Skip to content
Engine Wiki

Environment shaders

ForzaTechShadersApplies toFH6

environment is the family of city facades, building kits and most props; environment_plot, environment_skinned and the signage_* families are close relatives. Its compiled G-buffer shaders were disassembled and the formulas below read from them. Observed Pixel shaders of this family were also interpreted on real merged constants and real textures and matched the formulas; where a formula is marked untraced it depends on a run-time constant. Observed

The numbers come from a recorded census of the Tokyo track Brio, no longer re-checkable. Observed

Inputs

Section titled “Inputs”

TEXCOORD0 to TEXCOORD2 are the dequantised UV sets (v_d3d = 1 − v), COLOR is the raw vertex colour, INSTANCE_VARIATION.x is the placement record byte at +0x34 divided by 255 (Placement files), and AoWetBlendContextTex is sampled on TEXCOORD1 (its islands match the UV1 layout exactly). Observed

Swatch tint (the _ST permutations: city facades)

Section titled “Swatch tint (the _ST permutations: city facades)”
w      = 1 − base.alpha · (1 − COLOR.r)            base colour alpha = tint mask, vertex red = tint off
tint   = SwatchTintTexture.SampleLevel(u = COLOR.g, v = INSTANCE_VARIATION.x)      a 32 × 512 palette
albedo = base.rgb · lerp(tint, 1, w)

Without the tint every facade is near white: the tiling textures are deliberately pale and get their colour from the palette. Observed For the Shibuya facade SHOP_W7_02_A_Tiling_14_VARIANT_6 the albedo before modulate and AO is sRGB ≈ (214, 214, 213). Observed Its visible character comes from the AO and grime mask, deposits, modulate, normal map and lighting. Inferred

The tint’s inputs over the Shibuya box

Section titled “The tint’s inputs over the Shibuya box”

Read from the models of every prototype of the Shibuya Scramble box (summer), 568 of its 1,534 prototypes draw an environment material with SwatchTint on: 3,984 primitives, 729,706 triangles, 976,427 vertices. Observed

  • COLOR.r is 0 on 666,214 vertices and 1 on 306,530; 3,683 (0.4%) lie between. The triangles whose corners disagree on red number 499. Observed
  • COLOR.g is one value over the whole primitive in 3,196 of the 3,984, at most four in 772 more and five to 32 in 16; the vertices hold 47 distinct values over 0 to 255, 255 on 293,996 and 24 on 153,316 of them. Among the triangles whose red is 0, 813 have corners that disagree on green. Observed
  • A base colour with an alpha channel is the rule: 662,422 of the 729,712 triangles of these materials (90.8%) are drawn with a BC7 sRGB DiffAlphaTex_A, whose texels have alpha under 250 on average 46% at its third mip level; the other 67,242 use an opaque BC1 sRGB texture, so the tint there covers the whole surface where red is 0. Observed
  • The palette textures are RGBA8 or BC7 sRGB and range from 4 × 4 to 1024 × 1024 (32 × 512, 32 × 32, 64 × 64, 16 × 16 and 256 × 256 among them), with one to seven stored mips. Observed
  • Rule placements (.pgeo records) carry the byte at +0x34: 2,801 distinct (prototype, byte) pairs over 39,446 instances, at most 111 bytes for one prototype. Observed Hand placements (.pvs instances) have no such record: 26 instances of these prototypes. Observed Their transform’s tint (four floats) is not a whole byte over 255 in 25 of the 26; the values are round authored fractions (0.51, 0.514, 0.13, 0.66, 0.09) with the other components equal to each other or one, the same kinds of population as the rule records’ four bytes (all ones, all zeros, a varied first with the rest ones). Observed The first float is therefore the likely INSTANCE_VARIATION.x of a hand-placed instance, read as the palette row as it is. Inferred
  • Over all 835 hand placements of the box, of every prototype, the tint is (1, 1, 1, 1) in 281, (0, 0, 0, 0) in 232, (1, 0, 0, 0) in 90 and something else in 232, mostly of the shapes (a, a, 0, 0) and (a, b, b, 0). The box’s 61,292 rule records hold four bytes of 255 in 9,435, four zeros in 6,411 and mostly a varied first byte with the other three 255 in the rest. Observed The later tint components do not follow the rule record’s pattern, so which of them, if any, a hand-placed instance’s INSTANCE_VARIATION.y comes from is not established. Unknown

