Other world shader families

This page gives what each remaining world family writes to the G-buffer. Environment, rocks and animated props, spline deform and water have their own pages; cars are on Car materials. Each formula was transcribed from the family’s GBufferLightScenario pixel shader, and those marked [V] were reproduced by interpreting the compiled shader on real Tokyo materials (real merged constants, real textures, random points on real faces with the mesh’s own normal and tangent), with a maximum difference of 1e-7 over 49 runs. Observed Counts come from a recorded census of the Tokyo track Brio, no longer re-checkable. Observed

“Overlay” below is always a > .5 ? 1 − 2(1−a)(1−b) : 2ab. The alpha-test threshold is a global (cb2[128].x) found in no file. Unknown

Albedo and opacity

Family	Albedo (RT0.rgb)	Opacity	Confidence
uber_foliage	DiffuseTexture(uv0) (0x0940E415); _Tint adds a world tint map and LUT overlay (one material)	alpha test sat((a − AlphaStepMin)/(AlphaStepMax − AlphaStepMin))	Observed
uber_branch	DiffuseA(uv0 · DiffuseAUVTiling)	opaque	Observed
environmentdeferreddecal	DiffAlphaTex((uv0 + Uv0Offset) · DiffAlphaUVTiling) (0x0670EE0A); _ST swatch tint as the environment family; offset and tiling select an atlas cell (94 of 424 materials)	the deferred-decal pass: alpha a · DiffRoughIntensity · (VertexAlphaOnOff ? COLOR.a : 1), discard below 0.01	Observed
environment_plot	D = DiffAlphaTex_A((uv0 + Uv0Offset) · BaseColorAlpha_A_UVTiling); _M: M = Modulate(uv2 · ModulateUVTiling) (PavementCorner texture where uv3.x ≥ .999), b = sat((M.r − .5)·_BrightnessStrength + .5), s likewise with .g, c1 = overlay(b, D^(1/2.2)), L = luma(c1), mod = sat(L + (c1 − L)·2s)^2.2, albedo = D + (mod − D)(1 − RMA.y) (grey128 = identity)	AlphaTest variants: D.a	Observed
uber_roaddecal, uber_roaddecalretro	DecalDiffuseATexture(uv0 · RoadUVMultiplierA) (0xFCA19F4A); PerObjectTint0OnOff: overlay with ObjectTintColor^(1/2.2); T variants overlay with COLOR.b; modulate variants multiply by 1 + (Modulate(uvM).w − 1)·sat(DecalNormalA.w + 1 − HeightInfluence)·COLOR.r	alpha test on DecalMaskATexture(uv0).r (0xBB3F3690, raw uv0) × (AlphaFadeOnOff ? COLOR.g : 1); HA variants sat((mask + fade − 1)/MaxHeightAlpha). The diffuse alpha is not opacity	Observed
uber_road	layers A to D (0x83D46B62, 0x92A9011B, 0x2B52DAF3, 0xB053D5E9; normal alpha = height) on world-planar uv when UVSet < 0 (u = X·Tiling.x·WorldUVScale.x, v = −Z·Tiling.y·WorldUVScale.y), else uv0 or uv1; height blend with BlendTexture(uv2) channels and BlendSmoothness; then overlay(mix, TintTexture(uv2).rgb); _modnorm × (1 + (Mod.w − 1)·Tint.a)	opaque	Observed
anim_flag	DiffuseATexture(uv0 + DiffuseUVOffset) (the offset picks a flag in an atlas)	AlphaTest variants: a	Observed
projection_lod	uv = lerp(UVOutMinMax.xy, UVOutMinMax.zw, uv0) (atlas cell); DiffuseA(uv) · lerp(palette(vc.g, IV.x), 1, 1 − d.a(1 − vc.r)), vc = BakedVCTexture(uv)	AlphaTest variants: d.a	Observed
glass	_Standard: the constant TintColor; _BCT* (switch BaseColorTexOnOff): BaseColour(uv0); _M: modulate as the plot family	transparency is not in RT0	Observed
snow	BaseColor(uv0 · UVTiling)	opaque	Observed
emissiveonly, pylon_cables	constant Color	opaque	Observed
diff1	DiffuseA(uv0 · UVTilingA)	opaque	Observed
ocean, ocean_shoreline	RT0 holds water data, not colour: the global ocean renderer colours them		Observed
uber_water	RT0 = InScatterColourA; the colour comes from its forward pass		Observed

