diff --git a/goTool/ota_upload.go b/goTool/ota_upload.go
index b1f3ba1..a130990 100644
--- a/goTool/ota_upload.go
+++ b/goTool/ota_upload.go
@@ -20,6 +20,9 @@ const (
 	otaDistQueryInterval   = 500 * time.Millisecond
 	otaDistQueryTimeout    = 2 * time.Second
 	otaDistEmitMinInterval = 150 * time.Millisecond
+	// Pace host chunks so the master UART RX ring is not overrun (~20 frames/block).
+	otaHostChunkPace = 3 * time.Millisecond
+	otaBlockMaxRetries = 3
 )
 
 const (
@@ -189,52 +192,83 @@ func runOTAOnPortUnlocked(sp *serialPort, firmware []byte, onProgress otaProgres
 
 	var seq uint32
 	for offset := 0; offset < imageSize; {
-		bytesInBlock := 0
-		for bytesInBlock < otaFlashBlockSize && offset < imageSize {
-			n := otaHostChunkSize
-			room := otaFlashBlockSize - bytesInBlock
-			if n > room {
-				n = room
-			}
-			if offset+n > imageSize {
-				n = imageSize - offset
-			}
-			chunk := firmware[offset : offset+n]
+		blockStart := offset
+		blockStartSeq := seq
 
-			if err := writeUartMessage(sp, &pb.UartMessage{
-				Type: pb.MessageType_OTA_PAYLOAD,
-				Payload: &pb.UartMessage_OtaPayload{
-					OtaPayload: &pb.OtaPayload{Seq: seq, Data: chunk},
-				},
-			}); err != nil {
-				notify("error", "", 0, err.Error())
-				return err
-			}
-			seq++
-			offset += n
-			bytesInBlock += n
+		sendBlock := func() (fullBlock bool, err error) {
+			bytesInBlock := 0
+			for bytesInBlock < otaFlashBlockSize && offset < imageSize {
+				n := otaHostChunkSize
+				room := otaFlashBlockSize - bytesInBlock
+				if n > room {
+					n = room
+				}
+				if offset+n > imageSize {
+					n = imageSize - offset
+				}
+				chunk := firmware[offset : offset+n]
 
-			pct := offset * 100 / imageSize
-			if pct > 99 {
-				pct = 99
+				if err := writeUartMessage(sp, &pb.UartMessage{
+					Type: pb.MessageType_OTA_PAYLOAD,
+					Payload: &pb.UartMessage_OtaPayload{
+						OtaPayload: &pb.OtaPayload{Seq: seq, Data: chunk},
+					},
+				}); err != nil {
+					return false, err
+				}
+				time.Sleep(otaHostChunkPace)
+				seq++
+				offset += n
+				bytesInBlock += n
+
+				pct := offset * 100 / imageSize
+				if pct > 99 {
+					pct = 99
+				}
+				notify("uploading", otaStepMaster, pct,
+					fmt.Sprintf("Master: %d / %d bytes", offset, imageSize))
 			}
-			notify("uploading", otaStepMaster, pct, fmt.Sprintf("Master: %d / %d bytes", offset, imageSize))
+			return bytesInBlock == otaFlashBlockSize, nil
 		}
 
