diff --git a/docs/port/MISSION.md b/docs/port/MISSION.md
index 495d25c..3973765 100644
--- a/docs/port/MISSION.md
+++ b/docs/port/MISSION.md
@@ -137,12 +137,26 @@ the frame at 734 s has content but correlates only **0.15** with build 4's rende
 and 0.05 with the main menu, i.e. it is an attract-movie frame, not title art.
 
 So the Japanese run was **not** failing because of the locale — neither locale
-reaches the interactive title in ~12 minutes of no pad input. 🟡 What remains
-untested is a **pad press**: `pad.py` and `nav_to_flight.sh` exist, and
-`title_states_capture.sh` reaches title states with its own flag set
-(`--log_ui_draws --ui_draw_capture_frames`, plus xdotool window focus) which this
-probe did not replicate. That is the next thing to try, and it is a capture
-problem, not a locale one.
+reaches the interactive title in ~12 minutes of no pad input. **🔴 Tried the proven path too, and I am stopping this line (2026-08-29).**
+Run 3 launched exactly the way `boot_menu.sh` does — `DISPLAY=:98`, `--apu=sdl`,
+`/dev/shm/xenia_*` cleared, the existing profile signed in — and drove
+`skip_intro.sh`, the detector that is documented to work. It classified **every
+one of 20 samples over 604 s as "movie"** and waited them all out, then timed
+out. A direct check at 604 s confirms it was right: zero green-glyph pixels,
+`screen_id` = other, warm mean (69,53,40), correlation **0.09** with build 7. The
+game really was playing attract movies for ten minutes.
+
+So: **three runs, two locales, two launch paths, ~35 minutes of emulator time, no
+interactive title.** Either the attract loop is far longer than the 600–780 s
+windows tried, or something has regressed since `live-title-press-a.png` was
+captured (that one came from a pad-driven boot — commit `88b3ce9`, "booting to
+the main menu and walking it").
+
+⚠️ **Not worth more loop iterations.** This capture does not block any of the
+five menu screens, and it has now cost five. It is written down so a later
+session with a reason to spend an hour on the boot path can pick it up; the
+locale mechanism, the fixed oracle and the launch recipe are all in place, so
+what is left is patience with the attract loop, not tooling.
 
 The locale is restored to English; `set_console_language.py ja` flips it back in
 one command.
diff --git a/docs/re/METHOD.md b/docs/re/METHOD.md
index efd6dd8..d8d58b2 100644
--- a/docs/re/METHOD.md
+++ b/docs/re/METHOD.md
@@ -463,3 +463,10 @@ agent's loop prompt, i.e. nowhere durable. See [`README.md`](README.md) for the
   ran. Printing the oracle's value each tick turns "no title" into "75 samples,
   max glyph 0", which is a measurement — and it is what showed the black-screen
   and attract-movie phases were being sampled correctly all along.
+* **Know when to stop paying for a non-blocking answer.** A capture that would
+  settle two 🟡 residuals — neither of which blocks the deliverable — absorbed
+  five iterations. Each one produced a real finding (a broken oracle, a wrong
+  "impossible", a locale red herring), so none was wasted, but the *question*
+  never moved. The signal to stop is not "this is hard", it is "the thing this
+  unblocks is not blocking anything". Write down the recipe you built and leave
+  it for a session with a reason to spend the time.