diff --git a/gc_bba_test/README.md b/gc_bba_test/README.md
index e6b9717..97c51a3 100644
--- a/gc_bba_test/README.md
+++ b/gc_bba_test/README.md
@@ -163,7 +163,24 @@ one block per round trip, so block size sets the speed.
 
 A missing ACK is not fatal: the host retransmits, and the GC re-ACKs a
 duplicate block rather than ignoring it, so a lost ACK recovers in one
-timeout. `txto=N` in the status line counts ACKs where TI never set.
+timeout.
+
+**The GC stops with its summary before booting Swiss** — press a button to
+continue. Those counters are the only place the *reason* for a retry shows up:
+
+```
+desc@+0=N +1=M  bad=N  lapped=N  txto=N
+```
+
+- `txto` — ACKs where TI never set, i.e. the frame never reached the W5100.
+  High `txto` with many host-side retries means the TX path is the problem.
+- `txto` near zero with many retries means ACKs were sent but lost on the wire
+  or at the host.
+- `desc@+1` — frames whose ring descriptor landed at page offset +1 instead of
+  +0. Still unexplained; the reader handles both.
+
+Measured 2026-10-09: 299908 bytes in 293 blocks took 15.7 s with roughly a
+third of blocks retried — the retries, not the data, dominate the time.
 
 ### Stop-and-wait is load-bearing
 
diff --git a/gc_bba_test/source/main.c b/gc_bba_test/source/main.c
index 06ae956..89720ae 100644
--- a/gc_bba_test/source/main.c
+++ b/gc_bba_test/source/main.c
@@ -152,7 +152,8 @@ static void do_netboot(char *status, size_t nstatus)
 	 * round trips. */
 	printf(C_DIM "  temp+verify+rotate; keeps <name>.prev.dol; boots /boot.dol\n"
 	       C_RESET);
-	printf("  Blocks until done or it times out (30 s idle, 5 s mid-xfer).\n\n");
+	printf("  Blocks until done or it times out (30 s idle, 5 s mid-xfer).\n"
+	       "  Stops with the summary before booting Swiss.\n\n");
 
 	if (!s_stage) {
 		snprintf(status, nstatus, "net-update: staging region unavailable");
@@ -176,6 +177,12 @@ static void do_netboot(char *status, size_t nstatus)
 		return;
 	}
 
+	/* Stop here so the transfer summary can actually be READ.  Booting
+	 * straight into Swiss scrolled it away, which meant the one set of
+	 * numbers that says WHY blocks needed retrying -- txto, the descriptor
+	 * split, bad/lapped counts -- was never visible. */
+	nb_pause("update written to the SD card");
+
 	printf("\n  booting /boot.dol ...\n");
 	VIDEO_WaitVSync();
 	VIDEO_WaitVSync();
diff --git a/gc_bba_test/source/netboot.c b/gc_bba_test/source/netboot.c
index 0e00c12..0c4bd92 100644
--- a/gc_bba_test/source/netboot.c
+++ b/gc_bba_test/source/netboot.c
@@ -413,14 +413,23 @@ int netboot_receive(void *dst, u32 maxlen, void (*status)(const char *msg))
 				memcpy(s_dest, frame + NB_HDR, plen);
 				s_dest[plen] = 0;
 			}
-			if (status) { snprintf(line, sizeof(line),
-			                       "received %u bytes, %u blocks "
-			                       "(desc@+0=%u +1=%u, %u bad)",
-			                       (unsigned)got, (unsigned)nblocks,
-			                       (unsigned)s_shift_hits[0],
-			                       (unsigned)s_shift_hits[1],
-			                       (unsigned)bad);
-			              status(line); }
+			if (status) {
+				snprintf(line, sizeof(line),
+				         "received %u bytes in %u blocks",
+				         (unsigned)got, (unsigned)nblocks);
+				status(line);
+				/* txto distinguishes the two reasons an ACK can go
+				 * missing: TI never set (the frame did not reach the
+				 * W5100) versus sent-but-lost.  The host reports the
+				 * retries; only the GC knows which kind they were. */
+				snprintf(line, sizeof(line),
+				         "desc@+0=%u +1=%u  bad=%u  lapped=%u  txto=%u",
+				         (unsigned)s_shift_hits[0],
+				         (unsigned)s_shift_hits[1],
+				         (unsigned)bad, (unsigned)lapped,
+				         (unsigned)s_tx_timeouts);
+				status(line);
+			}
 			return (int)got;
 		}
 		if (frame[18] != NB_OP_DATA) continue;