diff --git a/docs/re/captures/menu-nav/live-jp-difficulty.png b/docs/re/captures/menu-nav/live-jp-difficulty.png
new file mode 100644
index 00000000..b7d21aed
Binary files /dev/null and b/docs/re/captures/menu-nav/live-jp-difficulty.png differ
diff --git a/docs/re/data/difficulty-is-a-language-pair.txt b/docs/re/data/difficulty-is-a-language-pair.txt
new file mode 100644
index 00000000..31af7b62
--- /dev/null
+++ b/docs/re/data/difficulty-is-a-language-pair.txt
@@ -0,0 +1,41 @@
+# Are GP_DIALOG entries 2/3 an ENGLISH/JAPANESE pair? ✅ YES -- MEASURED 2026-08-31.
+#
+# CLOSES the item left open when I withdrew "an EN/JP pair" as a bare assertion.
+# The byte comparison had established only "two builds, same element names, ~97.2 %
+# identical bytes" (dialog-0-1-is-a-duplicate.txt); that they are ENGLISH and
+# JAPANESE rested on the disc's convention, not on this screen. The untested step
+# was a `ja` capture, and it is now taken.
+#
+# ✅ THE ORACLE, both locales, same navigation
+# (captures/menu-nav/live-jp-difficulty.png against the English capture):
+#
+#   EN vs JP DIFFICULTY: 1.82 % of pixels differ, in FOUR bands and nowhere else:
+#     y 133..182  x 414..868   the HEADING      "DIFFICULTY"  ->  難易度選択
+#     y 376..410  x 527..566   the ring, 2 px   (EN row 395.0, JP 393.0)
+#     y 516..551  x 585..679   the BACK label   "BACK"        ->  戻る
+#     y 640..677  x 455..833   the FOOTER       Select/OK/Back -> 選択/決定/戻る
+#
+# 📌 EASY / NORMAL / HARD DO NOT APPEAR IN THE DIFFERING SET. They are the same
+# Latin text in both builds -- the Japanese release leaves the three difficulty
+# names untranslated and changes only the heading, the BACK item and the footer.
+#
+# ✅ SO 2/3 ARE A LANGUAGE PAIR, and the disc agrees quantitatively: 2.77 % of
+# BYTES differ between the two entries, against 1.82 % of PIXELS on screen. A few
+# sprites change and the layout, the element set and three of the four labels do
+# not.
+#
+# ✅ AND THE JP SCREEN OPENS ON NORMAL, like the English one -- so the opening
+# item is not locale-dependent either. The sweep also reproduced the reset finding
+# in Japanese: in-cursor 1.3 from opened against 93.9 from where left, where
+# English gave 1.0 against 93.9.
+#
+# ⚠️ REACH: one JP boot, one screen. It does not establish that every EN/JP pair
+# on the disc differs only in text -- GP_TITLE 4/7, the title art, is already
+# known NOT to (entry 7 carries nine sprites entry 4 lacks).
+#
+# 📌 The first attempt at this run FAILED and the fix is why it worked: the reach
+# probe was doing a full menu->title->menu round trip and not coming back, leaving
+# the game off-menu. --reach-only stops once the menu is reached. Arriving is the
+# cheap part; the round trip was that probe's own experiment, not this caller's.
+#
+# Locale restored and verified at language = 1.