Geographic summer-meadow candidate

Current result: the 02:22 R3/corridor/cut combination is retained as an incremental improvement, not a photorealism or AAA acceptance. Earlier rejected iterations below remain historical evidence.

Original geographic candidate

The native distant meadows still read as large uniform green surfaces. This bounded candidate replaces only summer terrain Layer1 with the unchanged attributed orthophoto, white base colour, and exactly the same absolute-XZ geographic mapping as Layers2/3. The separate R mask blends it in from 100 to 400 metres from the nearest of all three Alpine routes; that mask is owned and checked by the regional agent. Close Layer0 grass retains its real 1.4-metre surface. Geology, summer crown colour, winter, lighting and road geometry are unchanged.

Layer1's former repeating grass normal and roughness maps are removed because their geographic rescaling would be incorrect. No optional detail layer is added in this first comparison. The existing rough matte scalar surface remains. The photograph retains its original seasonal colour and baked shadows; distant buildings may still exist in its pixels. These limits must be assessed in the actual Alps01 image, not accepted from a source preview.

The exact previous material source is archived at ArtSource/Alps/SummerMeadow/GeographicMeadowCandidate/AlpineTerrainMaterial.before.cs. material-candidate.json records its SHA-256, candidate SHA-256 and unchanged orthophoto hash. Rollback consists of restoring this file and the separately archived original R mask. Native comparison performed; visual acceptance rejected.

Native candidate result, 16 September 01:13 Vienna

The actual native report and invocation belong to the 01:13:28 Windows build, assembly SHA-256 80D79B9BE9F99469B1633881763D5C02B1A558E3DAC221BE91E95FAC6FFE2DD0. Alps01 completed 3,625.6477 m in 1,486 rendered frames with zero crashes, recoveries or missed checkpoint frames; stable 1920×1080, exit 0, unchanged production profiles and no driver events. Intel UHD / D3D11 averaged 19.777 fps, with a 7.810 fps 1% low. This is a functional 1/1 pass, not a graphics or performance acceptance.

Root reviewed the actual captured images and rejected the art result: open fields still look too bright, lime green and flat. Material work continues; later source edits are not claimed to be included or tested in this build. No new Alps02/03, full-campaign or memory pass is attributed to this candidate.

Revision 2: authored summer grass SetColor inputs

The actual Docs/QA/20260916-011329-campaign/alps-01-pass0-view0.png and view1.png remain excessively uniform yellow-green. The mask and retained Layer1 variant were checked: the photo is active, but the wide 100–400-metre road exclusion leaves large visible meadows dominated by Grass005. Its measured mean linear RGB is (0.1352141, 0.2357573, 0.0265343), approximately 2.16 times the luminance of the existing Grass001 source.

R2 assigned only summer Layer0 the authored SetColor inputs (0.45, 0.46, 0.85), retaining the Grass005 blade texture. The former prediction (0.06085, 0.10845, 0.02255) and claimed match to Grass001's (0.06032, 0.10911, 0.02263) are superseded: that calculation incorrectly treated the inputs as direct shader-linear multipliers. Under Unity's documented material-colour conversion, the expected multiplier is approximately (0.170645, 0.178868, 0.692071), giving a source mean of approximately (0.023074, 0.042169, 0.018364) before lighting. These revised numbers are documentation-derived expectations, not measured GPU constants or final image values. The original texture bytes, normals, roughness and 1.4-metre scale remain unchanged. This is an explicit art choice following the native image review, not a colour-space bug fix or global exposure adjustment. Winter Layer0 stays unchanged.

The regional agent separately prepares an 18–75-metre all-route R transition to expose meadow macrostructure between route loops; closer aerial-image artifacts require the next native review. No mask is changed by the material edit. The exact immediately preceding material source is in ArtSource/Alps/SummerMeadow/GeographicMeadowCandidate/Revision2/AlpineTerrainMaterial.before.cs, with before/after byte hashes in that folder's material-candidate.json. Revision 2 is a bounded candidate, not an accepted final image.

Revision 3: summer near/far colour consistency

The native R2 run in Docs/QA/20260916-013504-campaign passed its two functional cases, but the visual result remains unaccepted. Its actual alps-01-pass0-view0.png was inspected: roadside grass is darker and less glaring, yet too gray; large open photographic fields still appear pale or washed white. This does not establish a UV or gamma defect. The photograph contains its own acquisition lighting and is not de-lit PBR albedo.