The modulate uv of environment_plot is WorldSpaceUV_Modulate ? world.xz · WorldSpaceUVTiling_Modulate : TEXCOORD2, with ModulateUVTiling or ModulateUVTiling_PavementCorner, selected by TEXCOORD3.x ≥ 0.999 (float32); the luma weights are (0.299, 0.587, 0.114) and the gamma 1/2.2 and 2.2 as float32. The road decals’ uvM is TEXCOORD[ModulateUV] · ModulateUVTiling (int ModulateUV 0, 1 or 2; any other value samples (0, 0)), and DecalNormalA is read at uv0 · RoadUVMultiplierA. Observed

uber_road _TB (terrain blend, 12 materials) and glass _M were transcribed but not run on a material; the road world-uv origin depends on camera constants. Unknown

Glass transparency

glass is drawn by the forward glassTrack pass (17 Tokyo permutations, each with a G-buffer and a forward shader), which writes two targets: target 0 the reflected radiance (the fresnel term between F0 and 1 times the lit sky, with Cloudyness) and its gloss, target 1 the transmittance the background is multiplied by, 1 − k·c per channel, c the colour of the G-buffer shader (TintColor, or BaseColour.rgb with BaseColorTexOnOff) and k = (0.4·Alpha + 0.5)·(0.98 − a view-dependent term)·(1 − Cloudyness)·(1 − fog). The _ATT permutations (AlphaTestTexOnOff) drop the Alpha factor and discard where BaseColour.a < 1e-4: a cut-out, not a coverage. Observed

Cloudyness scales only k, so at 1 the transmittance is 1 and the reflected radiance of target 0 is the whole contribution; in 61% of the triangles of the 119 Shibuya glass materials the transmission alone, k times the luminance of c, is below 2%. Observed The two-target output has the form of dual-source blending, dst = target0 + dst·target1, as the car glass group does (Car materials); the blend state is run time, so that the reflection is added whatever the transmittance is inferred. Inferred

Roughness, metalness and normal

uvK is TEXCOORDK (D3D v down).