The palette read

Section titled “The palette read”
  • The palette is the only texture the _ST pixel shaders sample through the static sampler register s11, and it is sampled with SampleLevel at level 0. Observed (the compiled pixel shaders of the _Standard_ST permutations)
  • The root signature of those shaders declares no static samplers, so the filter and the address mode of s11 are set at run time and are not in the shader. Observed The filter is not known. Unknown
  • Over the Shibuya box the (palette, green, row) combinations read number 1,267 (rows from the rule placements’ bytes over 255 and the hand placements’ first tint float); a bilinear read of the same positions differs from a nearest-texel read by more than two 8-bit steps in 477 of them with clamp addressing and 592 with wrap. Observed Which filter the game uses is not established, so a nearest-texel read is a guess. Unknown

G-buffer albedo (_StandardRCSM_ShadowsTriPlanar_ST_TPNAO_D)

Section titled “G-buffer albedo (_StandardRCSM_ShadowsTriPlanar_ST_TPNAO_D)”
uv    = TEXCOORD0 + Uv0Offset                         then RCSM cone-step parallax (view dependent)
base  = DiffAlphaTex_A(uv · BaseColorAlpha_A_UVTiling)            BC1 sRGB
tint  = SwatchTintTexture(u = COLOR.g, v = INSTANCE_VARIATION.x)  RGBA8 sRGB
a0    = base.rgb · lerp(tint, 1, 1 − base.a · (1 − COLOR.r))
aow   = AoWetBlendContextTex(TEXCOORD1 · MacroAOWetBlendMaskTiling)   x AO, y wetness mask, w deposit amount
rma   = RoughMetAoPorTex_A(uv · …)                                   x rough, y metal, z AO, w porosity
dep   = DepositsATexture(uv).x                                         BC4 SNORM

Triplanar modulate

Section titled “Triplanar modulate”
world = cb2[8].xyz + WORLDPOS: the absolute world position (game frame); n = normalize(WORLDNORMAL), no normal map
noise = simplex3D(world · TriPlanarNoiseVariationTiling) · 16
nf    = NoiseOnOff ? lerp(1, sat((0.5 − StepMin + noise) / (StepMax − StepMin)), Strength) : 1
frame = ModulateConstants {c0, c1, c2}: the .pvs frames table at the .pgeo record's u16 @ +0x3E
        (65535: a per-draw constant buffer instead, run time); c1.w · c2.w ≤ 0 (frame 0): uv = 0, k = 0
P = world · TriPlanarTextureTiling + R − c0.xyz;  A = c1.xyz, B = c2.xyz, C = cross(A, B)
    R = RandomOffsetOnOff ? frac((p.x·4.192 + p.z·2.624 + p.y·3.213, p.y·7.554 + p.x·4.523 + p.z·6.587,
                                  p.z·11.125 + p.y·10.474 + p.x·14.853)) · 123.456 : 0,  p = instance position
uA = dot(P,A)·c1.w + .5, uB = dot(P,B)·c2.w + .5, uC = dot(P,C)·c1.w + .5
|n·C| > |n·A| and > |n·B|:   uv = (uA·sign(n·C), uB),   w = sat(|n·C|−|n·A|) · sat(|n·C|−|n·B|)
else if |n·A| > |n·B|:       uv = (−sign(n·A)·uC, uB),  w = sat(|n·A|−|n·B|) · sat(|n·A|−|n·C|)
else:                        uv = (uA·sign(n·B), −uC),  w = sat(|n·B|−|n·A|) · sat(|n·B|−|n·C|)
k  = pow(w · SeamSize, SeamHardness) · nf · BrightSatRoughStrength
tm = TriPlanarModulateTexture(uv)  (global sampler s1)

