Start attempt 1/5 as an evidence-and-comment-only repair of a094bf2b7. Preserve behavior, assertions, shader/SPIR-V, register, package, sequential review, and no-landing-before-dual-PASS constraints.
Record a094bf2b7 as behaviorally retail-clean but unlanded because the binding evidence/prose contract failed. Preserve the sequential-review stop: the production lens did not run, no graphical gate ran, and G3/G4 remain unpassed.
Record the owner override after the required stop and bind the one-attempt repair to exact EnvCell pass membership, retail CLIP state, bounded source scratch, and production-path discriminators. Keep G3/G4 unpassed and the implementation unlanded pending dual review.
The 'empty plane under fog' frame the S4-c1 capture-pose self-gates produced
at frame 15 is the terrain seen from below by a falling player, not a draw
change: the camera probe holds playerCell=0xA9B4013F at (134.07,17.36) while
z drops 96 -> -20 m after materialization. The S3-state run had seated the
same request outdoors in 0xA9B40029 and stood. Placement is not the campaign's
domain; the gate pose is moved so the G3 frame is deterministic and the
original pose is kept in the issue as the reproduction. #462 recurrence on
the pre-S4 control build noted.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The two-list/router/flush core of 048d5b12f is exact at the bytes; the
failures are at the call sites (a data-driven row 5 turned into a throw;
detail-surface input hardcoded false while the drain applies detail),
the register (the structurally empty CLIP list is undeclared; AP-237's
example is wrong; AP-34 not retired by convention), the first-for-list
quantifier, the all-KnownFailure gate, and the skipped C4 that turns out
to be bounded.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Written for a fresh session on another account: read order, exact tips and
worktrees, what S1-S4-c1 landed with their proof, how to find the state of
the chunk 2 implementer dispatched this morning, the #464 attribution plan
(RenderDoc pixel history on the scripted artifact frame - the lead's own
job once ACE is up), the Holtburg doorway frame read, tooling gotchas, and a
paste-prompt.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The 2026-09-03 session landed S4 chunk 1 at 766f9e749 but ran out of
credits before writing the ledger row: round-2 lens verdicts, the landing
gates, the two capture-pose self-gates whose Holtburg doorway frame came
up as an empty plane under fog, the lead's read of that frame (the client
seated the player in the house-interior cell at the doorway threshold, so
the frame roots indoors with the eye outside), the crashed pre-S4 control,
and the S4-c2 dispatch that died with no commit.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
S4-c1 per docs/research/2026-09-01-overhaul/s4-depth-alpha-packet.md §6.
S3 chunk 2 already landed the persistent portalsDrawnCount latch, the
gated clear, the exit-seal counting, and the look-in isolation — this
chunk covers only what §1/§2 of the packet name as still owed: C0-C3.
C0 — far-punch depth constant (R1: DrawPortalPolyInternal @0x0059bc90's
tail). portal_depth.vert's punch branch carried the decimal 0.99999988,
which reinterprets as bits 0x3F7FFFFE — fifteen ULPs FARTHER from the
camera than retail's real constant, bits 0x3F7FFFEF. Now writes
`uintBitsToFloat(0x3F7FFFEFu)` so the exact bits survive the GLSL/SPIR-V
compiler instead of trusting a decimal literal to round-trip unchanged.
Recompiled via tools/compile-shaders.ps1 (glslc 1.4.350.0 backend
recorded, managed shaderc path used); portal_depth.vert.spv's SHA-256
re-pinned in VulkanShaderManifestTests
(51c60d0924d62c61548efcf5f9e7672a121b1b68ca0a06755e32f1a4d73a8acf,
was 4ac1c452e7ac0d08a32f67fb03f21229af2d1605baa81f407240a3626251dfd7).
T1 (new Fact PortalDepthVert_FarPunchConstant_MatchesRetailExactBits in
VulkanShaderManifestTests.cs): a SOURCE pin — reads portal_depth.vert's
punch line and reinterprets whatever literal it carries (uintBitsToFloat
hex or a plain decimal) as raw bits, asserts == 0x3F7FFFEF. Verified
against the PRE-CHANGE source by hand-reverting the line to
`clipPos.z = clipPos.w * 0.99999988;` and re-running just this test:
Assert.Equal() Failure: Values differ
Expected: 1065353199
Actual: 1065353214
(1065353199 = 0x3F7FFFEF, 1065353214 = 0x3F7FFFFE). Line restored and
the test re-confirmed green afterward. MUTATION: any other literal fails
the same way.
