fabi/Sylpheed
1
0
Fork 0
Code Issues 9 Pull Requests Actions Packages Projects Releases Wiki Activity

re: two sweeps, two instrument faults -- the three submenus are still unmeasured

Browse Source 
Sweep 1: ring_row.py scans x 500:542, the main menu's gutter. EXTRAS happened to
put its ring there; LOAD GAME, TUTORIAL and OPTIONS do not, so the reader found a
static element and all three voided on 'the ring did not move'. Differencing S1
against S2 shows the cursors moved at x 97..231, 338..1099 and 153..479.

Sweep 2: replaced the reader with a whole-frame comparison, controlled on the
EXTRAS frames whose answer is known. Then the guest crashed -- the already
documented STL map/set erase at PC 0x82307128, firing early in the boot, not
something B did -- and Xenia's crash dialog covers the screen centre, so a
whole-frame identity test can never match again. The narrow ring column the dialog
does not cover was reading correctly the whole time. I traded one blindness for
another, and did it because the first instrument had just failed.

Refutation attempt on 'B on a submenu restores the parent's focus' (4/4): it
SURVIVES, and the run I had written off is what corroborates it. The stuck frame
has the ring at y 303.5 = LOAD GAME, the item entered from. A fifth instance.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wuu56cE8vJGTBtn1ppsk8v
This commit is contained in:
sylph-decoder
2026-08-30 22:25:04 +00:00
parent 3f34c8c7f3
commit 83ace47d46
3 changed files with 263 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

70
docs/re/data/submenu-focus-sweep-unmeasured.txt Normal file
View File

@@ -0,0 +1,70 @@
# Do LOAD GAME, TUTORIAL and OPTIONS remember their cursors?
# 2026-08-30. ❔ STILL UNMEASURED. Two sweeps, two different instrument faults.
#
# The main menu PERSISTS; EXTRAS RESETS. Two screens, two behaviours, so every
# screen needs measuring and none can be inferred. NEW GAME stays deliberately
# untested. These three are what is left, and they are still open.
#
################################################################################
# SWEEP 1 -- voided by a reader that only works on one screen.
#
# All three voided on "the ring did not move" after a DELIVERED DOWN:
# LOAD GAME ring 288.0 -> 288.0
# TUTORIAL ring 385.0 -> None
# OPTIONS ring 287.5 -> 287.5
#
# 🔴 THE RING HAD MOVED. Differencing S1 against S2 shows exactly where:
# LOAD GAME 0.25% of frame changed, x 97..231
# TUTORIAL 2.31% x 338..1099
# OPTIONS 2.59% x 153..479
#
# ring_row.py scans x 500:542 -- the MAIN MENU's gutter. EXTRAS happened to put
# its ring in that column, which is why the EXTRAS measurement was sound; these
# three do not. The reader was finding a STATIC element and reporting no motion.
# ⚠️ A reader calibrated on one screen is not a reader for the others, and
# "it worked last time" is what made that invisible.
#
################################################################################
# SWEEP 2 -- the reader replaced, and defeated by an OVERLAY instead.
#
# Decision rule changed to need no ring at all: S1 and S2 differ ONLY by cursor
# position, so compare the re-entry frame to each over the whole frame.
# ✅ CONTROLLED on the EXTRAS frames, whose answer is already known:
# E1 vs E2 (the move) 2.70% -> localised, control passes
# E3 vs E1 (opened) 0.02%
# E3 vs E2 (left on) 2.71% verdict RESETS, and RESETS is the known answer.
#
# LOAD GAME S1 opened (glyph 12838), DOWN changed 6.00% -- control PASSED
# then: TIMEOUT waiting for the main menu. TUTORIAL and OPTIONS
# skipped, because the probe refuses to measure from an unknown
# state rather than pressing on.
#
# 🔴 THE GUEST HAD CRASHED -- and it is the crash this corpus already documents.
# PC 0x82307128, guest thread 9, Access Violation read at 0x000000010000000C,
# preceded by HostPathDevice::ResolvePath(\aab216c3) and "Guest attempted to
# throw a C++ exception!" -- the STL map/set erase of title-crash-stl-tree.md.
# ⚠️ NOT a new finding: that page has it, and autopilot-knowledge-sources.md
# counts 384 dumps at the same PC. The first dump here is at log line 2204, so it
# fired EARLY IN THE BOOT and the dialogs stacked; it is not something Ⓑ did.
#
# ⚠️ AND THE "MORE ROBUST" RULE WAS THE MORE FRAGILE ONE. Xenia's crash dialog
# covers the middle of the screen, so a WHOLE-FRAME identity test can never match
# again once it appears. The narrow ring column the dialog does not cover read the
# screen correctly the whole time. A global comparison is defeated by any overlay;
# a narrow measured feature survives it. I replaced the reader with the global
# rule *because* the reader had just failed, and traded one blindness for another.
#
################################################################################
# ✅ REFUTATION ATTEMPT -- "Ⓑ on a submenu returns to the parent with focus
# restored to the item you entered from" (4/4, another agent's). IT SURVIVES, and
# the run that looked like a failure is what corroborates it.
#
# The stuck frame -- main menu visible behind the crash dialog -- has the ring at
# y 303.5, which the calibrated reader names LOAD GAME. That is exactly the item
# the probe entered from. So Ⓑ did return AND did restore focus; only my screen
# test failed. A FIFTH instance, from a run I had written off.
#
# ❔ WHAT IS STILL OPEN: whether LOAD GAME, TUTORIAL or OPTIONS persist. Nothing
# here measures it. What a third attempt needs is a per-screen ring locator (the
# S1->S2 difference gives the cursor's own region on each screen) plus a screen
# test that ignores the dialog rectangle -- or a boot without the crash.