Modulate, deposits and AO

Section titled “Modulate, deposits and AO”
m2  = ModulateTexture(ModulateUVTiling · (WorldSpaceUV_Modulate ? world.xz · WorldSpaceUVTiling_Modulate : TEXCOORD2))
bri = sat((.5 + (tm.r − .5)·k) + (m2.r − .5)·_BrightnessStrength);  sat = the same with .g and _SaturationStrength
g = a0^(1/2.2); o = bri > .5 ? 1 − (1 − g)·2·(1 − bri) : g·2·bri; L = dot(o, (.299, .587, .114))
c1 = lerp(a0, sat(lerp(L, o, 2·sat))^2.2, 1 − rma.y)                  metal is never modulated
dn   = sat((simplex3D(world · DepositMacroNoiseTiling)·16 + .5 − MacroStepMin) / (MacroStepMax − MacroStepMin))
mask = sat(2·(1.5 + (dn − 1)·MacroStrength + (sat(aow.w) − .5)·MidStrength) − 2) · DepositsStrength
       · (1 + (sat(dep) − 1)·MicroStrength)
c2   = lerp(c1, DepositDiffuse, mask)                                 DepositDiffuse default (0.0529, 0.0482, 0.0395)
albedo = c2 · (1 + K · (max(..., AO(rma.z, aow.x), ...) − 1))          K = a per-frame constant (cb2[159]), run time

Wetness is zero when dry. Observed DualUv permutations sample base, normal and RMA twice (layer A on TEXCOORD0, layer B on TEXCOORD2 with DualUV_BaseColorAlpha_A_UVTiling) and blend lerp(A, B, COLOR.b); tint, modulate and deposits follow unchanged; _Standard_ST is base and tint only. Observed

Read from the shaders and checked on data: the permutation selection rule, the register and slot maps, every formula above, UV1 for the AO and context mask, and the ModulateConstants frames. Observed Not traced: the AO-into-albedo constant K and the AO strength globals (encrypted render settings), and the sampler filter modes (run-time descriptor tables; the palette’s 2 × 2 texel cells make nearest and bilinear agree at v = byte/255 except for bleeding at small rows). Unknown

The simplex noise is, op by op: skew s = dot(p, 1/3), i = floor(p + s), x0 = p − (i − dot(i, 1/6)); corner order from x0.x ≥ x0.y, x0.z ≥ x0.x, x0.y ≥ x0.z; per corner t = sat(0.6 − |x|²)⁴, h = uint(frac(sin(dot((c.x + c.z, c.y + c.z)/256, (12.9898, 78.233))) · 43758.5453) · 255) & 15, sum += t · dot(g_grad3[h], x), result sum · 16; g_grad3 is the 12 edge gradients of a cube followed by the first four again. The hash runs sin on arguments in the hundreds, where GPU sin implementations differ in the low bits; 93% of Shibuya’s lattice corners change gradient between float32 and float64 sin. The noise’s lattice, frequency and amplitude are exact everywhere; the blotch layout is not reproducible across GPUs. Observed

What each G-buffer channel depends on

Section titled “What each G-buffer channel depends on”

Reading every GBufferLightScenario pixel shader of environment in the Tokyo track’s bin.zip (379 permutations, one per variant) through the CBMP and TXMP maps, and following each render-target channel back to its texture reads and parameter rows, gives these dependencies. Observed

  • RT0.w, the metalness, is RoughMetAoPorTex_A.g in the 273 permutations that sample one texture set: alone in 199, and with the mega-skirt blend’s textures in the 74 _MS ones. The 106 _B (two-set) permutations add RoughMetAoPorTex_B.g, NormHeightTex_A.a and AoWetBlendContextTex.b (54 alone, 52 with the mega-skirt blend).
  • RT1.w, the roughness, is sat(RoughMetAoPorTex_A.r) plus the wet terms; with the rain constant at 0 it is exactly that.
  • RT2.xyz reads NormHeightTex_A.r and .g only (378 of the 379; the one permutation without a normal output is the ray-tracing hit shader), as x = 2r − 1, y = 2g − 1, z = sqrt(sat(1 − x² − y²)), brought to the world as x·T − y·B + z·N with B = (N × T)·w. The vertex shader gives w as sign(2·w_vertex − 1) when the tangent is stored as unsigned bytes and the stored w otherwise (_Standard).
  • RT0.xyz reads DiffAlphaTex_A.rgb in 378 of the 379 (the same one lacks it). The sample is taken through an sRGB view (linear), raised to 1/2.2 before the wet term and to 2.2 after it, so a dry surface returns the linear sample; the occlusion term of the albedo section multiplies it.
  • The three maps are sampled at Uv_k · (TEXCOORD0 + Uv0Offset) with their own tiling parameters, BaseColorAlpha_A_UVTiling, NormalHeight_A_UVTiling and RoughMetAoPoros_A_UVTiling; a time-driven vertical scroll is added to V when a parameter row is non-zero (AnimTranslate). The DualUV_* tilings enter the permutations that blend two UV sets.
  • AlphaTest permutations discard when DiffAlphaTex_A.a − cb2[128].x < 0: the reference is a constant of the draw’s buffer, not a material parameter, and the materials’ ATST bytes are 1, 0.
  • EmissiveTexture reaches only RT1.xyz (the lit colour), not the albedo.
  • The EmissiveTexture files bound by the EmissiveAltUV materials drawn in the Shibuya Scramble box (21 distinct textures, 256 x 256 or 512 x 512, linear) are all BC6H unsigned float, and they hold radiance rather than a 0 to 1 colour: the largest red, green or blue value of each is between 3.5 and 3,808, and between 4,700 and 141,400 texels of each map are above 1. Observed · every one of the 21 decoded to float32 and its extremes taken

