fix(3d): restore correct orientation in test_3d_render (direct-to-surface)

Add Renderer3D::set_direct_render(bool) flag (must be set before init()).
When true, uses standard CCW winding and un-negates perspective Y, restoring
the pre-ba7ea27 orientation for direct-to-surface rendering where the
post-process Y-flip is absent.

handoff(Gemini): set_direct_render() is the escape hatch for any future
renderer usage that bypasses the post-process chain.

4 files changed, 16 insertions, 4 deletions

diff --git a/src/3d/renderer.h b/src/3d/renderer.h
index 11a7931..8f933b5 100644
--- a/src/3d/renderer.h
+++ b/src/3d/renderer.h
@@ -78,6 +78,10 @@ class Renderer3D {
     bvh_enabled_ = enabled;
   }
 
+  // Call before init(). When true, renders directly to surface (no post-process
+  // Y-flip): uses standard perspective (Y not negated) and CCW winding.
+  void set_direct_render(bool v) { direct_render_ = v; }
+
   struct MeshGpuData {
     WGPUBuffer vertex_buffer;
     WGPUBuffer index_buffer;
@@ -120,6 +124,7 @@ class Renderer3D {
 
   BVH cpu_bvh_; // Keep a CPU-side copy for building/uploading
   bool bvh_enabled_ = true;
+  bool direct_render_ = false; // true = render to surface (no post-process flip)
 
   std::map<AssetId, MeshGpuData> mesh_cache_;
   const MeshGpuData* temp_mesh_override_ = nullptr; // HACK for test_mesh tool
diff --git a/src/3d/renderer_draw.cc b/src/3d/renderer_draw.cc
index d4a35a8..dca7113 100644
--- a/src/3d/renderer_draw.cc
+++ b/src/3d/renderer_draw.cc
@@ -9,7 +9,9 @@
 
 void Renderer3D::update_uniforms(const Scene& scene, const Camera& camera,
                                  float time) {
-  const mat4 vp = camera.get_projection_matrix() * camera.get_view_matrix();
+  mat4 proj = camera.get_projection_matrix();
+  if (direct_render_) proj.m[5] = -proj.m[5]; // undo Y-negate for direct surface
+  const mat4 vp = proj * camera.get_view_matrix();
   const GlobalUniforms globals = GlobalUniforms::make(
       vp,
       vec4(camera.position.x, camera.position.y, camera.position.z, time),
@@ -203,8 +205,9 @@ void Renderer3D::draw(WGPURenderPassEncoder pass, const Scene& scene,
       }
     }
 
-    // Calculate ViewProj matrix for the debug renderer
-    mat4 view_proj = camera.get_projection_matrix() * camera.get_view_matrix();
+    mat4 dbg_proj = camera.get_projection_matrix();
+    if (direct_render_) dbg_proj.m[5] = -dbg_proj.m[5];
+    mat4 view_proj = dbg_proj * camera.get_view_matrix();
     visual_debug_.render(pass, view_proj);
   }
 #endif
diff --git a/src/3d/renderer_pipelines.cc b/src/3d/renderer_pipelines.cc
index 3abc3bd..195baad 100644
--- a/src/3d/renderer_pipelines.cc
+++ b/src/3d/renderer_pipelines.cc
@@ -253,7 +253,10 @@ void Renderer3D::create_mesh_pipeline() {
   pipeline_desc.vertex.buffers = &vert_buffer_layout;
   pipeline_desc.primitive.topology = WGPUPrimitiveTopology_TriangleList;
   pipeline_desc.primitive.cullMode = WGPUCullMode_Back;
-  pipeline_desc.primitive.frontFace = WGPUFrontFace_CW;  // Y-flipped perspective
+  // CW winding compensates for Y-negation in perspective() (post-process chain).
+  // Direct-to-surface rendering uses standard CCW winding.
+  pipeline_desc.primitive.frontFace =
+      direct_render_ ? WGPUFrontFace_CCW : WGPUFrontFace_CW;
   pipeline_desc.multisample.count = 1;
   pipeline_desc.fragment = &fragment_state;
   pipeline_desc.depthStencil = &depth_stencil;
diff --git a/src/tests/3d/test_3d_render.cc b/src/tests/3d/test_3d_render.cc
index de9718e..ef799ec 100644
--- a/src/tests/3d/test_3d_render.cc
+++ b/src/tests/3d/test_3d_render.cc
@@ -140,6 +140,7 @@ int main(int argc, char** argv) {
 
   InitShaderComposer();
 
+  g_renderer.set_direct_render(true); // Renders to surface, no post-process flip
   g_renderer.init(g_device, g_queue, g_format);
   g_renderer.resize(platform_state.width, platform_state.height);