2
tools/re-capture/extras_focus_session.sh
View File

@@ -22,4 +22,4 @@ nohup run-canary --apu=sdl --log_mask=13 --log_level=2 \
# reach the title and the menu with the probe that already does it # reach the title and the menu with the probe that already does it
python3 "$SD/focus_persistence.py" "$LOG" "$OUT/reach" 900 >"$OUT/reach.log" 2>&1 python3 "$SD/focus_persistence.py" "$LOG" "$OUT/reach" 900 >"$OUT/reach.log" 2>&1
echo "-- reached the menu; now the EXTRAS question --" echo "-- reached the menu; now the EXTRAS question --"
python3 "$SD/extras_focus_persistence.py" "$LOG" "$OUT" 600 python3 "${SWEEP:-$SD/extras_focus_persistence.py}" "$LOG" "$OUT" 600

192
tools/re-capture/submenu_focus_sweep.py Normal file
View File

@@ -0,0 +1,192 @@
#!/usr/bin/env python3
"""Does EACH submenu remember its cursor across leave -> re-enter?
The main menu PERSISTS (focus-persists-across-title.txt); EXTRAS RESETS
(extras-focus-resets.txt). Two screens, two behaviours, so there is no menu-wide
rule and every screen has to be measured. This sweeps the three that are left:
LOAD GAME, TUTORIAL, OPTIONS.
❔ NEW GAME is deliberately NOT entered -- the corpus has held it untested
because it starts a game, and nothing here is worth breaking that for.
Controls, each of which exists because an earlier run failed without it:
* ABSOLUTE row check after EVERY navigation press, not just the total -- a
constant offset passes a differential control exactly
(menu-focus-reader-offset.txt);
* SCREEN IDENTITY against a reference frame captured in this same run -- the
glyph window cannot separate the main menu from a submenu (327 / 324 / 317);
* 🔴 NO RING READER INSIDE A SUBMENU. ring_row.py scans x 500:542, which is
the MAIN MENU's gutter. EXTRAS happened to put its ring in that column;
LOAD GAME, TUTORIAL and OPTIONS do not (their cursors move at x 97..231,
338..1099 and 153..479), so the first sweep read a STATIC element and all
three voided on "the ring did not move". They had moved. The decision here
needs no ring at all: S1 and S2 differ ONLY by cursor position, so compare
S3 to each of them over the whole frame;
* every press confirmed from the guest's own [RE-INPUT] log.
Any control failing SKIPS that screen and moves on; it never reports a number it
cannot justify.
submenu_focus_sweep.py LOG OUTDIR [wait_s]
"""
import os, re, subprocess, sys, time
import numpy as np
from PIL import Image
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
from ring_row import ring_row, main_menu_item, NAMES
LOG, OUT = sys.argv[1], sys.argv[2]
W, H = 1280, 720
PAD = os.path.join(os.path.dirname(os.path.abspath(__file__)), "pad.py")
TARGETS = [1, 2, 3] # LOAD GAME, TUTORIAL, OPTIONS. NEW GAME excluded.
T0 = time.time()
def deliveries(vk):
pat = re.compile((r"RE-INPUT\] XamInputGetKeystrokeEx -> user=\d+ vk=%s flags=0001" % vk).encode())
try:
return len(pat.findall(open(LOG, "rb").read()))
except FileNotFoundError:
return 0
def press(btn, vk, tries=5):
for _ in range(tries):
before = deliveries(vk)
subprocess.run([sys.executable, PAD, "tap", btn, "0.5"], check=False)
for _ in range(20):
time.sleep(0.25)
if deliveries(vk) > before:
return True
print(f"[{time.time()-T0:7.1f}s] 🔴 {btn} NEVER delivered", flush=True)
return False
def _open():
return subprocess.Popen(
["ffmpeg", "-loglevel", "error", "-f", "x11grab", "-draw_mouse", "0",
"-video_size", f"{W}x{H}", "-i", ":98", "-r", "4",
"-f", "rawvideo", "-pix_fmt", "rgb24", "-"],
stdout=subprocess.PIPE, bufsize=W * H * 3 * 2)
def fresh():
q = _open(); a = None
for _ in range(3):
b = q.stdout.read(W * H * 3)
if len(b) == W * H * 3:
a = np.frombuffer(b, np.uint8).reshape(H, W, 3).astype(int)
q.kill()
return a
def img(a):
return Image.fromarray(a.astype(np.uint8))
def differs(a, b):
return float((np.abs(a - b).max(axis=2) > 24).mean())
def glyph(a):
r, g, bl = a[:, :, 0], a[:, :, 1], a[:, :, 2]
return int(((g > 130) & (g - r > 45) & (g - bl > 45)).sum())
