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>
F1 (BLOCKING, punch-fan view alignment). ClipFrameAssembler.
ReassembleOutsideViewFromWalk now keeps exactly ONE outside-view slice per
walk view, index-aligned with RetailFrameWalk.DrawBuilding's own
ActiveViewIndex (retail building_view = portal_view_num @0x0059f3bf): a
collapsed view (ClipPlaneSet.IsNothingVisible — retail polyClipFinish under
a degenerate view leaves <3 vertices, 0x59BDBC-0x59BDBF) now appends its
own ClipViewSlice.NothingVisible=true slot instead of being skipped, which
used to shift every later view's slice down by one. DrawWalkPunchFan draws
NOTHING for a NothingVisible slice and throws ArgumentOutOfRangeException
when activeViewIndex is out of range (fail-loud; never draws unclipped).
New tests: WalkOutsideViewReassemblyTests.
FirstViewCollapses_SecondSurvives_SlicesStayIndexAligned and
PunchLeaf_UsesIndexAlignedSlice_DrawsNothingForCollapsed_ThrowsOutOfRange.
MUTATION M1 (revert the append-nothing-on-skip fix, restoring the old
skip): both new tests fail with
Assert.Equal() Failure: Values differ
Expected: 2
Actual: 1
MUTATION M2 (restore the old ": default" unclipped fallback in
DrawWalkPunchFan): the punch-leaf test fails with
Assert.Equal() Failure: Values differ
Expected: 0
Actual: 1
(both restored; verified clean afterward).
F2 (major, exit-seal pin through the real producer + both overflow cases).
The exit-seal CPU/GPU equivalence pin moved from ClipFrameLayoutTests (which
drove ClipFrame.AppendSlot(ClipPlaneSet) via the zero-caller
ClipFrameAssembler.Assemble) to the new
tests/.../Walk/WalkFrameDriverClipSealTests.cs, which drives
WalkFrameDriver.BeginFrame + the real IWalkEventSink.OnInteriorFloodDrawTurn
hook — the SAME path RetailPViewPassExecutor.DrawExitPortalMask resolves
through via WalkFrameDriver.InteriorFloodViewClipPlanesAt. The new pin
compares WalkFrameDriver's captured planes EXACTLY (not just geometrically)
against ClipPlaneSet.From's independent CCW-normalized-perpendicular
computation, since AppendClipSlot's per-edge formula is bit-identical to
ClipPlaneSet's own. Two overflow cases added: a 9-vertex portal view exit-seals
to exactly the 4 conservative AABB planes containing every source vertex
(over-include, never under-include — AppendClipSlot's fallback for a view too
complex for the 8-plane budget); a 9-vertex OUTSIDE view (the punch-fan
sibling, WalkOutsideViewReassemblyTests.
NineVertexOutsideView_PunchSliceHasZeroPlanes_DrawsUnclipped_NotNothingVisible)
produces a zero-plane, NOT-NothingVisible slice — draws fully unclipped,
distinct from F1's "draw nothing" state.
MUTATION M1 (flip the winding selection, `ccw = area2 < 0f`):
Assert.Equal() Failure: Values are not within 4 decimal places
Expected: 0.9191 (rounded from 0.919144988)
Actual: 0.8944 (rounded from 0.89442724)
MUTATION M2 (drop the normalize from the per-edge plane formula):
Assert.Equal() Failure: Values are not within 4 decimal places
Expected: 0.9191 (rounded from 0.919144988)
Actual: 0.7 (rounded from 0.700000048)
MUTATION M3 (delete the >8-plane AABB overflow branch — the per-edge loop
then indexes a fixed 8-plane stackalloc with a 9-edge polygon):
System.IndexOutOfRangeException : Index was outside the bounds of the array.
at WalkFrameDriver.AppendClipSlot(...) line 1915
at WalkFrameDriver.CaptureViews(...) -> CaptureCellViews -> CaptureCellViewRoute
-> EmitFloodTurns -> IWalkEventSink.OnInteriorFloodDrawTurn
(all three restored; verified clean afterward).
F3 (minor/major). RetailPViewPassExecutor.cs's false bracket comment
(neither KEEP clip — exit seals nor punch fans — ever called Enable/
DisableClipDistances; the six DisableClipDistances() call sites all sat
around the UNCLIPPED particle/weather draws, whose bodies are no-ops on the
only backend) is rewritten to the truth; the six no-op calls and the
private DisableClipDistances() wrapper are deleted.
IWorldPassSurface.EnableClipDistances/DisableClipDistances stay on the
INTERFACE — WorldScenePassExecutor (the separate flat-world path) still
calls them.