Family	Roughness	Metal	Normal	Evidence
uber_foliage _BushVertDeformHiQDoubleSided(_Tint)	NormalTexture(uv0).a · RoughnessScale	0	standard NormalTexture(uv0), then n = normalize(n + (up − n)·f) with f = sat(NormalFadeOffset + 1 − sat((d − FadedNormalEndDistance)/(FadedNormalStartDistance − FadedNormalEndDistance))), d the camera distance (Tokyo 4500/5000 m or 0/0, offset 0: f = 0 nearer than 4,500 m); no SV_IsFrontFace, so back faces keep the front normal	[V]
uber_foliage Tree	same	0	canopy normal (all four channels ·2 − 1, bounding-box normal cb9[23..24]): not traced	roughness [V]
uber_branch _BranchHiQ_NormalAlpha_AO	DiffuseA(uv0 · DiffuseAUVTiling).a	0	NormalA(uv0 · NormalAUVTiling): z from the unscaled x, y, then x, y ·= NormalAIntensity	[V]
environment	sat(RMA.x); modulate: sat(r − 1 + 2·mA), mA = sat((.5 + (tm.b − .5)k) + (Mod.b − .5)·_RoughnessStrength); deposits: r + (DepositRoughness − r)·mask	RMA.y	NormHeightTex_A (uv0 tangent)	[V] on two permutations
environment_plot	r0 = sat(RMA.r); _M: corner ? sat(r0 + 2mC − 1) : sat(r0 + 2mA − 1), mA = sat((Mod(uvM·ModulateUVTiling).b − .5)·_RoughnessStrength + .5), mC likewise with the corner texture	RMA.g	NormHeightTex_A(uvA · NormalHeight_A_UVTiling)	[V]; RCSM parallax omitted (≤ 0.003 rough, 0.04 normal)
uber_roaddecal, uber_roaddecalretro	D.a; Modulate: D.a(1 + (M.b − 1)k), k = sat(DecalNormalA.a + 1 − HeightInfluence)·COLOR.r; frost: + sat(4·sat((.5 − FrostHeightMin)/(FrostHeightMax − FrostHeightMin))·FrostAmount)(0.43 − r) (FrostAmount 0 in Tokyo)	Metalness switch: Mask_MetalnessATexture(uv).r (sRGB-decoded), else 0	standard DecalNormalA(uv)	[V]; POM parallax omitted
uber_road	r_L = UberRoad{L}DiffuseTexture.a, height-blended with the colour weights r = r_L + (r − r_L)·w_L; ModulateNormal: r ·= 1 + (M.b − 1)·Tint.a	0	per layer standard decode on the layer’s tangent, n = n_L + (n − n_L)·w_L (not renormalised); ModulateNormal: x,y = (2M.rg − 1)·ModulateNormalScale, z = sqrt(sat(1 − x² − y²)), n2 = normalize(Tm·x − cross(n, Tm)·w·y + n·z), n = n + (n2 − n)·Tint.a	[V]; TerrainBlend and winter Snow not traced
anim_flag	RoughnessMetalnessATexture(uv0).r (raw uv0)	.g	normalize(nA + COLOR.r·COLOR.a·nAnim); nAnim = NormalAnimTexture((uv2 + 1 − t)·NormalAnimUVTiling), t = time and wind scroll (run time)	[V] at t = 0
projection_lod	_RM: RoughnessMetalnessATexture(uvA).r; else the constant Roughness (0 in every Tokyo material)	_RM: .g; else 0	vertex normal	[V]
glass	RoughnessTexOnOff ? Roughness texture(uv0).r : Roughness; Modulate: sat(r − 1 + 2·sat((Mod(uv2·ModulateUVTiling).b − .5)·_RoughnessStrength + .5))	0	NormalTexOnOff ? standard Normal(uv0) : vertex; then a per-instance tilt lerp(n, normalize(n + r/2), GlassNormalOffsetStrength) with r = clamp(frac(hash(instance pos))·2 − 1, ±MaxPerturbAngle/90) (0 in Tokyo)	[V]; drawn in the transparent bucket by the forward glassTrack pass with the same roughness and normal
snow	RoughMetalAO(uv0·UVTiling).r	0	the sample used raw (no 2v − 1, no flip)	[V]
emissiveonly, diff1, pylon_cables	1	0	vertex normal; pylon _LineInner: constant game (0, 0, −1)	[V]
terr_mega _Default (far field)	sat(TerrainGlossTexture(uvB).r / 0.93)	0	world space, not a normal map: x = 2A.r − 1, z = 2A.g − 1, y = sqrt(sat(1 − x² − z²)) (game Y up), A = TerrainColourTexture(uvB)	[V]: 300 points, 0 error
window_linecastroom	see Environment	Metalness		window trace

All rows Observed; the evidence column gives the check. The standard normal decode is the one on Swatch textures.

Several families normalise B and some do not, which does not matter for orthonormal frames. The uber_road layers with UVSet < 0 use the constant game +X with w = +1. Observed

Terrain far field in the near range

The terr_mega row is the far field: no clipmap ring contains the pixel, the camera is far enough that the normal’s fade to the vertex normal has ended (sat((1 − d/cb7[34].y)·cb7[34].x), run-time clipmap constants), there is no twinkle (the sand and snow sparkle normal from TwinkleTexture, within 40 m of the camera), and the puddle path is off. The puddle path runs where TerrainGlossTexture.b and .a exceed 0.01 and a luminance mask of TerrainNormalTexture.a is non-zero, and its roughness and normal then depend on the run-time wetness globals cb2[91].y and cb2[140].y even in dry weather (interpreter: 78 of 100 such points change with random values of them). Observed The rural road families uber_megaroad and uber_megaroad_edge have no traced roughness or normal. Unknown

