fix(cnn_v3): shadow pass — 5 bugs fixed, labels in gbuf_viewHEADmain

1. Camera Y-inversion: proj.m[5] = -proj.m[5] in upload_scene_data
   + WGPUFrontFace_CCW on raster pipeline.
2. Shadow formula: replace shadowWithStoredDistance with 64-step
   IQ soft shadow (8*d/t, unbounded).
3. Local→world SDF scale: d *= length(obj.model[0].xyz).
4. Shadow bias: use rasterized normal from normal_mat_tex (binding 4)
   instead of light direction — fixes terminator self-shadow on spheres.
5. ShaderComposer: GBufViewEffect now resolves #include via
   ShaderComposer::Get().Compose().

Also: per-tile channel labels in gbuf_view.wgsl via debug_str.
Scene simplified to 1 cube + 1 sphere for debugging (restore TODO).
Scale propagation for pulsating sphere confirmed correct end-to-end.

handoff(Gemini): shadow validated. Next: restore full scene in
GBufferEffect::set_scene() (20 cubes + 4 spheres, 2 lights), then
run training pass per cnn_v3/docs/HOWTO.md §3.

2 files changed, 36 insertions, 46 deletions

diff --git a/cnn_v3/src/gbuf_view_effect.cc b/cnn_v3/src/gbuf_view_effect.cc
index 180919d..ccf80b0 100644
--- a/cnn_v3/src/gbuf_view_effect.cc
+++ b/cnn_v3/src/gbuf_view_effect.cc
@@ -10,6 +10,7 @@
 #endif
 
 #include "gpu/gpu.h"
+#include "gpu/shader_composer.h"
 #include "util/asset_manager.h"
 #include "util/fatal_error.h"
 
@@ -63,7 +64,8 @@ GBufViewEffect::GBufViewEffect(const GpuContext& ctx,
   // Shader module
   WGPUShaderSourceWGSL wgsl_src = {};
   wgsl_src.chain.sType = WGPUSType_ShaderSourceWGSL;
-  wgsl_src.code        = str_view(gbuf_view_wgsl);
+  const std::string composed = ShaderComposer::Get().Compose({}, gbuf_view_wgsl);
+  wgsl_src.code        = str_view(composed.c_str());
   WGPUShaderModuleDescriptor shader_desc = {};
   shader_desc.nextInChain = &wgsl_src.chain;
   WGPUShaderModule shader =
diff --git a/cnn_v3/src/gbuffer_effect.cc b/cnn_v3/src/gbuffer_effect.cc
index c49ab88..b059915 100644
--- a/cnn_v3/src/gbuffer_effect.cc
+++ b/cnn_v3/src/gbuffer_effect.cc
@@ -109,50 +109,29 @@ void GBufferEffect::set_scene() {
   };
   auto rrange = [&](float lo, float hi) { return lo + rnd() * (hi - lo); };
 
-  // 20 small cubes scattered in a [-2,2]×[-1.5,1.5]×[-1.5,1.5] volume.
-  static const int kNumCubes = 20;
-  for (int i = 0; i < kNumCubes; ++i) {
+  // 2 large cubes.
+  // 2 large static cubes for shadow debugging.
+  {
     Object3D obj(ObjectType::CUBE);
-    obj.position = vec3(rrange(-2.0f, 2.0f),
-                        rrange(-1.5f, 1.5f),
-                        rrange(-1.5f, 1.5f));
-    const float s = rrange(0.10f, 0.25f);
-    obj.scale = vec3(s, s, s);
-    obj.color = vec4(rrange(0.4f, 1.0f),
-                     rrange(0.4f, 1.0f),
-                     rrange(0.4f, 1.0f), 1.0f);
-
-    // Random rotation axis (avoid degenerate zero-length axis).
-    vec3 axis = vec3(rrange(-1.0f, 1.0f),
-                     rrange(-1.0f, 1.0f),
-                     rrange(-1.0f, 1.0f));
-    if (axis.len() < 0.01f) axis = vec3(0.0f, 1.0f, 0.0f);
-    axis = axis.normalize();
-    const float speed = rrange(0.3f, 1.5f) * (rnd() > 0.5f ? 1.0f : -1.0f);
-
+    obj.position = vec3(-1.0f, 0.0f, 0.0f);
+    obj.scale    = vec3(0.6f, 0.6f, 0.6f);
+    obj.color    = vec4(0.9f, 0.5f, 0.3f, 1.0f);
     scene_.add_object(obj);
-    cube_anims_.push_back({axis, speed});
+    cube_anims_.push_back({{0.0f, 1.0f, 0.0f}, 0.0f});
   }
-
-  // 4 pumping spheres at fixed positions; radius modulated by audio_intensity.
-  static const vec3 kSpherePos[4] = {
-    { 0.0f,  0.0f,  0.0f},
-    { 1.5f,  0.5f, -0.5f},
-    {-1.5f, -0.5f,  0.5f},
-    { 0.0f,  1.0f,  1.0f},
-  };
-  static const float kBaseSphereRadius[4] = {0.35f, 0.28f, 0.30f, 0.25f};
-  for (int i = 0; i < 4; ++i) {
+  {
     Object3D obj(ObjectType::SPHERE);
-    obj.position = kSpherePos[i];
-    const float r = kBaseSphereRadius[i];
-    obj.scale = vec3(r, r, r);
-    obj.color = vec4(0.85f, 0.60f, 0.95f, 1.0f);
+    obj.position = vec3(1.0f, 0.0f, 0.0f);
+    const float r = 0.9f;
+    obj.scale    = vec3(r, r, r);
+    obj.color    = vec4(0.3f, 0.6f, 0.9f, 1.0f);
     const int idx = (int)scene_.objects.size();
     scene_.add_object(obj);
     sphere_anims_.push_back({idx, r});
   }
 