F4a (EC transcript exactness). WalkFrameDriver.EmitFloodTurns now prints one
"EC" line per LIVE VIEW of the flood cell (retail's setup_view/DrawEnvCell
loop, PView::DrawCells @0x005a4ab1-0x005a4acc, fires once per live portal_view
slice, not once per cell) via the captured route's own SliceCount — read
directly off _lookInTurns[viewRouteIndex] rather than through
InteriorFloodViewSliceCountAt's flood-index indirection, because this same
loop also serves a building's LOOK-IN flood (WalkDrawStage.LookInStatic),
whose cells never populate InteriorFloodCells (the interior ROOT flood only)
— indexing through that accessor threw ArgumentOutOfRangeException on a real
look-in turn during verification, fixed before landing. The actual CellShell
DRAW event is unchanged (the frame stamp still dedupes it to one submission).
F4b (transcript exactness, trailing weather OC).
WalkTraceConformanceTests.Recorder now implements IWalkEventSink.
OnWeatherTurn (the interface's silent no-op default previously left the
replay side of Signature8 blind to the weather turn's own "OC" line), and
WalkTraceReplayContext.Signature8(WalkOracleFrame) no longer excludes the
trailing per-frame weather ObjectCellTurn — every ObjectCellTurn now reads
literally, on both sides. Implementing OnWeatherTurn alone left
Still_fixture_first_frame_reproduces_exactly(terrace-edge.walk) diverging
ONLY on the newly-un-excluded trailing OC (EXPECTED ...OC:f4180104|
OC:f418000b vs ACTUAL ...OC:f4180104 — a length-12 tail, nothing else
differs across a 16.6k-char signature) because
WalkTraceReplayContext.WeatherGateOpen had always defaulted to false (the
harness never needed to fire the walk's weather hook before). Per the
review round's own instruction not to weaken the pin, the harness itself is
fixed instead: WalkTraceReplayContext now implements ViewerCellId (= the
harness's own WalkFrame cameraCellId argument, matching production's
RetailPViewFrameInput.ViewerCellId invariant) and WeatherGateOpen
(retail's own SmartBox::is_player_outside gate, (cellId & 0xFFFF) < 0x100,
with no App-level render-toggle concept to AND against). MUTATION (delete
Recorder.OnWeatherTurn): Still_fixture_first_frame_reproduces_exactly
(terrace-edge.walk) fails with EXPECTED ending "...OC:f4180104|OC:f418000b"
vs ACTUAL ending "...OC:f4180104" (the trailing OC missing) — restored;
verified clean afterward.
Fixture-row status (InstalledDat lane, WalkTraceConformanceTests, 14 rows):
13 pass, 1 fails — Oh_doorway_still_first_frame_diff, the PRE-EXISTING
[Trait("Status","KnownFailure")] #458 row (a documented block-plane
precision boundary at token index 165, LC/SC content, upstream of any
EC/OC — unaffected by and unrelated to this round's OC change). Every OTHER
row, including every eight-kind (Signature8) row that now compares the
weather OC literally, passes exactly.
F5 (major, dead per-cell clip machinery). WbDrawDispatcher.SetClipRouting
(the only writer of the per-instance routing-active flag) had ZERO
production callers; EnvCellRenderer.SetClipRouting was called only with
null (WorldScenePassExecutor's BeginFrame/AbortFrame) — no path could ever
arm the per-cell clip-region table, on the walk path OR the flat path (the
S3 landing-hygiene AD-17 correction's "live for the flat path" clause is
itself corrected here: dormant in every path, deleted). Deleted:
WbDrawDispatcher's SetClipRouting/ClearClipRouting and their four backing
fields; EnvCellRenderer's SetClipRouting and its backing field (its
RenderModernMDIInternal write now unconditionally clears instanceClipSlot);
the ProbeClipRouteEnabled-gated per-frame [clip-route-disp] dispatch probe
and its three backing fields (the OTHER two ACDREAM_PROBE_CLIPROUTE
producers, [clip-route] and [clip-route-scis] in WorldRenderDiagnostics.cs,
are untouched — a different, still-live mechanism); ResolveEntitySlot,
IsIndoorCellId and the ClipSlotCull sentinel; the ACDREAM_CLIP_DEBUG probe
(RenderingDiagnostics.ClipDebugNoShellTrim) and its docs/launch-options.md
row, since the "clip shells" branch it toggled between is gone.
ResolveSlotForFrame is simplified to a parameterless
`(uint Slot, bool Culled) ResolveSlotForFrame() => (0u, false)` per the
review's own framing ("it becomes slot 0, never culled") rather than
deleted outright, since its callers (WbDrawDispatcher.cs and
WbDrawDispatcher.WalkClassify.cs) still want the same two-value shape.
tests/.../Wb/WbDrawDispatcherClipSlotTests.cs (12 tests, all exercising the
deleted routing arm via ResolveEntitySlot/the active ResolveSlotForFrame
branch) is replaced with one pin on the new parameterless behavior.