C1 — the ±12 local-input reject (R2: 0x59BCD6-0x59BD28 then
0x59BD40-0x59BD66). The Ghidra arbitration table in
oh1-depth-lifecycle.md governs over the pseudo-C's own nested-if reading
of the four x87 FCOM results (BinaryNinja's `test ah, 0x44` condition
synthesis is FPU-flag-ambiguous and reads backward at face value — see
feedback_bn_decomp_field_names.md on decompiler flag mush as an artifact
class, not semantics): the table's row says "whole poly on any
local-input x/y == +/-12 boundary is rejected before count/clip" — taken
as written, not re-derived from the pseudo-C's literal branch nesting.
Ported as one shared predicate,
WalkVisibilityMath.IsRejectedByPortalPolygonBoundaryGuard(ReadOnlySpan
<Vector3>): true iff any vertex's X or Y is exactly +12f/-12f (retail
tests LOCAL x/y before xformStart, the world transform). Wired at BOTH
producers that own the LOCAL polygon before it leaves cell/building
space:
- WalkFrameDriver.OnPunchGeometry (the walk's punch-event producer,
IWalkEventSink.OnPunchGeometry) — checked on the building-local
WalkPolygon.Vertices before TransformToWorld; a hit returns before
MarkIfGrown/any event append (retail's reject -> transform -> clip
-> count order).
- RetailPViewPassExecutor.DrawPortalDepthWrite (the exit-seal
enumeration behind DrawExitPortalMask, the sole caller) — checked on
cell.PortalPolygons[index]'s local vertices before the
Vector3.Transform loop; a hit `continue`s with no `submitted++`.
T2 (three layers):
1. WalkVisibilityMathTests.cs — direct unit tests of the predicate:
Boundary_guard_rejects_a_polygon_with_one_vertex_exactly_on_plus_minus_12
(Theory, x/y == +-12 each), Boundary_guard_admits_a_polygon_whose_
nearest_vertex_is_just_inside_12 (Theory, x/y == +-11.999),
Boundary_guard_rejects_the_whole_polygon_even_when_only_one_of_
several_vertices_hits_it, Boundary_guard_ignores_the_vertical_z_
component, Boundary_guard_admits_the_empty_polygon.
2. WalkFrameDriverTests.OnPunchGeometry_RejectsWholePolygonOnExact
PlusMinus12LocalVertex_ButPunchesJustInside — functional: feeds
OnPunchGeometry a polygon with a vertex at x=12 (no PunchFan/no
"PUNCH:" log line) then one at x=11.999 (punches normally,
leaf.Punches has exactly one entry, log has exactly one "PUNCH:3@v0").
3. RetailPViewPassExecutorTests.DrawPortalDepthWrite_RejectsDegenerate
LocalPolygons_BeforeTransformOrSubmission — a real functional test of
DrawPortalDepthWrite needs a live PortalDepthMaskRenderer the suite
has no fake for, so this is a compiled-call-graph pin (this file's
established pattern for exactly this situation): the guard call
precedes both the Vector3.Transform loop and
PortalDepthMaskRenderer.DrawDepthFan by IL offset, gated by a
conditional branch immediately after it.
MUTATION texts, all verified live during this session then reverted:
- OnPunchGeometry_RejectsWholePolygon... with the C1 guard deleted from
OnPunchGeometry:
Assert.Single() Failure: The collection contained 2 items
Collection: [WalkPolygon { Plane = WalkPlane { Normal = <0, 0, 1>, D = -3 }, Vertices = [<0, 0, 3>, <12, 0, 3>, <5, 5, 3>] }, WalkPolygon { Plane = WalkPlane { Normal = <0, 0, 1>, D = -3 }, Vertices = [<0, 0, 3>, <11.999, 0, 3>, <5, 5, 3>] }]
- DrawPortalDepthWrite_RejectsDegenerateLocalPolygons... with the C1
guard deleted from DrawPortalDepthWrite:
Expected call to WalkVisibilityMath.IsRejectedByPortalPolygonBoundaryGuard.
- Boundary_guard_admits_a_polygon_whose_nearest_vertex_is_just_inside_12
with the predicate widened to `MathF.Abs(x) >= 11.99f ||
MathF.Abs(y) >= 11.99f` (all four rows):
Assert.False() Failure
Expected: False
Actual: True
- Boundary_guard_rejects_a_polygon_with_one_vertex_exactly_on_plus_
minus_12 with the predicate narrowed to strict `x > 12f || x < -12f
|| y > 12f || y < -12f` (all four rows):
Assert.True() Failure
Expected: True
Actual: False
- Boundary_guard_rejects_the_whole_polygon_even_when_only_one_of_
several_vertices_hits_it with the guard checking only
localVertices[0] instead of looping every vertex:
Assert.True() Failure
Expected: True
Actual: False
C2 — no pipeline change for R3 (depth ALWAYS/write/no-cull, color writes
ENABLED with a zero-alpha SRCALPHA/INVSRCALPHA blend). acdream's
PortalDepthMaskRenderer.Rhi.cs:92,100 sets ColorWrite=false alongside
Blend=None; portal_depth.frag writes no color output at all. Provably
pixel-identical (retail's blend collapses to dst'=dst when srcAlpha is
fixed at 0, for any RGB) and the write mask is the SAFER mechanism going
forward (structurally blocks any future accidental color write,
independent of an authored zero-alpha invariant). Added register row
AD-119 to docs/architecture/retail-divergence-register.md (the next free
id after AD-118), citing DrawPortalPolyInternal @0x0059bc90 and
PortalDepthMaskRenderer.Rhi.cs; section 2's active-row count and running
header note updated (90 -> 91).
