# Original Milky Way density, cloud generator and material

This is offline evidence and an original-source-space rendering port. It does not establish complete original presentation or live game parity. No game functions, network requests, process control or OS input are used by these extractors or tests.

## Resource identity and exact integers

The source is the same original resource used by GZM and constellations, `d30f11c57448cff8ec3dd0c5c80d9868482ccadb`. Decoded FRES SHA256 is `745b9cfbeb501fc83eba4e64822ee7a2953d1cf1c293afc1ed5c32469a3fdf2d`; its82,232,836-byte frozen body SHA256 is `49a18ad1f339169391abe2a76475eea6731028b9809d7120fb32d57309196f22`. It contains3227 frozen pools,24 roots and814013 references, with pool bytes beginning at13127884.

The actual GDM root is name/type2854198236/193492573 at pool2075:2048. It is different from stardata1765098201/2090718614 at2647:4032, which supplies the original Sol offset `(49985,40985,24105)` light years. The GDM contains original14 integer coefficients and physical extents `(100000,81910,100000)`. Do not reinterpret stardata's string references at138/140 as density grids.

`Extract-OriginalGalaxyDensity.py` copies four2048² inputs without normalization: field38 mass-zero density with32-bit RLE words, field78 dust surface integral with16-bit words, field138 Milky Way surface with16-bit words, and field140 radial/vertical integral with16-bit words. Each row is expanded through its actual relocation pointer, with nonzero bounded run lengths, and the sum of encoded row bytes plus2048×8 original row pointers must equal the original descriptor's exact byte count. Both16-bit and32-bit RLE rules are needed; treating all channels as32-bit misreads the first run.

The embedded40MiB integer payload has SHA256 `55aa2bc69d9ae86a4262145ab5346a6212a0f77f9929debe228cae3c5b1298c5` and deterministic gzip size10,596,407bytes. This is intentionally different from the earlier exported normalized one-channel overview.

## Source interface and pure sampling

All function addresses in this document refer to pinned original image SHA256 `6891f70a021fa9f3a0e00458c284f48d1226e68ae7fc02ad48081aed47857e93`. `Extract-GalaxyMilkyWayEvidence.py` verifies its complete image hash, actual source-interface vtable entries, constants and hashes of22 decompiled functions, writing `power-outer-sources/milkyway-source-evidence.json`.

The generator's service is global5F05C18/interface5539D08 at component+60. Its resource pointer is interface+F8, the same component+158 accessed by5F05C1C interface+100 in the existing native overview exporter. Confusing these two registered interfaces gives different vtable layouts. Milky Way provider global5FA0A20/interface55272B0 is a separate component at+58, constructor3A08F00.

Original virtual68→3C41ED0→4CA5940 reads field138 at the centre of the original10ly cell using the original rounded fixed-point raster coefficient and16-bit bilinear sampler4CA5F30. Virtual78→3C41500→4CA5370 bilinearly reads mass-indexed density and combines it with4CA5A20's radial/vertical integral. The latter samples the original pre-integrated field140 at eight cell corners; it is not an estimated exponential height falloff. Virtual70→3C427C0→4CA5A20 provides that integral. Virtual88→3C411C0→virtual90→3C411F0→4CA5240 averages four field78 samples for the original dust density.

`GalaxyOriginalDensity` ports these integer multiplication, shift and truncation operations for the mass-zero addresses actually produced by the original cloud job. At original Sol's exact mass-zero cell5367098788688, it returns stellar density.9424896, surface.2746277, height integral526 and dust.73916626. At the galactic midplane/centre, stellar density139.5209 is retained rather than clamped to1. Values at absolute origin are zero. These are pinned source fixtures, not measurements of a running game's current renderer.

## Original layer job and geometry

Provider virtual18/20 are3A31D50/3A31B20. They derive bounds through3A64C60, reserve an original `MilkyWayLayerJob` task descriptor5FD85C8, and submit48-byte arguments. Descriptor vtable551E640+8 reaches39EFBA0→3A469D0. The tool never invokes these requests or their job; `GalaxyOriginalClouds` is a pure local port.

Constructor defaults are step1066ly, Milky Way height16000ly, dust height1280ly, one Milky Way layer and three dust layers. The source Y midplane is40955ly, distinct from Sol's40985ly. Grid points are quantized by layer, encoded into original mass-zero cell addresses, tested against the actual density channels, and jittered by original cell hash and Box-Muller helper39F25F0. The Park-Miller stream uses48271/44488/3399. The original Gaussian call order is Z,Y,X. Size/angle/atlas use a separate sequence from the same cell seed, not the post-Gaussian state.

