re: a menu/EXTRAS discriminator that failed its control, and the fix

sylpheed-port's BLOCKED.md ask #1 -- does (B) from EXTRAS also show no black
interval, or is "(B) has no black" one screen pair -- needs the harness to know
it is on EXTRAS. screen_id.py cannot tell: both are dark blue GP_TITLE screens
and it reports `menu` for either.

which_title_screen.py correlates a grab against our build 5 / build 6 renders.
First version FAILED its control: it called live-main-menu.png "extras" and
live-extras.png "main_menu", both backwards, margins under 1.1 on RMSE ~33.
Cause: it applied the y=45 game-surface offset unconditionally, but only a full
1280x720 display frame has the menu bar -- a 1279x675 grab IS the surface, and
two of the three reference captures are surface-sized. Offset made conditional;
the control now passes 4/4 with margins 9.9-11.7 against ~18 within-class.

The tool is a navigation aid for driving the emulator and says so: it identifies
a screen by agreeing with our own renders, so nothing measured may rest on it.

Also records the METHOD entry sylpheed-port offered from their own wedged check
script: an absence of output is not a status. They reported "still running, two
lines, both ok" for three iterations while the first attempt had died silently
under its own timeout with block-buffered output -- zero information, reported
as patience. An orphan from an older form of the same script was found running
after 9.5 hours. Line-buffer and bound long jobs, and check the artifact rather
than the exit code: the artifact reached its correct duration while the process
never returned.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wuu56cE8vJGTBtn1ppsk8v

docs/re/METHOD.md

@@ -203,6 +203,19 @@ agent's loop prompt, i.e. nowhere durable. See [`README.md`](README.md) for the
   or `-` is that bug's signature. Grep the corpus for readers of a structure
   before calling its fix done.
 
+* **An absence of output is not a status.** `sylpheed-port` reported "still
+  running, two lines, both ok" for three consecutive iterations of a check script.
+  The first attempt had already died silently under its own timeout with
+  block-buffered output — so there had been *zero* information from it, and "still
+  two lines" was being read as patience rather than as the alarm it was. The
+  underlying process was wedged on an ffmpeg that finishes its work and never
+  exits (an orphan from an older form of the same script was found still running
+  after **9.5 hours**). Two rules fall out, and this corpus has now paid for both:
+  a long-running job needs **line-buffered output and a bound**, and *"no new
+  output"* must be treated as **no information**, never as progress. Check the
+  artifact, not the exit code — the artifact reached its correct 8.0 s duration
+  while the process it came from never returned.
+
 ## Runtime / emulator
 
 * **Look at the PNG** — and check its dimensions.

tools/re-capture/fade_decompose.sh

@@ -23,6 +23,9 @@
 #
 # WHERE=menu (default) arms on the main menu and presses (B)  -> menu -> title.
 # WHERE=title            arms on the boot title and presses (A) -> title -> menu.
+# WHERE=extras           navigates to EXTRAS, arms there, presses (B) -> menu.
+#                        Tests whether "(B) has no black interval" is a rule or
+#                        one screen pair -- sylpheed-port's BLOCKED.md ask #1.
 # The two transitions have DIFFERENT declared fade-ins for the incoming screen --
 # build 4 is 0->16 (16 units, 8 frames), build 5 is 0->12 (12 units, 6 frames) --
 # which is what makes the pair a discriminator rather than a fit.
@@ -68,6 +71,31 @@ if [ "$WHERE" = "menu" ]; then
   done
   [ "$s" = "menu" ] || { echo "NO MENU (screen=$s)"; exit 2; }
   BTN=B
+elif [ "$WHERE" = "extras" ]; then
+  python3 "$SD/pad.py" tap A 0.5
+  for _ in 1 2 3 4 5 6; do
+    sleep 4; s="$(screen)"; echo "  after A: $s"
+    [ "$s" = "menu" ] && break
+  done
+  [ "$s" = "menu" ] || { echo "NO MENU (screen=$s)"; exit 2; }
+  # EXTRAS is the only main-menu button that stays inside GP_TITLE, and initial
+  # focus varies boot to boot, so search: press (A), ask which screen we got, and
+  # if it is not EXTRAS come back with (B) and step the cursor. screen_id.py
+  # cannot tell EXTRAS from the main menu -- both are dark blue GP_TITLE screens
+  # -- so the check is which_title_screen.py, whose control separates them by
+  # ~11 RMSE against ~18 within-class.
+  found=""
+  for try in 1 2 3 4 5 6; do
+    python3 "$SD/pad.py" tap A 0.5; sleep 5
+    shot "$OUT/try$try.png"
+    w="$(python3 "$SD/which_title_screen.py" "$OUT/try$try.png")"
+    echo "  try $try: $w"
+    case "$w" in extras*) found=1; break;; esac
+    python3 "$SD/pad.py" tap B 0.5; sleep 5
+    python3 "$SD/pad.py" dpad UP 0.3; sleep 2
+  done
+  [ -n "$found" ] || { echo "NEVER REACHED EXTRAS"; exit 3; }
+  BTN=B
 else
   # arm on the title itself and press (A): the incoming screen is then build 5,
   # whose declared fade-in is 12 units where build 4's is 16.

tools/re-capture/which_title_screen.py

@@ -0,0 +1,44 @@
+#!/usr/bin/env python3
+"""Is this screenshot the MAIN MENU or EXTRAS? screen_id.py cannot tell them apart.
+
+Both are dark-blue `GP_TITLE` screens, so `screen_id.py` reports `menu` for
+either. This correlates the grab against our own build 5 / build 6 renders and
+reports which is closer, plus the margin -- a small margin means "cannot tell",
+not "the closer one".
+
+⚠️ These are OUR renders, so this identifies a screen by agreeing with our
+decoding. It is a navigation aid for driving the emulator, NOT evidence about the
+game. Nothing measured is allowed to rest on it.
+
+    which_title_screen.py <grab.png> [ref_dir]
+"""
+import sys
+import numpy as np
+from PIL import Image
+
+def load(p, shape=None):
+    im = Image.open(p).convert("RGB")
+    a = np.asarray(im, dtype=float)
+    return a
+
+def main():
+    grab = load(sys.argv[1])
+    d = sys.argv[2] if len(sys.argv) > 2 else "/sylph-home/re/ref"
+    # The game surface sits at y=45 ONLY in a full 1280x720 display frame; a
+    # 1279x675 grab is the surface already. Hard-coding the 45 made this tool fail
+    # its own control -- it called the main menu "extras" and extras "main_menu",
+    # because two of the three reference captures are surface-sized.
+    off = 45 if grab.shape[0] >= 716 else 0
+    best = []
+    for b, name in ((5, "main_menu"), (6, "extras")):
+        r = load(f"{d}/build{b}.png")
+        h = min(grab.shape[0] - off, r.shape[0]); w = min(grab.shape[1], r.shape[1])
+        g = grab[off:off + h, :w]; rr = r[:h, :w]
+        best.append((float(np.sqrt(((g - rr) ** 2).mean())), name))
+    best.sort()
+    margin = best[1][0] - best[0][0]
+    print(f"{best[0][1]}  rmse={best[0][0]:.2f}  (other {best[1][1]} {best[1][0]:.2f}, margin {margin:.2f})")
+    return 0
+
+if __name__ == "__main__":
+    raise SystemExit(main())