C3 — the truth table + cross-frame latch tests. The (root kind,
draw_landscape, outside-view count, previous count) table's cells are
mostly already covered by S3 chunk 2's own tests — this chunk adds only
the genuinely missing rows/cases, and leaves every existing test
untouched:
Pre-existing coverage (named, not reproduced):
- interior, ov==0, prior==0 ->
RunFrame_InteriorFloodWithNoExitView_SkipsLandscapeAndNeverFlushesClearsOrSeals
- interior, ov>0, prior==0 ->
RunFrame_InteriorFloodWithExitView_FreshDriverSkipsTheGatedClearThenDrawsSealsAndFloodCells
and OnInteriorFloodDrawTurn_FirstOvFrameSkipsClear_SecondFrameArmedByFirstsSealsClears
(its own frame 1)
- interior, ov>0, prior>0 (T4's "frame 1 seals N>0 -> frame 2
clears" half) ->
OnInteriorFloodDrawTurn_FirstOvFrameSkipsClear_SecondFrameArmedByFirstsSealsClears
(its own frame 2)
- T4's "frame 1 seals 0 -> frame 2 does not clear" half (repeated
across three consecutive ov>0 frames, subsuming the two-frame
case) -> OnInteriorFloodDrawTurn_FloodWithNoExitPortal_NeverClearsAcrossFrames
- one look-in isolated from the root latch ->
LookInDrawCells_NeitherArmsNorConsumesThePortalsDrawnCounter
No further T4 test was added — the two existing facts above already
prove both halves of the two-consecutive-frames latch case exactly.
New rows added this chunk:
- WalkFrame_OutdoorRoot_NeverFiresTheInteriorClearSealMachinery: root
kind == OUTDOOR. RetailFrameWalk.WalkFrame's outdoor branch
((cameraCellId & 0xFFFF) < 0x100) calls DrawLandscape directly and
never calls DrawInside/OnInteriorFloodDrawTurn at all, so the whole
LFLUSH/stamp/CLEAR/SEALS mechanism structurally cannot fire —
driven end-to-end through RunFrame with an outdoor cameraCellId,
asserting SKY present, LFLUSH/CLEAR/SEALS absent, counter stays 0.
MUTATION (verified, then reverted): added a stray
`sink.OnInteriorFloodDrawTurn([], 1);` call to WalkFrame's outdoor
branch:
Assert.DoesNotContain() Failure: Item found in collection
↓ (pos 1)
Collection: ["SKY", "LFLUSH", "SEALS"]
Found: "LFLUSH"
- OnInteriorFloodDrawTurn_OvZeroAfterAPriorArmedCounter_LeavesTheLatch
CompletelyUntouched: interior, ov==0 immediately after an EARLIER
ov>0 frame armed the counter — proves the counter is left EXACTLY
as an earlier frame left it (not merely "not cleared this frame"),
since S3 §8.1 R3 gates the ENTIRE outside_view.view_count>0 block,
including the read-then-zero decision itself, on ov>0.
MUTATION (verified, then reverted): moved
`int armed = PortalsDrawnCount; PortalsDrawnCount = 0;` out of the
`if (outsideViewCount > 0)` gate in
WalkFrameDriver.OnInteriorFloodDrawTurn (unconditional
read-then-zero every call):
Assert.Equal() Failure: Values differ
Expected: 1
Actual: 0
(every OTHER WalkFrameDriverTests fact stayed green under this same
mutation — this new test is the only one that catches it).
- MultipleLookIns_WithinOneFrameAndAcrossFrames_NeverTouchTheRootLatch
(T5): extends the single-look-in fact to TWO look-ins in one frame
then a THIRD in a later frame. MUTATION (verified, then reverted):
a `_mutationLookInCalls` counter in HandleDrawCellsTurn's
LookInStatic branch that resets PortalsDrawnCount on the SECOND
look-in call:
Assert.Equal() Failure: Values differ
Expected: 1
Actual: 0
— while LookInDrawCells_NeitherArmsNorConsumesThePortalsDrawnCounter
(one look-in only) stayed green under the identical mutation,
confirming this test's incremental value over the existing single-
look-in fact.