Foliage, branch and facade-LOD permutations

A dependency census of every GBufferLightScenario pixel shader of three families in the Tokyo track’s bin.zip (uber_foliage 27 permutations, uber_branch 7, projection_lod 24; each render-target channel followed back to its texture reads and parameter rows through the CBMP and TXMP maps), with the texture formats the Shibuya Scramble box’s materials bind (18, 6 and 59 materials; 13,674, 8,213 and 10,002 triangles). Observed

uber_foliage

• RT1.w is NormalTexture.a · RoughnessScale in the 11 HiQ, static and punch-through permutations of tree, bush and grass and in the ray-tracing hit shader; RTAOTexture.b joins it in Grass3DNoFlattenVCO. The 7 very-low-quality (VL) and billboard permutations write no RT1.w and no normal (RT2), and the 7 impostor permutations read MaskTexture, NormalTexture.a and RTAOTexture for both. Observed
• RT0.xyz is DiffuseTexture.rgb times a factor of RTAOTexture.b, MediumDOPower, MediumDOIntensity and three scene constants (cb2[43].x, cb2[44].x, cb2[159].x); the _Tint permutations add LUTTex and RTAOTexture.g, and the static, punch-through and world-tint ones RTAOTexture.r. Observed
• The bush and grass permutations take the normal from NormalTexture.r and .g only; the tree permutations read all four channels (the canopy normal). Observed
• Every map is sampled at TEXCOORD0 with no tiling parameter; the alpha discard compares sat((DiffuseTexture.a − AlphaStepMin)/(AlphaStepMax − AlphaStepMin)) with cb2[128].x, a constant of the draw. Observed
• Formats bound in the box: DiffuseTexture BC7 sRGB; NormalTexture BC7 linear (BC5 unorm on 2 triangles, which has no alpha); RTAOTexture BC7 linear; LUTTex BC1 sRGB (10,444) or BC7 linear (3,230); WorldTintMap BC1 linear or BC7 sRGB; MaskTexture RGBA8 sRGB; NoiseTexture BC4 unorm. Observed
• The impostor permutations (Impostor on; FramesXY_Colour 4, FramesXY_Depth 6) draw a tree from an atlas of views, chosen by the object-space camera direction a per-draw constant holds. DiffuseTexture is 4 × 4 colour frames (256² each in a 1,024² texture), MaskTexture 6 × 6 depth frames (BC4 snorm), NormalTexture an object-space normal (the texture’s name ends objnorm) and RTAOTexture the occlusion, in the same frame layout as the colour. The pixel shader encodes the direction hemi-octahedrally, dir = (x, (1 − |x| − |z|)^p, z) with x = (u + v)/2, z = (v − u)/2 and p = HemiOctahedronRemapYPower, takes the colour frames at u, v in {−1, −⅓, ⅓, 1}, blends the four nearest frames, ray-marches the depth frames, and discards when a saturated coverage of the blended alpha is below cb2[128].x. Observed Of the 16 colour frames, the 12 with |u| or |v| = 1 have dir.y = 0: horizon views, and the 4 inner ones look down. Inferred

uber_branch

• RT0.xyz is DiffuseA.rgb (read at uv0 · DiffuseAUVTiling) times a factor of NormalA.a, the vertex colour’s alpha, MediumDOPower, MediumDOIntensity and the same scene constants (the NormalAlpha_AO permutations: NormalA.a is the occlusion). RT1.w is DiffuseA.a; NormalA.a enters only the wetness terms. RT2 is NormalA.r and .g read at uv0 · NormalAUVTiling. Observed
• The only discard compares a value that is not a texture channel with cb2[128].x: the surface is opaque. Observed
• The _MS (mega-skirt) permutations read the terrain’s textures (DeferredTerrainRoot_MegaSkirtTwinkleTexture and a texture at a fixed register, class 0 space 4 t0) for a height-gated blend; the box’s six materials all select them. Observed
• Formats bound in the box: DiffuseA and FuzzBaseColor BC7 sRGB; NormalA, MicroNormal and FuzzNormal BC7 linear; NormalB BC5 unorm; TintTexture and FuzzMask RGBA8 sRGB. Observed