+  // (sphere removed for shadow debugging)
+
   // Camera: above and in front of the scene, looking at origin.
   camera_.set_look_at(vec3(0.0f, 2.5f, 6.0f),
                       vec3(0.0f, 0.0f, 0.0f),
@@ -213,13 +192,13 @@ void GBufferEffect::render(WGPUCommandEncoder encoder,
   // Upload two directional lights.
   {
     GBufLightsUniforms lu = {};
-    lu.params = vec4(2.0f, 0.0f, 0.0f, 0.0f);
+    lu.params = vec4(1.0f, 0.0f, 0.0f, 0.0f);
     // Key: warm sun, upper-right-front.
     lu.lights[0].direction = vec4(0.408f, 0.816f, 0.408f, 0.0f);  // norm(1,2,1)
     lu.lights[0].color     = vec4(1.00f, 0.92f, 0.78f, 1.0f);
-    // Fill: cool sky, upper-left-back.
-    lu.lights[1].direction = vec4(-0.577f, 0.577f, -0.577f, 0.0f);  // norm(-1,1,-1)
-    lu.lights[1].color     = vec4(0.40f, 0.45f, 0.80f, 0.4f);
+    // Fill: cool sky, upper-left-back. (disabled for debugging)
+    // lu.lights[1].direction = vec4(-0.577f, 0.577f, -0.577f, 0.0f);
+    // lu.lights[1].color     = vec4(0.40f, 0.45f, 0.80f, 0.4f);
     lights_uniform_.update(ctx_.queue, lu);
   }
 
@@ -301,7 +280,7 @@ void GBufferEffect::render(WGPUCommandEncoder encoder,
 
   // --- Pass 2: SDF shadow raymarching ---
   if (shadow_pipeline_.get() != nullptr) {
-    WGPUBindGroupEntry shadow_entries[4] = {};
+    WGPUBindGroupEntry shadow_entries[5] = {};
     shadow_entries[0].binding = 0;
     shadow_entries[0].buffer  = global_uniforms_buf_.buffer;
     shadow_entries[0].size    = sizeof(GBufGlobalUniforms);
@@ -317,12 +296,15 @@ void GBufferEffect::render(WGPUCommandEncoder encoder,
     shadow_entries[3].buffer  = lights_uniform_.get().buffer;
     shadow_entries[3].size    = sizeof(GBufLightsUniforms);
 
+    shadow_entries[4].binding     = 4;
+    shadow_entries[4].textureView = normal_mat_view;
+
     WGPUBindGroupLayout shadow_bgl =
         wgpuRenderPipelineGetBindGroupLayout(shadow_pipeline_.get(), 0);
 
     WGPUBindGroupDescriptor shadow_bg_desc = {};
     shadow_bg_desc.layout     = shadow_bgl;
-    shadow_bg_desc.entryCount = 4;
+    shadow_bg_desc.entryCount = 5;
     shadow_bg_desc.entries    = shadow_entries;
 
     WGPUBindGroup shadow_bg =
@@ -448,7 +430,8 @@ void GBufferEffect::upload_scene_data(const Scene& scene,
                 : (size_t)kGBufMaxObjects);
 
   const mat4 view = camera.get_view_matrix();
-  const mat4 proj = camera.get_projection_matrix();
+  mat4 proj = camera.get_projection_matrix();
+  proj.m[5] = -proj.m[5];  // undo post-process Y flip: G-buffer uses integer reads
   const mat4 vp   = proj * view;
 
   GBufGlobalUniforms gu = {};
@@ -554,7 +537,7 @@ void GBufferEffect::create_raster_pipeline() {
   pipe_desc.depthStencil = &ds;
   pipe_desc.primitive.topology  = WGPUPrimitiveTopology_TriangleList;
   pipe_desc.primitive.cullMode  = WGPUCullMode_Back;
-  pipe_desc.primitive.frontFace = WGPUFrontFace_CW;  // Y-flipped perspective
+  pipe_desc.primitive.frontFace = WGPUFrontFace_CCW;  // standard (no Y flip)
   pipe_desc.multisample.count = 1;
   pipe_desc.multisample.mask = 0xFFFFFFFF;
 
@@ -584,7 +567,7 @@ void GBufferEffect::create_shadow_pipeline() {
   WGPUShaderModule shader = wgpuDeviceCreateShaderModule(ctx_.device, &shader_desc);
 
   // BGL: B0=GlobalUniforms, B1=ObjectsBuffer, B2=texture_depth_2d, B3=GBufLightsUniforms
-  WGPUBindGroupLayoutEntry bgl_entries[4] = {};
+  WGPUBindGroupLayoutEntry bgl_entries[5] = {};
 
   bgl_entries[0].binding = 0;
   bgl_entries[0].visibility =
@@ -607,8 +590,13 @@ void GBufferEffect::create_shadow_pipeline() {
   bgl_entries[3].buffer.type = WGPUBufferBindingType_Uniform;
   bgl_entries[3].buffer.minBindingSize = sizeof(GBufLightsUniforms);
 
+  bgl_entries[4].binding = 4;
+  bgl_entries[4].visibility = WGPUShaderStage_Fragment;
+  bgl_entries[4].texture.sampleType = WGPUTextureSampleType_Float;
+  bgl_entries[4].texture.viewDimension = WGPUTextureViewDimension_2D;
+
   WGPUBindGroupLayoutDescriptor bgl_desc = {};
-  bgl_desc.entryCount = 4;
+  bgl_desc.entryCount = 5;
   bgl_desc.entries = bgl_entries;
   WGPUBindGroupLayout bgl = wgpuDeviceCreateBindGroupLayout(ctx_.device, &bgl_desc);