The resulting28-byte records contain XYZ, raw alpha, raw size, rotation and atlas index. They are diffuse cloud records and are never added to `GalacticSystem`, query results, hit tests, system labels or stellar counts. Original MW raw size spans.75..1.5; renderer constructor defaults then multiply it by31×32 for world half-size. Original default raw-alpha clamp is.046 and multiplier10. The near/far bounds are200/300000ly; default transition width is1066×100. Original56-byte draw records convert rotation to half but retain size as float.

The Sol fixture samples141376 original Milky Way cells and emits15824 records in about42ms after data load. Dust layers sample8836/2304/2304 cells and emit140/100/142 records. The first original data load is about73ms. These counts certify deterministic constructor-default local generation over this frozen source, not runtime quality/debug overrides or live-game output.

## Textures, shader and current rendering

`Build-OriginalGalaxyMilkyWay.py` copies four actual BC1 textures with every padded mip from the already hash-verified original material bundle19b9e393…:1280×2048 diffuse atlas,512² MW RGB,1024×256 MW height RGB, and1024×256 dust height RGB. The combined2,324,480-byte payload SHA256 is `c2755a034d7eb251220ef83b8534eea346218f1d428c1b87d1cdb74a3e2abaaa`; gzip is666,162bytes. Internal texture format95 maps to BC1_UNorm through the independently recovered original graphics format table; it is not DXGI95.

The actual atlas table is render config295:14000+80→622:12672,15 original ushort entries. Columns/rows atC0/C1 are **3×5**, despite the texture's aspect ratio. High byte is atlas X and low byte Y. No atlas coordinates are inferred from image proportions.

The original `DistantMilkyWay` DXBC has been disassembled into `power-outer-sources/effects/`. The production `GalaxyMilkyWayGraphics.cs` and new VSMilkyWay/PSMilkyWay use its source-facing billboard basis, half rotation/sincos, unit corner geometry, near/far transition, camera-height alpha, exact RGB/Y texture lookups and squared colour combination. Actual original map-space coordinates are retained; only the actual tool camera's focal projection is supplied. The diffuse atlas uses original sample alpha and bright clamp with a source-alpha/one accumulation pass. It is separate from the local stellar-stream query bounds and from catalogue filtering. The captured256px cockpit sky cube remains absent.

`artifacts/galaxy-milkyway-shader-validation` compiles and renders the **production partial**, original textures and production HLSL on D3D11 WARP, with Sol, galactic-centre and above-galaxy cameras. All17 production shader entry points compile. Original1600×900 captures and a Sol detail crop were viewed: the source creates structured warm light, clouds and spiral structure, without a magnified sky cube, square backgrounds or synthetic stars. The central portions of the near views remain substantially dark; this is not a visual-equivalence pass against the user's grey reference.

## Explicit remaining limits

`OriginalMilkyWayFullRestored=false`. Original3AB44E0's render-config90 vector `(.29,.05,.07,2.79)` is now bound by actual constructor names to `gAlphaFadeRange/Radius/RangeFar/RadiusFar`; these are not the MW CB2 brightness/height parameters. `Extract-GalaxyMaterialPropertySlots.py` verifies name-load/default-store instructions and the common named publisher. Current MW rendering retains its authored FGM brightness/height values. Shared CB1[90].y/z are now identified as HDR Write/Read, with constructor defaults1/256 and256. The default U2 final transform and production MW Write input have actual original-PS comparisons in `galaxy-original-hdr-evidence.md`. Exact runtime global fade values, material/quality overrides, post-process options and complete live presentation remain unclosed. No brightness value is guessed to match a screenshot.

`OriginalDustEnabled=false`. The raw angle,56-byte Half4 layout and actual Dust shader-key association are now independently closed. The original VS really evaluates `sqrt(1-x²)` on producer angles beyond1; actual original DXBC WARP tests observe those records producing no RGB fragments. Its frame blend is Zero/SourceColor and authored pDustColour0 gives an RGB identity under the reciprocal original HDR defaults. No sine/clamp or invented dark tint is inserted. Later runtime tint overrides and complete multi-layer fades remain unknown; see`galaxy-original-dust-evidence.md`. Original Plane/Core and the default U2 final-HDR path now have their own source proofs and actual VS/PS comparisons.

Three independent Core background regressions pass: original coordinate/integer sampling, BC1 atlas/mip data, and deterministic/cancellable generation. No test success is presented as complete original appearance or live coverage.