The _Emissive permutations (_Emissive, _Emissive_EAUV, _Emissive_EF) add this emission to RT1.xyz, in the g(d) encoding of Emissive: Observed[the expressions of those three permutations’ RT1 channels]

E = DiffAlphaTex_A.rgb · EmissiveTexture.rgb · EmissiveTint.rgb · GIEmissiveScale · V · H · g(d)
V = 1 + LightEmisVariationOnOff · (sat((INSTANCE_VARIATION.y − EmissiveVariationMin)
                                       / (EmissiveVariationMax − EmissiveVariationMin)) − 1)
H = EmissiveTempVariationOnOff ? √3 · c / |c| : 1
c = lerp(lerp(1, WarmHue.rgb, sat(h)), ColdHue.rgb, sat(−h))     h = 2 · frc(hash) − 1
  • DiffAlphaTex_A is read at BaseColorAlpha_A_UVTiling · (Uv0Offset + TEXCOORD0), as in the other permutations, but EmissiveTexture at Uv0Offset + TEXCOORD0, without a tiling: the two maps line up only when the base colour map is not tiled. Observed
  • The hue term H is a direction of length √3, so it leaves white unchanged and its magnitude never depends on h. h runs from −1 (fully ColdHue) to 1 (fully WarmHue). Its hash is the sum of three terms ceil(row_k.x · q + row_k.w), one per row of a scene constant buffer, weighted 14.853, 10.474 and 11.125. Observed Those are the weights of the x, y and z of the instance position in the third component of the random offset of Triplanar modulate, so the rows likely quantise the three axes of a position and the hue changes in steps from one place to the next. Inferred
  • _Emissive_EF (EmissiveFlash) scales the emission by 1 + EmissiveFlashIntensity · (on − 1), where on is 1 while frc(t · FlashTimeScale + FlashOffset) ≤ FlashOnDuration and 0 otherwise, t being a scene constant. The emission therefore drops to 1 − EmissiveFlashIntensity of its value for the rest of each period. Observed t is likely the clock. Inferred

Mega-skirt (the _MS permutations)

Section titled “Mega-skirt (the _MS permutations)”

Following _Standard_MS’s albedo channels back to their inputs gives the blend toward the ground near the bottom of a mesh. Observed

h  = worldpos.y + cb2[8].y − height(xz)          height: a texture at cb7[29] · (worldpos.xz + cb2[8].xz + cb7[29].xy)
H  = MegaSkirtUseVariationOnOff ? INSTANCE_VARIATION.z : MegaSkirtHeight
s  = sat((h − H) / −H) · cb7[28].x               1 at the ground, 0 at the height H
m  = lerp(clip.r, 0.5, f) · 2 − 1                clip: the terrain clipmap (class 0, space 4, t0, a 2D array)
k  = sat((sat(S · m − S + (2S + 1) · s) − MegaSkirtHeightMin) / (MegaSkirtHeightMax − MegaSkirtHeightMin))
albedo = lerp(DiffAlphaTex_A, clip · (1 − f), k)                         S = MegaSkirtHeightStrength
  • clip is read in the clipmap ring that holds the pixel: a loop over the rings at cb7[36] onwards compares the position with each ring’s origin and extent, the coordinate is (pos − origin) · scale from that ring’s row, and the array slice and mip come through a Texture3D<uint> page table (space 12). f fades out the ring at its edge. Observed
  • The ring rows, the height texture’s mapping (cb7[29]) and cb7[28].x are scene constants written by the engine, and the clipmap is the near-field ground the engine writes at run time (Near-field ground), so the blend’s colour, its height source and its overall factor are not in any file. Observed The material holds only MegaSkirtHeight, MegaSkirtHeightMin, MegaSkirtHeightMax, MegaSkirtHeightStrength, the normal limits MegaSkirtNormalMin and MegaSkirtNormalMax (read by the normal channels) and MegaSkirtUseVariationOnOff. Observed