projection_lod

• Every map (DiffuseATexture, RoughnessMetalnessATexture, NormalATexture, EmissiveTexture, BakedVCTexture) is read at UVOutMinMax.xy + (UVOutMinMax.zw − UVOutMinMax.xy) · TEXCOORD0, the atlas cell of the building. Observed
• RT0.xyz is DiffuseATexture.rgb · ((1 − a(1 − v))(1 − s) + s) with a = DiffuseATexture.a, v = BakedVCTexture.r and s = SwatchTintTexture.r (read at row BakedVCTexture.g and the instance variation), through the wetness terms; the alpha is not the surface’s opacity. Observed
• 18 of the 24 permutations (every _RM one) write RT0.w = RoughnessMetalnessATexture.g and RT1.w = RoughnessMetalnessATexture.r before wetness; the other 6 use the constant Roughness and no metalness. The 6 HiQ permutations read NormalATexture.r and .g, the others the vertex normal. The AlphaTest permutations discard when DiffuseATexture.a is below cb2[128].x. Observed
• Over the 142 distinct BC5 SNORM RoughnessMetalnessATexture maps the whole box binds, 5,707 of 112,664,608 channels (0.0051%) in level 0 are below −0.06 (the 36 maps of every twentieth prototype: 19 of 34,701,312): the stored values are the positive half of the signed range. Observed
• Formats bound in the box: DiffuseATexture BC1 sRGB; RoughnessMetalnessATexture BC5 SNORM linear (9,650 triangles) or BC7 sRGB (352); NormalATexture BC5 unorm linear; BakedVCTexture BC5 unorm linear (9,842) or RGBA8 sRGB (160); EmissiveTexture BC1 linear (9,400), BC1 sRGB (12) or BC7 sRGB (518); SwatchTintTexture and LightmapTexture RGBA8 sRGB. Observed

Ground permutations: plots, terrain and roads

A dependency census of every GBufferLightScenario pixel shader of four families in the Tokyo track’s bin.zip (environment_plot 19 permutations, terr_mega 65, terr_distant 2, uber_road 50; each render-target channel followed back to its texture reads and parameter rows through the CBMP and TXMP maps), with the texture formats the Shibuya Scramble box’s materials bind (all of the box: 91,811, 14,462, 10,481 and 66,426 triangles). Observed

In environment_plot a material’s texture index dwords start 4 dwords before the end of its 384 bytes of constants: with that base every texture read is sampled with the tiling parameter of its own role, which the end of the constants as base does not give. In the other three the base is the end of the constants (64, 48 and 288 bytes). Observed

environment_plot is the environment family’s one-texture-set material on a terrain plot.

• It reads DiffAlphaTex_A, NormHeightTex_A and RoughMetAoPorTex_A with environment’s tiling parameters and offset (BaseColorAlpha_A_UVTiling, NormalHeight_A_UVTiling, RoughMetAoPoros_A_UVTiling, Uv0Offset), and RT0.w is RoughMetAoPorTex_A.g in all 19 permutations. Observed
• TerrainGlossTexture (sampled at BlendMapOffset + BlendMapScale · world xz) and NormHeightTex_A.a against BlendHeightMapRange make the mask of the snow term, which mixes the albedo, the roughness (RoughMetAoPorTex_A.r when dry) and the normal; AoWetBlendContextTex.r (TEXCOORD1 at MacroAOWetBlendMaskTiling) enters the occlusion factor with RoughMetAoPorTex_A.b. Observed
• The _M permutations add ModulateTexture and ModulateTexture_PavementCorner (TEXCOORD2 at ModulateUVTiling): red moves the brightness, green the saturation and blue the roughness around one half, scaled by _BrightnessStrength, _SaturationStrength and _RoughnessStrength. The _B permutations blend a second set (DiffAlphaTex_B, NormHeightTex_B, RoughMetAoPorTex_B) by BlendHeightMapBRange, and the _W ones add Snow_WetMax. Observed
• The alpha test discards DiffAlphaTex_A.a below cb2[128].x. Observed
• Formats bound in the box: DiffAlphaTex_A BC1 or BC7 sRGB; NormHeightTex_A BC7 linear; RoughMetAoPorTex_A BC7 linear; TerrainGlossTexture BC7 linear; AoWetBlendContextTex BC7 linear. Observed

terr_mega and terr_distant

• Every texture is sampled at BlendMapOffset + BlendMapScale · (x, z), the surface point’s position in the ground plane; terr_mega’s Alpha permutations take it from TEXCOORD1 and TEXCOORD4 instead. No other mesh coordinate is read. Observed
• TerrainNormalTexture.rgb (BC7 sRGB: a colour) times TerrainColourTexture.a squared is the albedo before the gamma 1/2.2 and snow terms; TerrainColourTexture.rg are decoded as x = 2 r − 1 and z = 2 g − 1, lerped with the vertex normal’s x and z by a distance term, and y = sqrt(sat(1 − x² − z²)) completes the vector, which makes them a world-space normal’s horizontal components Inferred; TerrainGlossTexture.r · 1.0752689 (r / 0.93) is the dry roughness; RT0.w is not written. The dependency sets of these channels are the same in terr_mega’s Default, Distant, Winter, Alpha_NR and Noise permutations and in terr_distant’s Default. Observed
• Formats bound in the box: TerrainColourTexture and TerrainGlossTexture BC7 linear; TerrainNormalTexture BC7 sRGB. Observed

