diff --git a/goldens/p0_quad.png b/goldens/p0_quad.png
new file mode 100644
index 0000000..bbe5863
Binary files /dev/null and b/goldens/p0_quad.png differ
diff --git a/main.sx b/main.sx
index 0535515..62ac332 100644
--- a/main.sx
+++ b/main.sx
@@ -22,6 +22,60 @@ g_viewport_h : f32 = 600.0;
 g_uikit_plat : *UIKitPlatform = null;
 g_metal_gpu  : *MetalGPU      = null;
 
+// ── Render-path proof (P0.2) ─────────────────────────────────────────────
+// Clear to a solid blue, then draw one centered orange quad covering the
+// central 50%×50% of the drawable. Geometry is in NDC ([-0.5, 0.5]²) so the
+// quad stays screen-size independent across simulator devices, which keeps
+// the screenshot golden unambiguous. This is the GPU protocol's
+// clear+quad path: an MSL pipeline state plus a 6-vertex (2-triangle)
+// buffer, created lazily once the MTLDevice exists.
+g_quad_shader : ShaderHandle = 0;
+g_quad_vbuf   : BufferHandle = 0;
+
+BG_CLEAR :: ClearColor.{ r = 0.10, g = 0.20, b = 0.55, a = 1.0 };
+
+// Vertex layout matches QUAD_MSL's `Vertex`: packed_float2 pos +
+// packed_float4 color = 24 bytes. `packed_*` avoids the 16-byte alignment
+// padding a plain `float4` would force. 6 vertices = 2 triangles.
+QUAD_VERTS : [36]f32 = .[
+    -0.5,  0.5,   1.0, 0.6, 0.0, 1.0,
+     0.5,  0.5,   1.0, 0.6, 0.0, 1.0,
+    -0.5, -0.5,   1.0, 0.6, 0.0, 1.0,
+     0.5,  0.5,   1.0, 0.6, 0.0, 1.0,
+     0.5, -0.5,   1.0, 0.6, 0.0, 1.0,
+    -0.5, -0.5,   1.0, 0.6, 0.0, 1.0,
+];
+
+// Pass-through shader: the vertex stage emits NDC positions directly (no
+// projection), the fragment stage returns the interpolated vertex color.
+// Entry-point names vmain / fmain are what MetalGPU.create_shader looks up.
+QUAD_MSL :: #string MSL
+#include <metal_stdlib>
+using namespace metal;
+
+struct Vertex {
+    packed_float2 pos;
+    packed_float4 color;
+};
+
+struct RasterizerData {
+    float4 position [[position]];
+    float4 color;
+};
+
+vertex RasterizerData vmain(uint vid [[vertex_id]],
+                            constant Vertex* vertices [[buffer(0)]]) {
+    RasterizerData out;
+    out.position = float4(vertices[vid].pos, 0.0, 1.0);
+    out.color = float4(vertices[vid].color);
+    return out;
+}
+
+fragment float4 fmain(RasterizerData in [[stage_in]]) {
+    return in.color;
+}
+MSL;
+
 frame :: () {
     fc := g_plat.begin_frame();
     g_viewport_w = fc.viewport_w;
@@ -49,12 +103,26 @@ frame :: () {
         } else if g_metal_gpu.pixel_w != g_uikit_plat.pixel_w or g_metal_gpu.pixel_h != g_uikit_plat.pixel_h {
             g_metal_gpu.resize(g_uikit_plat.pixel_w, g_uikit_plat.pixel_h);
         }
-        clear : ClearColor = .{ r = 0.12, g = 0.12, b = 0.15, a = 1.0 };
-        if !g_metal_gpu.begin_frame(clear) { return; }
+        // Compile the quad pipeline + upload its vertices once. The MTLDevice
+        // was created eagerly in main(), so both only need a valid device.
+        if g_quad_shader == 0 {
+            g_quad_shader = g_metal_gpu.create_shader(QUAD_MSL, "");
+            if g_quad_shader == 0 { return; }
+        }
+        if g_quad_vbuf == 0 {
+            g_quad_vbuf = g_metal_gpu.create_buffer(size_of([36]f32));
+            if g_quad_vbuf == 0 { return; }
+            g_metal_gpu.update_buffer(g_quad_vbuf, xx @QUAD_VERTS, size_of([36]f32));
+        }
+
+        if !g_metal_gpu.begin_frame(BG_CLEAR) { return; }
+        g_metal_gpu.set_shader(g_quad_shader);
+        g_metal_gpu.set_vertex_buffer(g_quad_vbuf);
+        g_metal_gpu.draw_triangles(0, 6);
         g_metal_gpu.end_frame(fc.target_present_time);
     } else {
         glViewport(0, 0, fc.pixel_w, fc.pixel_h);
-        glClearColor(0.12, 0.12, 0.15, 1.0);
+        glClearColor(0.10, 0.20, 0.55, 1.0);
         glClear(GL_COLOR_BUFFER_BIT);
     }
     g_plat.end_frame();