Signage

Section titled “Signage”

signage_* families (the *_CM permutations) colour a one-channel or RGBA mask atlas of AtlasImages.zw (columns, rows) cells with constants:

cell  = min(cols·rows − 1, AtlasIndex − 1)                      AtlasIndex is 1-based
uv    = (cell_xy + TEXCOORD3) / (cols, rows)                    TEXCOORD3 = the sign face's 0..1 UV
m     = bicubic(mask, uv)
if UseSteppedMask:   m = saturate((m − 0.4) · 5)
c = lerp(BackgroundColor, ColorR, m.r)
if UseFiveColorMask: c = lerp(c, ColorG, m.g); c = lerp(c, ColorB, m.b); c = lerp(c, ColorA, m.a)

Constant registers (byte offset ÷ 16, from the constant map): background 1, R 2, G 3, B 4, A 5, AtlasImages 10.zw, AtlasIndex 11.xy, UseFiveColorMask 48.x, UseSteppedMask 57.z. TEXCOORD3 exists only on a model’s sign meshes, so attribute sets are per mesh, not per model. Observed

Around that, over the 19 Tokyo signage_* G-buffer permutations (each render target channel followed back through the constant and texture maps; the texture indices of the parameter buffer start at dword 236): Observed

  • Without _CM the sampled cell is the colour itself; with _CM it is the mask above. The sample is taken through the format’s own view, raised to 1/2.2, darkened by wetness and raised back to 2.2, so a dry surface returns the linear sample; the BiCubic permutations filter with Catmull-Rom, the others bilinearly. The result is saturated.
  • The metalness is RoughMetAoPorTex_A.g, the roughness saturate(RoughMetAoPorTex_A.r) (wetness off) and the normal the standard decode of NormHeightTex_A.rg, all at TEXCOORD0 with the environment family’s tilings; the albedo is multiplied by a factor of AoWetBlendContextTex.r read at TEXCOORD1. A sign face is one coordinate set (TEXCOORD3) and the surface maps are others.
  • The Signage*Mural* permutations lerp the wall colour (DiffAlphaTex_A, TEXCOORD0) to the sign colour by MuralTexture.r read in the same cell.
  • The _SignageEmissive* permutations write EmissiveTexture.rgb · EmissiveTint.rgb · (the sign colour above), times GIEmissiveScale and an instance-variation factor (LightEmisVariationOnOff), to RT1.xyz as g(d) of Emissive; EmissiveTexture is read in the same cell of the same TEXCOORD3 atlas, or at TEXCOORD4 with EmissiveAltUV.
  • The only discard compares a constant of the draw with one: the surface is opaque. The scroll, rotation and scale of the four mask channels, RandomiseIndexOnOff and _AnimatedSign are driven by the clock.

Windows and other relatives

Section titled “Windows and other relatives”

window_linecastroom (parallax interiors): the base colour comes from FrameDiffuseandMask (else FittingsDiffAlphaTex); roughness is max(FrameGlassRoughness(uv2 frame).r, lerp(1 − a, GlassRoughness(uv0 · tiling).r, a)) with a = COLOR.a; FrameGlassNormalTexture is sampled on uv2 (frame atlas) but built on WORLDTANGENT0, the only tangent the window shaders take. The room’s x centre and width come from INSTANCE_VARIATION.z/.w (Placement files). Observed