R3 changes only the three summer authored SetColor input triplets: Layer0 (0.58, 0.64, 0.60), Layer1 (0.45, 0.58, 0.32), and neutral Layer2 (0.70, 0.70, 0.70). This is an authored summer palette: near grass regains green, photographic meadows move toward the existing summer crown palette, and the bright rock source receives a neutral colour adjustment. These triplets are not presented as direct shader-linear multipliers. Winter Layers0/1/2, Layer3, source photographs, masks, scale, DEM, roads and global lighting are unchanged.

The exact R2 material source and hash are archived in ArtSource/Alps/SummerMeadow/GeographicMeadowCandidate/Revision3/AlpineTerrainMaterial.before.cs and material-candidate.json. R3 was subsequently observed in the 02:22 combined run documented below; it remains short of final graphics acceptance.

Before the R3 build, summer Layer1 also receives the existing AlpineMeadow_detail texture to complement the nearer 18–75-metre photo transition. Its generator, texture hash and pinned HDRP sampling agree: R is centred photographic luminance, G normal Y, B neutral smoothness and A normal X; the texture imports as linear data with Repeat wrapping. LinkDetailsWithBase1=0 and independent detail ST 1/1.4 retain a 1.4-metre detail scale under the layer's world-planar scale of 1. Albedo strength is 0.28, normal strength 0.38 and smoothness strength 0. No original photo pixel or winter layer changes. This adds local surface response; it cannot reconstruct missing aerial survey detail or establish native quality before the next image review.

Colour-space audit and preserved candidate values

The pinned HDRP 17.3.0 LayeredLit properties are ordinary Color values; this project uses Linear colour space. Unity 6.3 documents conversion of material sRGB colours when passed to shaders. The SetColor API description concerns the storage step; reading back a CPU material value would not by itself establish the GPU constant. HDRP then directly multiplies the sampled base colour by the supplied constant in LitDataIndividualLayer.hlsl.

R2/R3 material-candidate.json records now distinguish authored inputs from expected shader-linear values, preserve the original R2 calculation as superseded, and retain the original JSONs as material-candidate-before-color-space-audit.json. R3 records the new source hash after a comments-only correction, its preceding hash and exact equality of the non-comment source. All R3 numeric values remain unchanged; no automatic .gamma conversion was applied. The source mean statistics remain offline measurements; final shading also depends on mask/detail blending, filtering, lighting and tone mapping. Native R2 is a functional 2/2 pass with unaccepted art; R3's later native review is recorded below; the material values remain unchanged by the colour-space audit.

Native R3, corridor and cut result, 16 September 02:22 Vienna

The 02:22:47 native report and invocation match the 02:22:46 Windows build, assembly SHA-256 91D6BB40B36C858FBB59826AEFFFAAC8F707F94ED82FE4975404598552C8809F, 524,279,386 build bytes. Both actual solo rides passed: Alps01 3,625.6477 m / 1,567 frames, Alps02 4,024.8096 m / 1,655 frames. Each has zero crashes, recoveries, missed checkpoint frames or report errors, stable 1920×1080, unchanged production profiles and exit 0 with no driver events. Intel UHD / D3D11 averages 20.901 / 20.030 fps and 1% lows 8.958 / 9.292 fps. Production AI drives the real BikeController without position correction. This time-trial fixture seeds alps-01; it is not a fresh campaign-progression test.

The Player log confirms the additional corridor photograph at 8192×8192, BC7, 14 mip levels, retaining the original geology and winter source. It also records the bounded Alps02 visual cut: 9 affected meshes, 26,267 additional render triangles, maximum relief 0.3499995 m. All nine original collider references and vertex/index hashes are retained. This is evidence of construction bounds and collision preservation, not proof of final landscape quality.

Root inspected the actual Alps01 view0, view1 and Alps02 view1. The combination is kept as an incremental improvement: less glaring near grass, greener field variation, less white peaks and a softer cut toe. The cut still reads as a hard, very regular plane; forest silhouettes remain sparse and dark. Photorealism/AAA is not accepted. Because colour, photograph resolution and bounded cut geometry were evaluated together, this is not an isolated attribution of every visual improvement to R3 colour alone.

No current-build Alps03/full-campaign, hardware-target or memory-plateau pass is claimed. The 00:41 memory result belongs to the older 64F078F3… assembly and predates the additional 8k photo. Remaining M0–M5 acceptance and architecture gaps are unchanged; see implementation status.