docs(issues): #464 owner's tilted frame with the walk transcript — same admission set, seventh exit view; the open face shows the hall interior where retail shows the shell wall

The frame is saved as logs/464-owner-tilted-frame.walk.txt. Narrowed to
the building shell versus interior-cell order/clip at the stairwell's open
face in the outside pass (S4), with a possible building-portal admission
component only the cathedral-stair-arch retail capture can settle.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

docs/ISSUES.md

@@ -109,6 +109,8 @@ then `/teleloc F4180113 39.502899 19.589222 181.119247 1 0 0 0`, zoom out, tilt
 
 **Lead transcript at the pose (2026-09-03 22:07, `logs/selfgate-20260903-220639-464-owner`, `ACDREAM_DUMP_WALK_TRANSCRIPT=1` + both probes, code `27ff6bb04`):** the automation's `drag at` verb does not drive mouse-look, so the four frames all carry the DEFAULT camera (eye `(39.50,17.33,183.18)`, 0.8 m higher and pitched further down than the owner's) and `464-owner-00-default.png` does not show the hall above the arch — the artifact needs the owner's slight upward tilt, so the owner's screenshot stays the record of the frame. The walk transcript of that default frame (1,582 lines: `F`, `P f4180114`, `DI f4180114`, 16 `DC`, 36 `EC`, 34 `OC`, 13 `BLD`) is still the useful part: the interior pass is `DC pv=0 ov=6 n=3: f4180114 f4180113 f4180112` (root flood of the three stacked stair cells with SIX outside views), and the outside pass (`pv=1`) draws fifteen building-portal groups including the cathedral's OTHER interior cells `0xF4180106`–`0xF4180110` (the hall: `f4180109 f4180108 f4180107 f418010e f418010b`, `f4180108 f4180107 f418010f f4180110 f4180106`, `f4180107 f418010c f418010f f4180110 f4180106`, …) after `f3180105`, `f4180100`–`f4180105`. So the hall's cells are admitted through the building's exterior portals in the outside pass, as retail also does at the two captured cathedral poses where the S3 transcript gate is exact; the tilted frame differs in what the DEPTH lets through — the exit punches and seals and the gated Z clear at the stairwell's open face (S4), or a building-portal admission retail would refuse from this eye (only the requested `cathedral-stair-arch` retail capture can tell the two apart). Plan: land S4-c1 and S4-c2, add the capture as the fifth transcript pose, re-gate this pose.
 
+**Owner's TILTED frame with the transcript (2026-09-03 22:17, owner client armed by the lead with `ACDREAM_DUMP_WALK_TRANSCRIPT=1` + both probes; log `logs/launch-464-transcript-20260903-221619.log`, the frame saved as `logs/464-owner-tilted-frame.walk.txt`):** `/loc` player `(39.50,20.08,180.78)`, eye `(39.76,17.51,182.31)` (2.60 m back, uncontacted), root `0xF4180114`. Walk: `P f4180114`, `DI f4180114`, interior `DC pv=0 ov=7 n=3: f4180114 f4180113 f4180112` (SEVEN outside views — the default camera at the same spot had six; the tilt brings one more exit of the open face into view), then thirteen `BLD` turns whose outside pass (`pv=1`) draws sixteen building-portal groups — `f3180105`, `f4180102`, `f4180101/f4180100` ×3, `f4180103/f4180105`, `f4180104` ×3, and under building `f418000a` the hall: `f4180109`, `f4180109 f4180108 f418010e f418010b`, `f4180109`, `f4180108 f4180107`, `f4180108 f4180107 f418010f f4180110 f4180106`, `f4180107`, `f4180107 f418010f f4180110 f4180106` — and the interior repaint `EC f4180112 f4180113 f4180114` LAST. The admission set is the SAME as the default frame's; the tilt only adds the seventh exit view. So what differs on screen is not which cells draw but what the open face shows: through `0x114`'s exit portals (the far face is portal polygons only, `emitPos=False emitNeg=False`, nothing of the cell itself is drawn there) acdream shows the hall's INTERIOR cells drawn in the outside pass, while retail shows the hall's EXTERIOR wall — building-shell geometry of the outside pass — solid above the arch. **Narrowed: the building shell versus interior-cell order/clip at that face in the outside pass (which of the cathedral's building portals retail admits from an exit view, and whether the shell's wall occludes the hall cells), S4 depth/alpha with a possible building-portal admission component.** The `cathedral-stair-arch` retail capture (walk + alphadepth at this pose) decides it: retail's `DC` groups under the cathedral's `BLD` turn versus ours, and its `PM`/`FL` order at the seven exit views.
+
 ## #463 — Soak `turn` samples can report a 0.1 ms GPU p50 (the turn window catches non-world frames)
 
 **Status:** OPEN — found 2026-09-03 by the Campaign OVERHAUL v2 S3 review's perf re-measure.

tools/overhaul-selfgate/route-464-look.txt

@@ -0,0 +1,61 @@
+# #464 reproduction, part 9: pitch the camera UP with instant mouse-look plus
+# synthetic mouse moves (the keyboard Raise/Lower shortens the boom; drag at
+# is not mouse-look). Run with ACDREAM_DUMP_WALK_TRANSCRIPT=1 + both probes.
+
+wait world-visible 180000
+sleep 4000
+
+command /teleloc F4180113 39.502899 19.832422 180.976700 1.000000 0.000000 0.000000 0.000000
+sleep 2000
+wait world-visible 60000
+sleep 4000
+screenshot 464-look-00-default 15000
+
+mousemove 880 520
+sleep 200
+input down CameraInstantMouseLook
+sleep 200
+mousemove 880 500
+sleep 100
+mousemove 880 480
+sleep 100
+mousemove 880 460
+sleep 100
+mousemove 880 440
+sleep 600
+screenshot 464-look-01-up80 15000
+mousemove 880 420
+sleep 100
+mousemove 880 400
+sleep 100
+mousemove 880 380
+sleep 100
+mousemove 880 360
+sleep 600
+screenshot 464-look-02-up160 15000
+mousemove 880 330
+sleep 100
+mousemove 880 300
+sleep 100
+mousemove 880 270
+sleep 100
+mousemove 880 240
+sleep 600
+screenshot 464-look-03-up280 15000
+input up CameraInstantMouseLook
+sleep 500
+screenshot 464-look-04-released 15000
+
+input press CameraInstantMouseLook
+sleep 300
+mousemove 880 500
+sleep 100
+mousemove 880 440
+sleep 100
+mousemove 880 380
+sleep 600
+screenshot 464-look-05-toggle-up140 15000
+input press CameraInstantMouseLook
+sleep 300
+
+close-client