diff --git a/mpcore/src/main/cpp/emu/gles_shim.cpp b/mpcore/src/main/cpp/emu/gles_shim.cpp
index 13de449..bc884cd 100644
--- a/mpcore/src/main/cpp/emu/gles_shim.cpp
+++ b/mpcore/src/main/cpp/emu/gles_shim.cpp
@@ -4,6 +4,7 @@
 #include <cstring>
 #include <vector>
 #include <set>
+#include <tuple>
 #include <unordered_map>
 #include <atomic>
 #include <thread>
@@ -63,6 +64,10 @@ uint32_t GuestCallerLR(GuestEngine& eng) {
 // eglSwapBuffers - see Shim_glClear's own comment for why presenting a
 // back buffer that received none of them is what made the screen cycle
 // between stale, black and real frames.
+// Task #52 probe switch - see Shim_glViewport. Opt-in per this
+// project's standing rule that diagnostics default to off.
+constexpr bool kLogViewportChanges = false;
+
 std::atomic<int> g_drawsSinceSwap{0};
 
 // Shadowed GL state (2026-09-19). These exist so the per-draw code paths
@@ -1722,7 +1727,36 @@ uint32_t Shim_glVertexAttribPointer(GuestEngine& eng, uint32_t r0, uint32_t r1,
 }
 
 uint32_t Shim_glViewport(GuestEngine& eng, uint32_t r0, uint32_t r1, uint32_t r2, uint32_t r3, uint32_t sp) {
-    glViewport((GLint)eng.ReadIncomingArg(0, r0, r1, r2, r3, sp), (GLint)eng.ReadIncomingArg(1, r0, r1, r2, r3, sp), (GLsizei)eng.ReadIncomingArg(2, r0, r1, r2, r3, sp), (GLsizei)eng.ReadIncomingArg(3, r0, r1, r2, r3, sp));
+    GLint vx = (GLint)eng.ReadIncomingArg(0, r0, r1, r2, r3, sp);
+    GLint vy = (GLint)eng.ReadIncomingArg(1, r0, r1, r2, r3, sp);
+    GLsizei vw = (GLsizei)eng.ReadIncomingArg(2, r0, r1, r2, r3, sp);
+    GLsizei vh = (GLsizei)eng.ReadIncomingArg(3, r0, r1, r2, r3, sp);
+    // Task #52 probe (2026-09-19, temporary). A Xiaomi 14 capture showed the
+    // game rendering into viewport=[0,0,2136,960] while the surface was
+    // 2670x1200 - i.e. a reduced internal resolution being upscaled. Whether
+    // that is the game's own choice or a wrong size we fed it is unknown, and
+    // the project already has form here: task #39 was exactly this shape (the
+    // UI projection used 2000x1000 instead of the real 2400x1080 because a
+    // JNI float argument was arriving as zero). Logging each DISTINCT
+    // viewport against the real EGL surface size, so the comparison is a fact
+    // rather than an inference.
+    if (kLogViewportChanges) {
+        static std::mutex mu;
+        static std::set<std::tuple<GLint, GLint, GLsizei, GLsizei>> seen;
+        std::lock_guard<std::mutex> lock(mu);
+        if (seen.insert({vx, vy, vw, vh}).second && seen.size() <= 24) {
+            EGLint sw = -1, sh = -1;
+            EGLDisplay dpy = eglGetCurrentDisplay();
+            EGLSurface surf = eglGetCurrentSurface(EGL_DRAW);
+            eglQuerySurface(dpy, surf, EGL_WIDTH, &sw);
+            eglQuerySurface(dpy, surf, EGL_HEIGHT, &sh);
+            GLint fb = g_shadowBoundFramebuffer.load(std::memory_order_relaxed);
+            Log("gles_shim: VIEWPORT [%d,%d %dx%d] fb=%d | EGL surface %dx%d%s",
+                vx, vy, vw, vh, fb, sw, sh,
+                (fb == 0 && (vw != sw || vh != sh)) ? "  <<< SMALLER THAN THE SCREEN >>>" : "");
+        }
+    }
+    glViewport(vx, vy, vw, vh);
     return 0;
 }