def wait_until(pred, what, limit=60):
t = time.time()
while time.time() - t < limit:
a = fresh()
if a is not None and pred(a):
return a
print(f"[{time.time()-T0:7.1f}s] 🔴 TIMEOUT waiting for {what}", flush=True)
return None
os.makedirs(OUT, exist_ok=True)
MAIN = wait_until(lambda a: main_menu_item(ring_row(img(a))) is not None
and 250 <= glyph(a) <= 420, "the main menu", 150)
if MAIN is None:
sys.exit("never identified the main menu")
img(MAIN).save(f"{OUT}/main-ref.png")
print(f"[{time.time()-T0:7.1f}s] MAIN MENU reference, focus = "
f"{NAMES[main_menu_item(ring_row(img(MAIN)))]}", flush=True)
results = {}
for tgt in TARGETS:
name = NAMES[tgt]
print(f"\n=========== {name} ===========", flush=True)
a = wait_until(lambda a: differs(a, MAIN) < 0.15, "the main menu", 60)
if a is None:
print(f" SKIP {name}: not on the main menu"); continue
cur = main_menu_item(ring_row(img(a)))
if cur is None:
print(f" SKIP {name}: no main-menu ring row"); continue
ok = True
for step in range((tgt - cur) % 5):
if not press("DOWN", "5811"):
ok = False; break
time.sleep(1.2)
y = ring_row(img(fresh())); i = main_menu_item(y)
want = (cur + step + 1) % 5
print(f" step {step+1}: ring y {y} -> {NAMES[i] if i is not None else '??'} "
f"(want {NAMES[want]})", flush=True)
if i != want:
print(f" 🔴 CONTROL FAILED walking to {name}"); ok = False; break
if not ok:
results[name] = "skipped (navigation control failed)"; continue
if not press("A", "5800"):
results[name] = "skipped (A not delivered)"; continue
if wait_until(lambda x: differs(x, MAIN) > 0.20, f"{name} to open", 40) is None:
results[name] = "skipped (A did not change the screen)"; continue
time.sleep(3.0)
S1 = fresh(); img(S1).save(f"{OUT}/{name.replace(' ','_')}-S1.png")
print(f" S1 opened: glyph {glyph(S1)}, "
f"{100*differs(S1, MAIN):.1f}% from main", flush=True)
if not press("DOWN", "5811"):
results[name] = "skipped (DOWN not delivered inside)"; continue
time.sleep(2.5)
S2 = fresh(); img(S2).save(f"{OUT}/{name.replace(' ','_')}-S2.png")
moved = differs(S1, S2)
print(f" S2 after 1 DOWN: {100*moved:.2f}% of the frame changed", flush=True)
# CONTROL: the cursor must have moved, and by a LOCALISED amount -- a whole
# screen changing means the DOWN left the screen, not moved a cursor.
if moved < 0.0005:
results[name] = f"VOID (control): nothing changed on DOWN ({100*moved:.3f}%)"
press("B", "5801"); time.sleep(3); continue
if moved > 0.20:
results[name] = f"VOID (control): {100*moved:.1f}% changed — DOWN left the screen"
press("B", "5801"); time.sleep(3); continue
print(f" ✅ CONTROL PASSED: a localised change ({100*moved:.2f}%)", flush=True)
if not press("B", "5801"):
results[name] = "skipped (B not delivered)"; continue
if wait_until(lambda x: differs(x, MAIN) < 0.15, "the main menu", 60) is None:
results[name] = "VOID: B did not return to the main menu"; continue
if not press("A", "5800"):
results[name] = "skipped (A not delivered on re-entry)"; continue
if wait_until(lambda x: differs(x, MAIN) > 0.20, f"{name} to reopen", 40) is None:
results[name] = "VOID: re-entry did not change the screen"; continue
time.sleep(3.0)
S3 = fresh(); img(S3).save(f"{OUT}/{name.replace(' ','_')}-S3.png")
d1, d2 = differs(S3, S1), differs(S3, S2)
print(f" S3 re-entered: {100*d1:.2f}% from S1 (opened on), "
f"{100*d2:.2f}% from S2 (left on)", flush=True)
if min(d1, d2) > 0.15:
results[name] = f"VOID: re-entry matches neither ({100*d1:.1f}% / {100*d2:.1f}%)"
elif d2 < d1 * 0.5:
results[name] = f"PERSISTS (S3 {100*d2:.2f}% from where left, {100*d1:.2f}% from opened)"
elif d1 < d2 * 0.5:
results[name] = f"RESETS (S3 {100*d1:.2f}% from opened, {100*d2:.2f}% from where left)"
else:
results[name] = f"UNDECIDED ({100*d1:.2f}% / {100*d2:.2f}%)"
print(f" => {results[name]}", flush=True)
press("B", "5801"); time.sleep(3)
print("\n================ SUMMARY ================")
for k, v in results.items():
print(f" {k:12} {v}")
print("SUBMENU FOCUS SWEEP DONE", flush=True)