diff --git a/issues/two-amdvi-boots-in-a-kvm-suite-wrote-not-one-byte.md b/issues/two-amdvi-boots-in-a-kvm-suite-wrote-not-one-byte.md
new file mode 100644
index 0000000000..7a7acaf43f
--- /dev/null
+++ b/issues/two-amdvi-boots-in-a-kvm-suite-wrote-not-one-byte.md
@@ -0,0 +1,30 @@
+---
+status: open
+kind: defect
+opened: 2026-10-28
+---
+
+# Two amdvi boots in a KVM suite wrote not one byte
+
+A full `cargo test --test toyos-build` at `0eb2c09d0` on a Linux x86_64 KVM
+host went red on `amdvi_selftest` and `amdvi_firmware_left`, and on nothing
+else. Both boots used `tests/images/testcases.toml`, finished building in the
+same second and timed out together after 63 s. Both reported `the console
+carried: nothing at all` and `uart: nothing at all`. So the 16550 file was
+empty, without even the firmware's screen-clearing escapes that the netcase
+boots in
+`issues/a-netcase-boot-under-host-load-said-nothing-past-the-firmwares-screen-clears.md`
+wrote. Both tests passed in that host's suites at `e4bdb0a5d` and `7745af98d`.
+`7745af98d` differs from `0eb2c09d0` only in a host-side wait inside
+`the_log_lands_on_the_boot_stick`. That suite ran alone on the host, and its
+load during these boots was not recorded. Log:
+https://github.com/ToyOSOrg/evidence/tree/d06821d0cb6e754f1fec16a5921bc4378d9ba600/pr-1107/r2
+
+Not reduced: whether QEMU or the firmware ever ran in those boots. A QMP read
+of a silent guest's registers would say. No `red-run-serial` file was empty,
+so the empty 16550 files were not kept.
+
+**Exit**: a silent boot's registers read, naming where it stopped, and the fix
+at that cause. If it is the netcase boots' cause, fold this record into that one.
+
+**Owner**: the orchestrator.
diff --git a/tests/toyos.rs b/tests/toyos.rs
index 9ff6c2ccbe..117b7aac09 100644
--- a/tests/toyos.rs
+++ b/tests/toyos.rs
@@ -2954,7 +2954,12 @@ fn the_log_lands_on_the_boot_stick(test_config: &Path) -> Result<(), String> {
     // power button is pressed, as the metal row's deadline finds it.
     writeln!(qemu.stdin_mut(), "run {}", metal::STICK_LOAD_JOB).map_err(|e| e.to_string())?;
     qemu.flush_stdin();
-    qemu::await_marker(&mut qemu, &mut console, "stick_load: line 1500 ", "the load to write the stick")?;
+    // Any line past the first 1500, not one exact line: logkeeper sheds its
+    // oldest console lines when the console falls behind the load.
+    let from = console.len();
+    qemu::await_guest(&mut qemu, &mut console, "the load to write the stick", |log| {
+        log[from..].lines().any(|l| stick_load_line(l).is_some_and(|n| n >= 1500))
+    })?;
     qemu::QmpShutdown::open(qemu.qmp_socket(), qemu.budget(qemu::GUEST_QUIET)).power_button();
     let by = qemu.budget(Duration::from_secs(60));
     console.push_str(&qemu.await_exit(by)?);
@@ -2981,6 +2986,12 @@ fn the_log_lands_on_the_boot_stick(test_config: &Path) -> Result<(), String> {
     usb_stopped_before_the_reset(&log, &console)
 }
 
+fn stick_load_line(line: &str) -> Option<u64> {
+    let (_, rest) = line.split_once("stick_load: line ")?;
+    let digits = rest.find(|c: char| !c.is_ascii_digit()).unwrap_or(rest.len());
+    rest[..digits].parse().ok()
+}
+
 /// usbd ended for the stop with every connected port's reset finished and its
 /// controller halted and reset (its exit code, which the kernel records,
 /// because its own line may reach no log), and it ended before the kernel's