uber_road

• A layer L of A to D is UberRoadLDiffuseTexture (colour in RGB, roughness in A) and UberRoadLNormalTexture (x = 2 r − 1, z = 2 g − 1, occlusion in B, a height in A). Both are sampled at TEXCOORD0 or TEXCOORD1 times UberRoadLUVTiling when UberRoadLUVSet is 0 or 1, and at the surface point’s position in the ground plane times UberRoadLWorldUVScale when it is −1. Observed
• The layers combine from the last: lerp(D, lerp(C, lerp(B, A, s1), s2), s3) of the albedo, the roughness, the occlusion and the decoded normals, with s1 = smoothstep(sat((s + (1 − hB) − w0 (1 + s)) / s)) for s = BlendSmoothness, hB the layer’s normal alpha and w0 the channel BlendShuffle[0] of BlendTexture (TEXCOORD2); s2 and s3 are the same with hC, w1 and hD, w2. Observed
• TintTexture (TEXCOORD2) is blended over the albedo as an overlay before the snow and frost mixes (UberRoadLFrost, UberRoadLSnow, FrostAmount); RT0.w is not written. Observed
• The permutations are named by the layer count (blend1 to blend4), M and modnorm (modulate maps), TB (terrain blend), winter and S (frost and snow). In the box Blend2 is on for 9,157 triangles and Blend3 for 582, Blend4 for none; UberRoadAUVSet is −1 for 22,433 triangles, UberRoadBUVSet for 9,095 and UberRoadCUVSet for 235, and 0 for the rest. Observed
• Formats bound in the box: the diffuse maps BC7 sRGB; the normal maps BC7 linear; BlendTexture BC7 linear (23,021) or RGBA8 sRGB (43,405); TintTexture BC7 linear. Observed

Emissive

The emissive term seeds RT1.xyz: g(d) = d ≥ 1 ? 1/(d + 1e-5)² : (d + 1e-5)², d the distance to the camera. emissiveonly: Color.rgb · EmissiveMultiplier · g(d) (Tokyo: 0). projection_lod _E: EmissiveTexture(uvA).rgb · EmissiveTextureScale · sat((INSTANCE_VARIATION.y − EmissiveVariationMin)/(EmissiveVariationMax − EmissiveVariationMin)) · g(d) (152 of 186 Tokyo materials bind the black ubershader_default_emissive). pylon_cables: Lightmap(uv1).rgb · Color.rgb · 0.96 · g(d) when LightmapActive (false in Tokyo). uberlod _Default: EmissiveTexture(uv0).rgb · EmissiveScale · g(d) (see Heightfield far view). Observed Whether a later pass multiplies g(d) back is not determinable statically (FPlusPlus_CombineSeperateDiffuseSpec_Final_Tile0 and the deferred composite FPlusPlus_Composite_NoDepthWrite_DirectLight_DXR_PixelClassify0 do not; no other pass was read), and the source of INSTANCE_VARIATION.y for hand-placed instances is not traced. Unknown The environment family’s _Emissive permutations are on Environment.

In the Shibuya box, 174 of the 1,366 placed prototypes hold an emissive material (306,233 triangles), and 87 of them (213,088 triangles) hold one whose emission depends on INSTANCE_VARIATION.y: 795 projection_lod, 2 environment and 1 environment_signage materials. Those 87 are placed 106 times. The 22 rule placements all store the byte 255, so the factor is 1. The 84 hand placements have no such byte (see the tint’s inputs). Their tint’s second float is 0 in 79 of them, 0.098 in one, 0.246 in one, and 0.648, 0.922 or 1 in the other three. The ramps (EmissiveVariationMin, Max) of the materials concerned are (0.2, 0.4) in 508, (0, 0.2) in 122, (0.6, 0.8) in 76 and (0.7, 0.9) in 62. Observed · the prototypes and placements of the box If that float were y, 80 of the 84 hand-placed buildings would emit nothing. Nothing establishes whether it is y or a field left at zero. Unknown The preset scales that turn the authored values into lighting units are on Time-of-day presets and light units.

