XFE Git
XFE Studio Git
Git 首页 全局搜索
XFE 主站 文档 NuGet

XFEEDToolBox

【WinUI】XFE精英危险工具箱

公开
关注 0 Fork 0 Star 0
UTF-8

Original authored Galaxy background models

This is read-only evidence from the pinned4.4.2.0 client, SHA2566891f70a021fa9f3a0e00458c284f48d1226e68ae7fc02ad48081aed47857e93. No running-game memory, game functions or network requests were used. The current executable has a different hash; these are source/resource proofs, not permission to use old RVAs in the current native adapter.

Extract-GalaxyAuthoredModels.py verifies constructor3A06740's actual main vtable551F2D8. Its slot10 points to complete initializer3A5BAF0. Earlier source at3A5BD80 was only a middle fragment of that initializer. The original function loads models through3B067D0 into the same renderer's model array6D0:

Model Index Scale Axes Original world centre
MilkyWayCore 0 40000 identity 50000,40955,50000
MilkyWayPlane 1 100000 -X,+Y,-Z 50000,40955,50000
LocalGroupModel 3 1 identity 49985,40985,24105

The Core/Plane centre is15,-30,25895 in the tool's Sol-relative coordinates. This comes from the actual Galaxy renderer initializer, not from another model/map's unit quad. The similar3A5C2D0 implementation separately agrees on these transforms but is not used as proof of active Galaxy class identity.

Mesh and material resources

Bundle19b9e393c96329f63707aeade6a46e8cf3b71398 has archive SHA2567da2211c8223413f94cb6b32ec9a72845268bbf0d05eeb8f4bcc824491480b96 and frozen stream SHA256f258703e3d5e7fad602ce33ff218bc4e86d0385a0fb43a72b83a14fd9df3ce87. Its model buffers are column-major, with separate material name,304-byte mesh metadata and GPU blocks. Both use vertex format8,40-byte packed vertices and16-bit triangle indices after a24-byte GPU header.

Plane has4 vertices and6 indices. Its original material isMilkyWayPlane, shaderAlpha_UnlitNoScale_StarMapFacingAlpha_NearFade, with2048² BC7-sRGB RGBA texture. Its packed positions bound an XZ plane of approximately±.5. The selected VS alias,40-byte input format and336-byte packed instance have since been closed by actual loader keys/native producers; seegalaxy-original-plane-evidence.md and its six-view original VS/PS comparison. The incompatible generic Alpha_UnlitNoScale VS was explicitly rejected as a substitute.

Core has150 vertices and576 indices, packed position bounds approximately±2.54679. The original materialMilkyWayCoreMat uses shaderGalaxyGlow, and its16² BC1 noise texture retains all5 padded mip levels. Build-OriginalGalaxyCore.py exports exact source-derived positions/indices and3072 texture bytes with hashes, independently of the FGM's CPU parameter storage offsets.

GalaxyGlow arithmetic and comparison

The actual VS reads only PACKEDVERTEXDATAA. The existing packed-position decoder matches the original bit extraction, including its per-vertex scale exponent and radial format64 branch. The VS emits clip.xyw and the transformed world/view vector. The PS evaluates two25-sample ellipsoid integrals, not an arbitrary background colour:

  • Inner radius20000, shape(1.871,.273,1.859), colour(.874,.686,.503,.407), scale9.7 and Gaussian power.03.
  • Outer radius80000, shape(.95,.09,.95), colour(.601,.603,.673,.3), scale10 and power.16.
  • Internal amount.5, distance blend(.1,.848), original noise/dither(.5552,8).

The named FGM values and the original PS vector operations independently constrain the8 material vectors. The source PS applies sqrt(CB1[90].z) and then CB1[90].y. The independent source-space probe retains explicit values1. Their names have since been proven as original HDRScalingRead/Write, with constructor defaults256 and1/256 and the actual embedded final U2 PS independently compared. Seegalaxy-original-hdr-evidence.md; runtime values/options remain unread.

GalaxyCoreShaderProbe renders the actual source mesh with both the production port and the original VS+PS DXBC. It compares an R32G32B32A32_FLOAT render target before SDR clamping. At640×360, Sol, centre and above-plane cases passed with maximum component differences3.57e-5,9.96e-5 and3.22e-5, mean differences below2.5e-7, and no nonfinite values. SourceColor/One was used explicitly on both comparison paths. This verifies the shader port under those controlled inputs, not live game parity.

Original513D70 creates D3D11 blend state through table4E2E460. Enum2 maps RGB to SourceColor and alpha to SourceAlpha; enum1 maps both to One. Core's FGM retains Src2/Dest1/Op0. The compiled effect bytecode now independently proves these named fields are copied to the actual blend descriptor (material-slots/original-blend-application.json). Full frame scheduling and runtime material overrides remain separate. OriginalCoreFullRestored remains false.

The Sol, centre and above PNGs underartifacts/galaxy-core-shader-validation are actual offscreen WARP output. The Sol image was inspected: its grey/purple background and warm core are produced by the original shader equations, not by a grey clear colour or added star points.

Dust boundary

Cloud job3A469D0 writes a raw angle0..2π at record14. The original3A21BC0 converts that angle directly to half4; format38 is Half4, not UNorm. The actual shader-key binding and56-byte input descriptor have now both been closed: the original DistantMilkyWayDust VS really evaluates sqrt(1-x*x) on that value. Actual original VS/PS WARP tests observe out-of-domain geometry producing no RGB fragments. No guessed sine, clamp or random rotation replacement is introduced. Seegalaxy-original-dust-evidence.md.

Initial construction3A643D0 temporarily supplies primary MilkyWay to the Dust quad because its runtime handle is created afterwards in3A5BAF0. Every-frame3AB44E0 explicitly publishes runtime950 and replaces Dust2E0's material through6CA5B0. It also overrides the blend toSrc0/Dest2. With the actual authored pDustColour0 and reciprocal default HDR scales, the original PS is an RGB identity; this is independently tested using original DXBC. Later runtime tint overrides and full multi-layer fade remain unknown, so production Dust stays disabled without inventing visible absorption.