diff --git a/docs/agents/CONTAINER-NOTES.md b/docs/agents/CONTAINER-NOTES.md
index a5a36c34..647467f4 100644
--- a/docs/agents/CONTAINER-NOTES.md
+++ b/docs/agents/CONTAINER-NOTES.md
@@ -96,6 +96,13 @@ python3 tools/re-capture/gmem.py find hex:820af844 400
   silence and 0.01 gaps/s, against 9.98 % / 8.37 rendered. ⚠️ No video, so
   screen-based provenance is unavailable (use the XMA probe), and `--gpu=null`
   runs here die at ~70 s with `PM4_DRAW_INDX: Failed in backend`.
+  🔴 **REFUTED 2026-08-30 — that lifetime does not hold.** A `--gpu=null` capture
+  ran **148.02 s** and ended on its own probe's timer with the emulator still
+  alive, having decoded the whole `ADV` movie
+  ([`intro-audio-output-census.md`](../re/structures/intro-audio-output-census.md)).
+  More than twice the quoted figure. Whatever produced the ~70 s was fixed or was
+  never general, and this note had been the reason not to use `--gpu=null` for
+  anything long — which is exactly the configuration a clean audio capture needs.
 
   ⚠️ **Do not read Canary's 6-channel PulseAudio stream as evidence the GAME is
   5.1.** `pactl` will show `float32le 6ch 48000Hz`, channel-mapped to a full 5.1
diff --git a/docs/re/structures/menu-bgm-loop-not-yet-captured.md b/docs/re/structures/menu-bgm-loop-not-yet-captured.md
new file mode 100644
index 00000000..8b4330a9
--- /dev/null
+++ b/docs/re/structures/menu-bgm-loop-not-yet-captured.md
@@ -0,0 +1,68 @@
+# 🟡 The menu BGM loop point — not captured, and why the obvious rig cannot take it
+
+**Classification: not an answer.** Recorded per *"do not improvise around a
+blocker"*: this says what was tried, what it cost, and the route that is left.
+
+## The question
+
+`BGM_103` is the menu track: two waves of **87.750 s** that play *together*
+([`bgm-two-stems.md`](bgm-two-stems.md)). That page establishes the track **is not
+a seamless loop** — `BGM_001` fades out with 6.15 s of silence after it — and that
+**no loop-point field has been found in the XMA header**, so a menu loop is
+currently **authored**. The port measures its bed restarting at 87.8 s with a
+**3.4 s near-silent seam**.
+
+What is missing is what the *game* does at that seam. Sitting on the menu for two
+loop periods and recording would answer it.
+
+## 🔴 The rig that seemed obvious does not work
+
+Audio needs the ALSA `file` tee; detecting the title needs **video**, so
+`--gpu=null` is unavailable. That combination was measured, twice:
+
+| | |
+|---|---|
+| guest speed | **~0.20× real time** (76.5 s of audio in 378 s of wall clock) |
+| title reached | **no** — not in 300 s, even after tapping Ⓐ to skip the movie |
+| ALSA state at the end | `avail_update returned error: -32 (Broken pipe)`, then `ALSA underrun detected, recovering...` |
+
+⚠️ **The emulator was not crashing** — memory was fine (rss 701 MB, 9.5 GB free) and
+the `Killed` line in the log is this harness's own cleanup after the probe timed
+out. The failure is simply that **rendering plus a PulseAudio-paced ALSA tee plus an
+x11grab sampler is too slow to reach the menu**, and the tee's slave breaks under it.
+
+## ✅ Refuted along the way — "`--gpu=null` runs here die at ~70 s"
+
+[`CONTAINER-NOTES.md:98`](../agents/CONTAINER-NOTES.md) says `--gpu=null` runs
+*"die at ~70 s with `PM4_DRAW_INDX: Failed in backend`"*. **That is not true of this
+container now.** The intro-audio capture
+([`intro-audio-output-census.md`](intro-audio-output-census.md)) ran `--gpu=null`
+for **148.02 s** and ended because its probe's timer expired, with the emulator
+still alive and the whole `ADV` movie decoded. More than twice the quoted lifetime,
+so the claim does not hold as written; whatever produced it was either fixed or was
+never general.
+
+That matters because `--gpu=null` is the only configuration that gives a clean
+capture (0.31 % silence, ~0.96× real time), and this note had been the reason not to
+use it for anything long.
+
+## The route that is left, for whoever picks this up
+
+**Use the XMA probe log as the screen oracle, and drop video entirely.**
+`bgm-two-stems.md` records that sitting on the main menu decodes exactly two
+streams — **3 876 864** and **3 930 112 B**, `BGM_103`'s two declared waves. So:
+
+1. boot `--gpu=null --apu=alsa --mute=false --xma_param_probe=true`, profile signed
+   in (`title-a-press-fault.md`);
+2. tap Ⓐ on a schedule to skip the movie and enter the menu — blind is acceptable
+   because the guest runs near real time here;
+3. **confirm arrival from the log, not the screen**: `BGM_103`'s two `byte_size`s
+   appearing is the menu.
+
+That replaces the video oracle with a log one, which is better provenance for an
+audio question anyway — it evidences what is being *recorded* rather than what was
+on screen. Not attempted; the iteration ran out.
+
+⚠️ **Do not read this page as "the menu is unreachable".** Ⓐ into the menu is
+measured and works ([`title-a-press-fault.md`](title-a-press-fault.md), leg B,
+final glyph 327). What failed is one specific *recording* configuration.