The authored magnitudes in the prototypes of the Shibuya box (every prototype placed there, counted by material and triangle) are small: GIEmissiveScale is 1 in all 9,467 environment, 286 environment_plot and 188 environment_signage materials; EmissiveTint is at most 1 per channel (white in all but about 35 of those materials, whose channels run from 0.002 to 0.982); EmissiveTextureScale of projection_lod is 1 in 1,166 of its 1,178 materials and between 0.16 and 1.4 in the other 12; the emissiveonly materials (12) have EmissiveMultiplier 0 and a black Color; the anim_ledflipbookpanel materials (8) have Brightness 1 or 10; Emissive is 1 in 227 environment (18,966 triangles), 90 environment_signage (2,462), 912 projection_lod (218,071), all 10 uberlod and 1,337 window_linecastroom materials (25,167 triangles), whose EmissiveVariationMin and Max are fractions between 0 and 1 (most often 0.2 to 0.4, 0.6 to 0.8 or 0.7 to 0.9). Observed · a census of the box’s prototype materials

Heightfield far view (uberlod)

The track’s uberlod shader file (the family of the autouberlod_* clusters of the far heightfield) holds two G-buffer permutations, _Default and _Rock_NM. Both were dumped and their render-target channels followed. Observed

• _Default reads DiffuseATexture, RoughnessMetalnessATexture and EmissiveTexture, every one at TEXCOORD0 as it is, and no normal map; _Rock_NM reads DiffuseATexture and NormalATexture at TEXCOORD0, takes roughness and metalness from the parameters Roughness and Metalness, and reads no emission texture. The parameter switches RoughnessMetal, Emissive, NormalMap and Modulate are not rows of the constant buffer. Observed
• _Default: RT0.w is RoughnessMetalnessATexture.g (the metalness) and RT1.w is RoughnessMetalnessATexture.r raised to a power that depends on the vertex normal’s y and a scene constant (the roughness). RT0.xyz is the diffuse colour moved away from or toward its luma (weights 0.3, 0.59, 0.11) by a factor of the normal’s y and two scene constants, times a scene constant. No texture alpha is read, so nothing is alpha tested. Observed
• The emission of _Default adds g(d) · EmissiveScale · EmissiveTexture.rgb to RT1.xyz, g(d) the distance encoding of Emissive. It has no instance variation, no tint, no GIEmissiveScale, no lightmap. Observed
• The Shibuya box holds 10 uberlod materials (8,568 triangles) of one signature: RoughnessMetal, Emissive on, NormalMap, Modulate off, EmissiveScale 1, Roughness 0.8, Metalness 0, LightSwitchOnStart 0.01, LightSwitchOnEnd 0.02 (the last two are read by neither permutation). DiffuseATexture, EmissiveTexture and RoughnessMetalnessATexture are BC1 sRGB, NormalATexture BC5 linear. The switches match _Default, not _Rock_NM. Inferred · the switch names; which permutation a material takes is not read from the shader bin

Road decals

Over the 30 G-buffer permutations of uber_roaddecal (decal, winter, metal, T, AM, Modulate, POM in their combinations) and the 5 of uber_roaddecalretro (decal, HA, M, T, T_M) of the Tokyo track, each render-target channel was followed back to its texture reads and parameter rows. Observed

