diff --git a/configs/test_cam1_face_det_rtsp_server.json b/configs/test_cam1_face_det_rtsp_server.json
index 2c9716f..d6e0d32 100644
--- a/configs/test_cam1_face_det_rtsp_server.json
+++ b/configs/test_cam1_face_det_rtsp_server.json
@@ -36,11 +36,12 @@
           "role": "filter",
           "enable": true,
           "model_path": "./models/RetinaFace_mobile320.rknn",
-          "conf": 0.6,
+          "conf": 0.3,
           "nms": 0.4,
           "max_faces": 10,
           "output_landmarks": true,
-          "input_format": "rgb"
+          "input_format": "rgb",
+          "debug": { "stats": true, "stats_interval": 30, "log_outputs": true }
         },
         {
           "id": "osd_cam1",
@@ -51,7 +52,8 @@
           "draw_text": true,
           "line_width": 2,
           "font_scale": 1,
-          "labels": []
+          "labels": [],
+          "debug": { "draw_test_rect": true, "force_clone_on_dma": true }
         },
         {
           "id": "post_cam1",
diff --git a/plugins/ai_face_det/ai_face_det_node.cpp b/plugins/ai_face_det/ai_face_det_node.cpp
index a67711a..684ce12 100644
--- a/plugins/ai_face_det/ai_face_det_node.cpp
+++ b/plugins/ai_face_det/ai_face_det_node.cpp
@@ -674,6 +674,9 @@ private:
             std::cerr << "[ai_face_det] frame=" << frame->frame_id
                       << " faces=" << n
                       << " best=" << best
+                      << " max=" << last_max_score_
+                      << " cand=" << last_candidates_
+                      << " prior_mismatch=" << (last_prior_mismatch_ ? 1 : 0)
                       << " conf_thr=" << conf_thresh_
                       << " nms=" << nms_thresh_
                       << "\n";
@@ -685,6 +688,10 @@ private:
                           int orig_w, int orig_h,
                           int in_w, int in_h,
                           FaceDetResult& out) {
+        last_prior_mismatch_ = false;
+        last_candidates_ = 0;
+        last_max_score_ = 0.0f;
+
         // Find loc/conf/landms tensors.
         std::vector<NcTensor> locs;
         std::vector<NcTensor> confs;   // Nx2
@@ -753,6 +760,7 @@ private:
         if (!priors.empty() && static_cast<int>(priors.size()) != n) {
             // Mismatch: can't reliably decode.
             std::cerr << "[ai_face_det] prior mismatch: priors=" << priors.size() << " n=" << n << "\n";
+            last_prior_mismatch_ = true;
             return;
         }
 
@@ -784,7 +792,9 @@ private:
                 if (s >= 0.0f && s <= 1.0f) score = s;
                 else score = Sigmoid(s);
             }
+            if (score > last_max_score_) last_max_score_ = score;
             if (score < conf_thresh_) continue;
+            ++last_candidates_;
 
             const Prior p = priors.empty() ? Prior{0, 0, 0, 0} : priors[static_cast<size_t>(i)];
 
@@ -898,6 +908,11 @@ private:
     bool log_outputs_ = false;
     uint64_t processed_ = 0;
     int printed_outputs_ = 0;
+
+    // Debug stats per last decoded frame
+    bool last_prior_mismatch_ = false;
+    int last_candidates_ = 0;
+    float last_max_score_ = 0.0f;
 };
 
 REGISTER_NODE(AiFaceDetNode, "ai_face_det");
diff --git a/plugins/osd/osd_node.cpp b/plugins/osd/osd_node.cpp
index 2a5d76e..c743d20 100644
--- a/plugins/osd/osd_node.cpp
+++ b/plugins/osd/osd_node.cpp
@@ -295,6 +295,10 @@ public:
             stats_log_ = dbg->ValueOr<bool>("stats", stats_log_);
             stats_interval_ = std::max<uint64_t>(
                 1, static_cast<uint64_t>(dbg->ValueOr<int>("stats_interval", static_cast<int>(stats_interval_))));
+
+            // Debug helpers
+            draw_test_rect_ = dbg->ValueOr<bool>("draw_test_rect", draw_test_rect_);
+            force_clone_on_dma_ = dbg->ValueOr<bool>("force_clone_on_dma", force_clone_on_dma_);
         }
 
         // Load custom labels from config
@@ -333,6 +337,9 @@ public:
 
         FramePtr out = frame;
         if (out->data && (out->det || out->face_det || out->face_recog)) {
+            if (force_clone_on_dma_ && out->dma_fd >= 0) {
+                out = CloneFrameForWrite(out);
+            }
             if (out.use_count() > 1) {
                 out = CloneFrameForWrite(out);
             }
@@ -345,6 +352,26 @@ public:
             DrawDetections(out);
             DrawFaceDet(out);
             DrawFaceRecog(out);
+
+            if (draw_test_rect_) {
+                int w = out->width;
+                int h = out->height;
+                uint8_t* data = out->planes[0].data ? out->planes[0].data : out->data;
+                PixelFormat fmt = out->format;
+                int stride;
+                if (fmt == PixelFormat::RGB || fmt == PixelFormat::BGR) {
+                    stride = out->planes[0].stride > 0 ? out->planes[0].stride : (out->stride > 0 ? out->stride : w * 3);
+                } else if (fmt == PixelFormat::NV12 || fmt == PixelFormat::YUV420) {
+                    stride = out->planes[0].stride > 0 ? out->planes[0].stride : (out->stride > 0 ? out->stride : w);
+                } else {
+                    stride = 0;
+                }
+                if (data && stride > 0) {
+                    const Color c{255, 0, 255};
+                    DrawRect(data, w, h, stride, fmt, 10, 10, 220, 120, std::max(2, line_width_), c);
+                    DrawText(data, w, h, stride, fmt, 14, 14, "OSD", font_scale_, c);
+                }
+            }
             if (out->dma_fd >= 0 && out->data) {
                 DmaSyncEndFd(out->dma_fd);
             }
@@ -538,6 +565,9 @@ private:
 
     bool stats_log_ = false;
     uint64_t stats_interval_ = 100;
+
+    bool draw_test_rect_ = false;
+    bool force_clone_on_dma_ = false;
 };
 
 REGISTER_NODE(OsdNode, "osd");