) & 1)][c & 3], c = int(TEXCOORD3.x·8): the characters are run-time constants	Observed; characters Unknown
car_metal_colorshift	car_standard with the tint multiplied per channel by `Reflected_RTint(Reflected_RTint_UVTiling·(R.B0, R.N0)/	(R.T0, R.B0, R.N0)
car_mirror	the planar reflection target is used only when cb4[29].x ≠ 0 && cb2[48].y == 0 (run time); fallback albedo 0.871367, metal 1, roughness 0	Observed
car_blackhole	no parameters; folded constant: diffuse 0.000483078 → albedo 0.0015176, metal 0, roughness 1	Observed
diffonly	albedo path(TEXCOORD0).rgb, metal 0, roughness 1	Observed
car_digitalscreen	folded albedo 0.0030353, roughness 0.22, metal 0; emission (StaticEmissive × cb2[99].z, screen and backup-camera render targets × cb2[60].y, PowerSwitch) is run time	Observed
car_interiorlod1	t = DiffuseMetalnessMap(TEXCOORD0); albedo 0.6·lerp(t.rgb, (0.549585, 0.556114, 0.554256), t.a), metal sat(t.a), roughness sat(0.7 − 0.3·t.a)	Observed
car_rotatingfan	uv rotated about UVCenterOfRotation by time·deg·(sat(2·cb?)·AdditionalSpinWithTireVelocity + TimeScale)·(light parameter > 79 ? light group : 0) (run time), then car_standard inputs without tiling	Observed
car_wheelblur	coverage sat(0.4(1 − min(cb4[42].z, cb8[0].z)) + blurRT.a …) from a run-time wheel-blur render	Observed
car_nodraw	colour 0 and transmittance 1: invisible	Observed

car_tires_pg

Two variants selected by the material’s int Legacy: _Standard_L (LegacyHeightAO, LegacyNormal, LegacyGlobalBase, LegacyGlobalWear) and _Standard (UniqueNorm, UnqiueMasks, GlobalBase and three tyre-line colours). All 197 tyre models use them (157 + 70 legacy sidewall materials, 116 + 36 standard sidewall, 122 legacy tread, 94 standard tread; SidewallOnOff on every sidewall, TextSwitch on 106 sidewalls); the only tyre material placed by a car file is CHE_BelAir_57’s spare (tireSidewall_vintage). D3D uv: Observed

uvt   = TEXCOORD0·UniqueMapTiling
gate  = sat(10(1 − |n_obj.x|))         n_obj = normalize(cb9[4..7] rows · N): the normal in the per-object frame, x = axle
k     = gate·ParallaxStrength·(h0 − c)       legacy: h0 = LegacyHeightAO.x(uvt), c = 0.8, ParallaxStrength@0x8780443B
                                             standard: h0 = UniqueNorm.w(uvt), c = 0.5, ParallaxStrength@0x5C6763CB
uvp   = uvt + k·(V·T, V·B)              V = normalize(−WORLDPOS), T = normalize(WORLDTANGENT0), B = cross(N, T)·w (unnormalised)
proj  = SidewallOnOff ? (dot(cb9[6].xyz, Pw) + cb9[6].w + 0.5, dot(cb9[5].xyz, Pw) + cb9[5].w + 0.5) : uvt        Pw = cb2[8] + WORLDPOS
sws   = SidewallOnOff ? SideWallScalars : (1, 1, 1, 1)
base  = TextSwitch ? path@0x33A205F2(TextUVTiling.x·u0, TextUVTiling.y·sat((TextSlideAmount + v0)·16/min(16, cb4[61][i]))) : BaseTyreColour   (sampler s12: U wrap, V clamp)
legacy   albedo = lerp(base, LegacyGlobalBase(proj).rgb, LegacyHeightAO(uvp).x·sws.x·LegacyGlobalWear(proj).x)
         rough  = sat(LegacyGlobalBase(proj).w·sws.y·UnwornRoughnessMultiplier)
         normal map LegacyNormal(uvp);  ao_in = LegacyHeightAO(uvp).y·sws.w
standard albedo = lerp(base, GlobalBase(proj).rgb, UnqiueMasks(uvp).x·sws.x), then towards TireLine1/2/3Color by
                  g·(1 − sat(|m − c|·14.285714)) for c = 0.25 / 0.5 / 0.75, m = UnqiueMasks(uvp).z·sws.z,
                  g = sat(10 − 20|v0 − 0.5|)·sat((wear − 0.06 + (SurfaceFX(uv2.x, 4·uv2.y).y − 0.5)·0.05)·(−50))
         rough  = sat(UnqiueMasks(uvp).y·sws.y·UnwornRoughnessMultiplier)
         normal map UniqueNorm(uvp);  ao_in = UnqiueMasks(uvp).w·sws.w
n     = normalize(x·T − y·normalize(B) + z·N), x, y = 2t.rg − 1, z = sqrt(sat(1 − x² − y²))
lighting: Lambert albedo/π, dielectric GGX F0 = 0.04, alpha = rough²
AO    = sat(CH3AOTexture(uv2.x, uv2.y − cb4[47][i]).x·ao_in), DO = sat((that + DOFromAOOffset)^DOFromAOPow)

i is the wheel index cb11[0].x (0 to 3). Clean state: new tyre (per-wheel wear cb4[67][i] = 0), stationary (per-wheel blur cb4[31+i].w = 0), dry and clean (cb4[21,28,42,46], cb2[70].w = 0), exterior view, fully faded in. Observed The per-object matrix cb9[4..7] (rows 4 to 7 of the per-draw pose buffer, whose rows 0 to 2 are the world matrix the vertex shader uses and rows 12 to 14 the previous frame’s) is read by no shader other than the tyre pixel shader and no readable file names it; its use fits the inverse of the object’s world matrix, i.e. the mesh’s own object space, which is inferred. Inferred The run-time surface map (bindless space 43 index 9, sampled at TEXCOORD1, its normal blended in by sat((N.y + 0.3)·5.6958)) has a neutral value (0.5, 0.5, −, 0). Unknown

Paint, slots and missing textures

Paint colour comes from the factory paint list or a livery; the paint type values 3 and −1 of a livery slot (Materials.zip painttypes\*.materialbin use another shader, carpaint_user) have no readable index → file table. Unknown 289 references of 261 materials on 129 cars are textures the install does not ship and are read outside any quality switch; 396 references to 215 missing paths were counted in all (see Swatch textures). In car_label the missing Normal sits behind cb4[43].y and in car_glass_detailed the missing BackNormal behind cb4[43].z (traced with a forward slice of the DXIL: the texture sample feeds only a quality-switch phi). Observed

What do the run-time quality switches and light groups hold?

cb4[43].x/.y/.z, the light-group table indexed by Emissive_Trigger and the per-wheel tyre rows are filled by CPU code each frame; their values are in no file.