Over the 51 Tokyo G-buffer permutations (texture indices from dword 124): Observed

albedo   = lerp(t, F.rgb, F.a) · (1 − (1 − Fit.a)(1 − F.a))     F = FrameDiffuseandMask at uv2 in its atlas
t        = FittingsDiffAlphaTex.r at uv1 (in its atlas) · WindowTintColour
metal    = Metalness

so the albedo is 0 where neither the frame nor a fitting covers the pane. FrameUseAtlasOnOff picks the frame cell (FrameAtlasIndex + uv2) / FrameAtlasCount. The room behind the pane (BoxDiffRoughTex, BoxEmissiveTex, up to three furniture layers with parallax) is traced in the pixel shader from the world position, the camera and the material’s RoomPos, RoomScale, RoomRotation, RoomAtlasIndex, RoomAtlasCount and RoomRandomisationOnOff (the room is a box intersected with the eye ray per wall; its atlas cell is (RoomAtlasIndex + h) mod RoomAtlasCount, h a hash of the ceiled RoomPos under a per-draw matrix when RoomRandomisationOnOff, else 0), and reaches RT1.xyz, not the albedo. The only discard compares a constant of the draw with one. Observed

A cell of the room atlas (BoxDiffRoughTex, BoxEmissiveTex) is four columns wide: the first column holds two side walls (one above the other), the second a back wall in its lower half and black in its upper half (the front, which no view reaches), the third the floor and the fourth the ceiling, each one tile spanning both rows of its column. Viewed as decoded images, the BoxDiffRoughTex textures of the Shibuya Scramble box’s windows that were looked at (2048 x 512 for a one-cell texture) show this layout; the back wall is the tile at u 0.25 to 0.5, v 0.5 to 1 of the cell, v down the image. Without BoxUseAtlas the whole texture is one cell; with it RoomAtlasCount gives the columns and rows of cells and RoomAtlasIndex a zero-based (column, row), wrapped by the count. Observed

The room’s glow

Section titled “The room’s glow”

In _Standard_E the traced room’s emission (BoxEmissiveTex) reaches RT1.xyz scaled as follows: Observed · the expressions of the permutation’s three RT1 colour channels

E = room · g(d) · GIBakeEmissiveScale · F · WindowTintColour.rgb · H · ExposureValues · COLOR.x · S
S = sat((INSTANCE_VARIATION.y − EmissiveVariationMin) / (EmissiveVariationMax − EmissiveVariationMin))

g(d) is the distance encoding of Emissive and H the warm/cold hue term of the _Emissive permutations (WarmHue, ColdHue, length √3), which the shader applies or leaves out by a switch, as there. Observed F has the form (1 − a)(1 − b⁵), the shape of one minus a Schlick Fresnel term, so less of the room shows at grazing angles. Inferred So a pane’s room is dark where its vertex colour’s red is 0 or where its placement’s INSTANCE_VARIATION.y is at or below EmissiveVariationMin, and nothing in the term depends on the time of day: none of the 51 permutations reads LightSwitchOnTime or LightSwitchOffTime. Observed

Over the Shibuya box (the 3,259 window_linecastroom materials loaded with its prototypes, their 1,361 window primitives and 22,662 placements): Observed · a census of the box’s prototypes and placements

Input Values Confidence
ExposureValues 1.4 in 2,908 materials, 0.352 in 183, 11.2 in 156, 0.704 in 12: a material constant Observed
GIBakeEmissiveScale, EmissiveTempVariationOnOff 1 in every material Observed
Emissive 1 in 3,163, 0 in 96 Observed
LightSwitchOnTime, LightSwitchOffTime −1 in 2,643; 0.01 and 0.02 in 616 Observed
EmissiveVariationMin–Max 0–0.2, 0.1–0.3, 0.2–0.4, 0.6–0.75, 0.6–0.8, 0.7–0.9, 0.7–1, 0.75–0.9, 0.8–0.9 and 1–1 over the placed primitives; most often 0.7–0.9, 0.75–0.9, 0.7–1 and 0.6–0.8 Observed
COLOR.x Every window primitive has COLOR_0. Over the 25,281 triangles of the prototypes, 27.6% are 0, 21.9% are 1 and 50.4% between (multiples of 1/255, 120/255 the commonest); weighted by placements, 96.4% of 145,639 placed triangles are 1 and 1.3% are 0 Observed
INSTANCE_VARIATION.y The placement record’s variation byte over 255, spread evenly over 0 to 1 on the 22,654 rule placements; the 8 hand placements have no variation byte Observed

With those ramps S is 0 on 73.4% of the 140,366 placed triangles of rule placements, 1 on 11.0% and between on 15.6%; COLOR.x · S is 0 on 73.5%. So about three windows in four show a dark room wherever the box is drawn. Observed