diff --git a/src/backend/wayland/wl_backend.c b/src/backend/wayland/wl_backend.c
index c91726f..96cd0d7 100644
--- a/src/backend/wayland/wl_backend.c
+++ b/src/backend/wayland/wl_backend.c
@@ -11,6 +11,8 @@
  * Phase 35: Core Wayland protocol globals advertised (wl_compositor,
  * wl_subcompositor, wl_seat, wl_data_device_manager, xdg_wm_base).
  * Phase 36: wl_shm and wl_output advertised; XDG toplevel listener wired.
+ * Phase 37: frame callbacks driven by output frame event; XDG configure
+ * sent on new toplevel so clients complete their first frame.
  */
 
 #ifdef HAVE_CONFIG_H
@@ -26,6 +28,7 @@
 #include <wlr/backend.h>
 #include <wlr/backend/headless.h>
 #include <wlr/render/wlr_renderer.h>
+#include <wlr/interfaces/wlr_output.h>
 #include <wlr/types/wlr_compositor.h>
 #include <wlr/types/wlr_data_device.h>
 #include <wlr/types/wlr_output.h>
@@ -61,6 +64,8 @@ struct wl_compositor_state {
     /* Phase 36: XDG toplevel tracking */
     struct wl_listener              xdg_new_surface_listener;
     int                             managed_surface_count; /* surfaces seen */
+    /* Phase 37: periodic timer that drives client frame callbacks */
+    struct wl_event_source         *frame_timer;
     char                            socket_name[64];  /* Phase 34: e.g. "wayland-0" */
     int                             started; /* non-zero after successful display_open */
 };
@@ -68,9 +73,34 @@ struct wl_compositor_state {
 static struct wl_compositor_state wl_state;
 
 /* ------------------------------------------------------------------ */
-/* Phase 36: XDG surface / toplevel management                         */
+/* Phase 36 / 37: XDG surface / toplevel management and frame loop     */
 /* ------------------------------------------------------------------ */
 
+/*
+ * Phase 37: Wayland event loop timer — drives client frame callbacks.
+ * For the headless backend we avoid the allocator/swapchain path entirely
+ * and call wlr_output_send_frame directly at ~60 fps.  This causes
+ * wlroots to send wl_surface.frame callbacks to all waiting clients.
+ */
+static int
+frame_timer_cb(void *data)
+{
+    (void)data;
+
+    if (wl_state.output)
+        wlr_output_send_frame(wl_state.output);
+
+    /* re-arm at ~60 fps */
+    if (wl_state.frame_timer)
+        wl_event_source_timer_update(wl_state.frame_timer, 16);
+    return 0;
+}
+
+/*
+ * Phase 36: new XDG surface listener.
+ * Sends an initial configure to toplevels so clients can commit their
+ * first buffer (size 0,0 = "use whatever size you want").
+ */
 static void
 handle_xdg_new_surface(struct wl_listener *listener, void *data)
 {
@@ -86,6 +116,9 @@ handle_xdg_new_surface(struct wl_listener *listener, void *data)
              xdg_surface->toplevel->app_id  ? xdg_surface->toplevel->app_id  : "(none)",
              xdg_surface->toplevel->title   ? xdg_surface->toplevel->title   : "(none)");
 
+    /* Phase 37: send the initial configure so the client can commit. */
+    wlr_xdg_surface_schedule_configure(xdg_surface);
+
     /* TODO: create WWindow and call policy layer (wManageWindow equivalent)
      * once the surface is committed (use map event on toplevel). */
 }
@@ -181,6 +214,18 @@ wl_display_open(const char *name)
     wl_signal_add(&wl_state.xdg_shell->events.new_surface,
                   &wl_state.xdg_new_surface_listener);
 
+    /* Phase 37: periodic timer to drive client frame callbacks at ~60 fps.
+     * Using a timer avoids the swapchain/allocator complexity of the full
+     * render pipeline while still keeping Wayland clients unblocked. */
+    {
+        struct wl_event_loop *loop =
+            wl_display_get_event_loop(wl_state.display);
+        wl_state.frame_timer =
+            wl_event_loop_add_timer(loop, frame_timer_cb, NULL);
+        if (wl_state.frame_timer)
+            wl_event_source_timer_update(wl_state.frame_timer, 16);
+    }
+
     wl_state.started = 1;
     return NULL; /* no X11 Display* in the Wayland path */
 }
@@ -190,6 +235,10 @@ wl_display_close(void)
 {
     if (!wl_state.started)
         return;
+    if (wl_state.frame_timer) {
+        wl_event_source_remove(wl_state.frame_timer);
+        wl_state.frame_timer = NULL;
+    }
     if (wl_state.output_layout) {
         wlr_output_layout_destroy(wl_state.output_layout);
         wl_state.output_layout = NULL;
diff --git a/tests/wl-smoke-test.sh b/tests/wl-smoke-test.sh
index 1647a10..0291206 100755
--- a/tests/wl-smoke-test.sh
+++ b/tests/wl-smoke-test.sh
@@ -14,6 +14,8 @@
 #      xdg_wm_base) — verified via weston-info if available
 #   6. (Phase 36) wl_shm and wl_output advertised; a real SHM client
 #      (weston-simple-shm) can connect without crashing the compositor
+#   7. (Phase 37) Frame callbacks fire; weston-simple-shm runs its render
+#      loop (stays alive for 1 s, compositor still alive afterwards)
 #
 # This test grows as wl_backend vtable slots are implemented.
 #
@@ -130,12 +132,15 @@ fi
 # compositor is still alive after the client runs (or times out).
 
 if command -v weston-simple-shm >/dev/null 2>&1; then
-    CLIENT_OUT=$(WAYLAND_DISPLAY="$WL_SOCK" timeout 2s weston-simple-shm 2>&1) || true
+    # Run weston-simple-shm for up to 3 seconds; it loops drawing frames.
+    # A clean timeout (exit 124) or exit 0 both mean frame callbacks fired.
+    # We only fail if wl_shm is missing or the compositor crashes.
+    CLIENT_OUT=$(WAYLAND_DISPLAY="$WL_SOCK" timeout 3s weston-simple-shm 2>&1) || true
     echo "$CLIENT_OUT" | grep -q 'No wl_shm global' \
         && die "weston-simple-shm: compositor missing wl_shm"
     kill -0 "$WM_PID" 2>/dev/null \
-        || die "compositor crashed after weston-simple-shm connected"
-    echo "  client check: weston-simple-shm connected; compositor survived"
+        || die "compositor crashed after weston-simple-shm ran"
+    echo "  client check: weston-simple-shm ran 3 s; compositor survived"
 else
     echo "  client check: skipped (weston-simple-shm not installed)"
 fi