• Only the AM permutations of uber_roaddecal and every uber_roaddecalretro permutation test a mask per pixel. The other uber_roaddecal permutations (_decal, _T, _metal_T, …) have two discards, both on 1 − cb2[128].x < 0, which holds for the whole draw or for none of it; they write every pixel of a draw otherwise, the diffuse alpha is the roughness and no opacity is derived. Observed · all 35 permutations hold two discards; the mask test is the first one in the permutations named, and the 4 non-AM ones without winter have no noise code at all
• The discard is a dithered coverage test. With s the test value (mask · (AlphaFadeOnOff ? COLOR.g : 1), or the HA form; the retro permutations also multiply by a camera-distance smoothstep of scene constants) and the scene’s dither on, the pixel is kept where s·sat(5(s − 0.2)) − 1 + n ≥ 0, with n = frac(52.98·frac(0.0671·x + 0.00584·y)) (interleaved gradient noise of the pixel coordinates; when a second scene flag is set both coordinates are first shifted by (frame & 63)·5.588). The test value is then 0.5 or −0.5 and is compared with the per-draw reference cb2[128].x. With the dither off the pixel is kept where s reaches that reference. s·sat(5(s − 0.2)) equals s from 0.4 up, so a hard 0/1 mask is exact either way, and a plain test at one half is the median of the dither. Observed · the discard code of the _decal_AM permutation read; the constants are those of the compiled code
• Dry albedo goes toward FadeColor by a smoothstep between FadeStart and FadeEnd of the screen-space derivative of uv0.x (a far fade); wetness (WetScale, Porosity, HeightPuddlingOnOff) acts on run-time rain state, so the other formulas on this page are the dry surface. Snow and Frost change nothing while SnowAmount and FrostAmount are 0, as in every material of the Shibuya box. Observed
• The retro permutations’ RT1.xyz carries a retro-reflective term from RetroReflectivity, RetroReflectivityFalloff and Hardness; the retro permutations write metalness 0, as do the uber_roaddecal ones without metal. Observed
• Enabled is a CBMP entry of uber_roaddecal (byte 228), but no render-target channel and no discard of any of the 35 permutations depends on it. It is 0 on 1,161 of the 20,743 road-decal triangles of the Shibuya box (the parameter is stored as a switch). Whether the draw path honours it before the shader runs is not established. Observed[the dependencies of all 35 permutations’ outputs and discards; the materials of one Shibuya area]
• In every material of the Shibuya box RoadUVMultiplierA is (1, 1), so the diffuse, normal and mask reads coincide with the mask’s raw uv0. Observed · the materials of one Shibuya area

Mesh decals

The mesh permutations _Standard_DR, _DR_NAO and _DR_ST of environmentdeferreddecal sit in draw bucket 0x1C (transparent, no opaque bit; all 917 Shibuya groups) and carry the AlphaBlended technique tag, so they are drawn by the deferred-decal pass (environmentdeferreddecalDeferredDecalsLightScenario), not the alpha-tested G-buffer scenario (which also exists, with discard a < cb2[128].x). Observed

uv  = uv0 + Uv0Offset;  c = DiffAlphaTex(uv · DiffAlphaUVTiling)         (_ST: c.rgb swatch-tinted as environment)
a   = c.a · DiffRoughIntensity (r4.w) · (VertexAlphaOnOff ? COLOR.a : 1)
NAO: aN = c.a · NormalAOIntensity (r5.x) · (VertexAlphaOnOff ? COLOR.a : 1);   discard max(a, aN) < 0.01 (else a < 0.01)
SV_Target0 = (c.rgb, a)
SV_Target1 = NAO ? (normalize(T·x − B^·y + N·z) of NormalTex(uv · NormalUVTiling), aN) : 0
SV_Target2 = (1 − RoughMetAOTex(uv · RMAOUVTiling).r, NAO ? 1 − max(cb2[43].x, RMAO.b) : 0, 0.5, NAO ? max(a, aN) : a)

The ten pixel shaders of the family are these two mesh groups (_DR, _DR_NAO and _ST; the RCSM ones add a cone-step march of ConeMapTexture for uv) and the two projected ones. The permutations with the DiffRough switch off are different: _Standard discards every pixel and writes zero to all three targets, and _Standard_NAO writes zero to targets 0 and 2 and only the normal and its weight to target 1. Observed · the dumped bodies of all ten permutations Every material of the Shibuya box has DiffRough on and Projected off (the draws are the _DR ones); NormalAOIntensity is 0 in 1,056 of its 38,263 triangles, where the normal’s weight is 0 and the occlusion’s weight is a. Observed · the materials of one Shibuya area

The blend state and the resolve of the decal buffer into the G-buffer are in no shipped file. Unknown

What does the decal buffer resolve do?

The depth, raster and blend state (names such as DecalGreaterEqual and ZBiasDecalOpaque appear in the executable) and the decal resolve are in encrypted settings or code, so how decals composite is not traced.