Gates: dotnet build (App.Tests and App) 0 warnings/0 errors; hermetic
lane 6832/6832 passed; InstalledDat lane against
C:/Users/erikn/Documents/Asheron's Call — exactly the four known
failures (TowerAscentReplayTests.TowerAscent_StaircaseStaysConeVisible_
EveryStep, LayoutImporterMediaBearingChildSweepTests.
MainGameUiAndChatInput_MediaBearingChildrenNowBuildAsRealWidgets and
LayoutImporterInvisibleSweepTests.EveryAuthoredInvisibleWidget_
StartsHiddenAcrossAllLayouts — both #383 — and
WalkTraceConformanceTests.Oh_doorway_still_first_frame_diff #458),
243 passed / 1 skipped / 4 failed / 248 total, no new failures; shader
tests (VulkanShaderDescriptorContractTests/VulkanShaderManifestTests/
RenderPackSpirvValidatorTests/SkyVertexLayoutTests) 35/35; register
tests (Divergence|Register filter) 52/52.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Sixteen DrawPortalPolyInternal attempts per frame, vertex-identical to
acdream's sixteen collect-time punches; twelve lie entirely on the local
y=12 plane and are rejected by the ±12 guard in both clients, so the hall
building's doorways are never punched in either.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Retail admits exactly acdream's cells and building-portal groups in the
same order; its alpha-depth frame punches each portal polygon to far depth
right before that group's cells and seals the exit views last. The
capture becomes S4-c1's fifth transcript pose.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Root 0xF4180114 with seven outside views; retail admits the same seven
hall building-portal groups in the same order as acdream, yet shows the
solid far face above the arch. The fifth transcript pose for S4.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
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>
The automation cannot tilt the camera, so the owner's screenshot stays the
record of the frame; the transcript shows the interior root flood of the
three stacked stair cells with six outside views and fifteen building-
portal groups including the hall cells. The retail capture at this pose
decides depth (S4) versus admission.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Owner client with both probes: root 0xF4180114, eye (39.89,17.25,182.36),
sweep uncontacted; the hall's interior shows above the bottom arch where
retail shows the solid far face, which the DAT gives as five exit portals.
The retail capture cathedral-stair-arch is requested as the fix oracle.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The eye is legitimate (retail's sweep passes the same opening); the defect
is the draw of the stairwell cells seen back through the building's
exterior portals. Matrix row invariant re-worded; the one retail capture
to request is the oh-capture walk + alphadepth at exactly this eye.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Read-only decomp + real-DAT replay: eleven sweep/root differences, each
unreachable at the pose or more constrained than retail; the replay seats
the pivot in 0xF4180114, stops the boom on its east pier at y=16.448, and
the walk from that root floods 114/113/112 with seven exit views. The
owner's probe launch line now also sets ACDREAM_PROBE_FACILITY_STAIRS.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Drafted from the five alphadepth captures (10,556 AM lines, every clip=0;
FL by return address 13,705 / 309 / 18 / 23) and the OH1 alpha-list
contract. Dispatch waits for S4-c1 to land.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Four probe-on self-gate rounds (run/zoom/tilt/mid) with the camera cell,
root and eye logged every frame: sweep ok, eye in root, no fallback, in
every frame. The DAT shows 0xF4180113/0xF4180114 are one stairwell split
horizontally; the zoomed-out eye stops 0.31 m in front of 0x114's
nine-vertex EXIT portal. Owner asked for one probe-on reproduction.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The held-forward route captured the owner's running symptom: the first frame
after the press has the chase camera above and outside the stairwell, the
next is clean. Same defect as the zoom-out. The runs also showed the
character running in place at one corridor spot for 5+ s (#467, movement,
outside the campaign).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The round-0 retail and production lenses both found the guard's
quantifier inverted (the lead's paraphrase); the production lens's DAT
scan found 2,163 exit polygons lying entirely on a ±12 plane, which makes
the guard the likely mechanism behind retail's never-sealed seam family.
The PM/PC transcript comparison becomes the chunk's own gate.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The S4-c1 retail lens caught the lead's paraphrase inverting retail's
predicate (any vertex on any plane vs every vertex on the same plane);
the contract and T2 now state the decomp's four per-plane predicates.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Owner decision 2026-09-03: an improvement on retail, deferred until G4
passes; retail mode off; registered when built.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The two lenses passed 6575cfcee and left these: AD-23 described the
ClipSlotCull the fix round deleted; AD-17's 'dormant, live for the flat
path' clause was false (the routing had no caller anywhere); the
ProbeClipRouteEnabled doc named a deleted producer; the per-view EC
print ran its loop with the flag off; the replay-side derivation still
prints one EC per flood cell while production prints one per live view,
which no fixture can tell apart today — recorded at the harness for S5.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