-		if bytesInBlock == otaFlashBlockSize {
-			st, err := waitOtaStatus(sp, otaStBlockAck, otaDefaultTimeout, nil)
-			if err != nil {
-				notify("error", "", 0, err.Error())
-				return err
-			}
-			pct := offset * 100 / imageSize
-			if pct > 99 {
-				pct = 99
-			}
-			notify("uploading", otaStepMaster, pct,
-				fmt.Sprintf("Master: Block geschrieben (%d bytes)", st.GetBytesWritten()),
-				OTAProgress{Bytes: st.GetBytesWritten()})
+		fullBlock, err := sendBlock()
+		if err != nil {
+			notify("error", "", 0, err.Error())
+			return err
 		}
+
+		if !fullBlock {
+			continue
+		}
+
+		var st *pb.OtaStatusPayload
+		var ackErr error
+		for attempt := 0; attempt < otaBlockMaxRetries; attempt++ {
+			if attempt > 0 {
+				offset = blockStart
+				seq = blockStartSeq
+				if _, err := sendBlock(); err != nil {
+					notify("error", "", 0, err.Error())
+					return err
+				}
+			}
+			st, ackErr = waitOtaStatus(sp, otaStBlockAck, otaDefaultTimeout, nil)
+			if ackErr == nil {
+				break
+			}
+		}
+		if ackErr != nil {
+			notify("error", "", 0, ackErr.Error())
+			return ackErr
+		}
+
+		pct := offset * 100 / imageSize
+		if pct > 99 {
+			pct = 99
+		}
+		notify("uploading", otaStepMaster, pct,
+			fmt.Sprintf("Master: Block geschrieben (%d bytes)", st.GetBytesWritten()),
+			OTAProgress{Bytes: st.GetBytesWritten()})
 	}
 
 	masterPct = 100
diff --git a/main/cmd/cmd_ota.c b/main/cmd/cmd_ota.c
index eae4a3d..5146081 100644
--- a/main/cmd/cmd_ota.c
+++ b/main/cmd/cmd_ota.c
@@ -175,8 +175,18 @@ static void handle_ota_payload(const uint8_t *data, size_t len) {
     return;
   }
 
-  ota_feed_result_t r =
-      ota_uart_feed(req_ptr->data.bytes, req_ptr->data.size);
+  ota_feed_result_t r = ota_uart_feed_chunk(req_ptr->seq, req_ptr->data.bytes,
+                                            req_ptr->data.size);
+  if (r == OTA_FEED_SEQ_GAP) {
+    send_ota_failed(16);
+    return;
+  }
+  if (r == OTA_FEED_SEQ_DUP) {
+    if (ota_uart_block_ready_for_reack()) {
+      send_ota_status(OTA_UART_ST_BLOCK_ACK, 0);
+    }
+    return;
+  }
   if (r == OTA_FEED_ERROR) {
     send_ota_failed( 13);
     return;
@@ -189,15 +199,6 @@ static void handle_ota_payload(const uint8_t *data, size_t len) {
     led_ring_show_ota_progress(done, total, OTA_LED_UART_R, OTA_LED_UART_G,
                                OTA_LED_UART_B);
     send_ota_status(OTA_UART_ST_BLOCK_ACK, 0);
-    return;
-  }
-
-  if (r == OTA_FEED_OK) {
-    uint32_t total = ota_uart_total_size();
-    if (total > 0) {
-      led_ring_show_ota_progress(ota_uart_bytes_received(), total,
-                                 OTA_LED_UART_R, OTA_LED_UART_G, OTA_LED_UART_B);
-    }
   }
 }
 
diff --git a/main/esp_now_core.h b/main/esp_now_core.h
index f459ecc..f5a00c6 100644
--- a/main/esp_now_core.h
+++ b/main/esp_now_core.h
@@ -23,8 +23,14 @@ esp_err_t esp_now_core_ensure_broadcast_peer(void);
 
 esp_err_t esp_now_core_send(const uint8_t *dest_mac,
                             const alox_EspNowMessage *msg);
+/** Like send but does not wait for esp_now send-done (lower latency). */
+esp_err_t esp_now_core_send_fast(const uint8_t *dest_mac,
+                                 const alox_EspNowMessage *msg);
 esp_err_t esp_now_core_send_wait(const uint8_t *dest_mac,
                                  const alox_EspNowMessage *msg);
+/** OTA payloads: wait for send-done with extra NO_MEM backoff (queue drain). */
+esp_err_t esp_now_core_send_reliable(const uint8_t *dest_mac,
+                                     const alox_EspNowMessage *msg);
 
 esp_err_t esp_now_core_init_radio(uint8_t channel);
 void esp_now_core_init_send_done(void);