Shaders: mesh_modern.vert, mesh_atmospheric.vert, mesh_detail.vert,
particle.vert and particle_mesh.vert each lose their CellClip struct +
binding=2 ClipRegionBuf fetch, gl_ClipDistance write loop, and gl_PerVertex
redeclaration (portal_depth.vert's own, separate gl_PerVertex/gl_ClipDistance
pair — the KEEP mechanism, its planes handed through the TerrainClip UBO at
binding=2 in the UBO namespace — is untouched; verified it never read the
deleted SSBO). Recompiled via tools/compile-shaders.ps1 (glslc backend
detected; managed shaderc path ran); 24/24 pairs compiled. Re-pinned in
VulkanShaderManifestTests.RetailOracleSpirvSha256 (the only three of the
five edited shaders that carry a byte-exact retail-oracle pin):
mesh_modern.vert.spv = f9ed4ee7140ccd136130559dbea68545f733f6022f52b085b6df4bcb787223c6
particle.vert.spv = 95ce6ecf834930a92da5c5fe9aef513b38b5ba104704b98c1606af71fe17eaf3
particle_mesh.vert.spv = 043482b97c2ed036511692f89c75a0a6c298aba48cb519e5e3aff7fe7ba6371b
(mesh_atmospheric.vert.spv / mesh_detail.vert.spv changed too but carry no
retail-oracle pin, so no re-pin was needed for them.)
tests/.../ParticleBindlessInstanceTests.cs's
Assert.Contains("clipRegions[aClipSlot]", vertex) — a real breakage the
grep sweep caught — is deleted with a note; the surrounding aClipSlot
attribute-declaration assertion stays.
Vertex-layout residue kept THIS round, fed 0 by the CPU, unread by any
shader (S5's instance-buffer-layout revisit): instanceClipSlot[] (binding=3
SSBO) in mesh_modern.vert/mesh_atmospheric.vert/mesh_detail.vert; aClipSlot
(a per-vertex attribute) in particle.vert/particle_mesh.vert.
Grep sweep — "SetClipRouting|_clipRoutingActive|ClearClipRouting|
BeginScissor|NdcScissorRect|SetTerrainClip|BindTerrainClip|TerrainBytes|
ScissorNdcAabb|UseScissorFallback" over src/ and tests/: EMPTY (also swept
and rewrote several PRE-EXISTING prose mentions of BeginScissor/
ClearClipRouting in RetailPViewPassExecutor.cs, RetailPViewRenderer.cs,
WorldPassSurface.cs and RetailPViewPassExecutorTests.cs that predate this
commit's own work, purely to satisfy the literal sweep). "clipRegions|
instanceClipSlot|aClipSlot": every remaining match is the vertex-layout
residue named above, or a comment/assertion describing that same residue.
Deviations from a fully literal reading of F5 (recorded here since none of
them cross the mandatory automated gates, all of which pass):
- The C#-side ClipRegions SSBO publish/bind pipeline
(RhiWorldPassSurface.PrepareClipFrame's publish, WorldFrameSections.
ClipRegions, and ClipFrame's internal std430 byte-packing —
RegionBytes/RegionBytesForTest/CellClipStrideBytes) is NOT deleted this
round; it is left as documented dead residue (verified zero shader
readers) rather than rewritten. The mandatory grep sweep targets the
shader-source identifier `clipRegions` (lowercase), which is fully clean;
the C# publish machinery sits behind ~8 files (ClipFrame.cs,
WorldPassSurface.cs, WbDrawDispatcher.Rhi.cs, WorldPassScope.cs,
GpuBindingModel.cs, EnvCellRenderer.Rhi.cs, ParticleRenderer.Rhi.cs,
WbDrawDispatcher.OrderedStream.cs, plus the ClipFrameLayoutTests std430
pins) and CellClipPlanesOffset/MaxPlanes are the SAME constants
portal_depth.vert's KEEP mechanism uses for its own, unrelated UBO layout
— a full rewrite carried materially higher regression risk than this
round's mandatory scope justified. ClipFrame.cs and WorldPassSurface.cs
now both name this residue explicitly in their doc comments for a future
round to finish.
- SetClipRegionSsbo/_sharedClipRegionSsbo in WbDrawDispatcher.cs (a
pre-existing, already fully dead GL-era write-only field/method with zero
callers, unrelated to per-instance clip-slot routing) is left untouched —
outside F5's named scope and outside the mandatory grep sweep.
- WalkTraceReplayContext.ViewerCellId/WeatherGateOpen (F4b, detailed above)
is an addition beyond the literal §12 text, made to keep the InstalledDat
lane at exactly the four established known failures without weakening
the newly-un-excluded OC pin.
Gates: dotnet build tests/AcDream.App.Tests/AcDream.App.Tests.csproj -c
Release and dotnet build src/AcDream.App/AcDream.App.csproj -c Release both
0 warnings/0 errors; dotnet build AcDream.slnx -c Release 0 warnings/0
errors; hermetic lane 6815/6815 passed; InstalledDat lane 243 passed / 4
failed / 1 skipped (exactly the four known failures: two #383 layout tests,
TowerAscentReplayTests, Oh_doorway_still_first_frame_diff #458);
VulkanShaderDescriptorContractTests/VulkanShaderManifestTests/
RenderPackSpirvValidatorTests/SkyVertexLayoutTests 34/34 passed.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Two live cdb captures on the paired 2013 client at the Holtburg doorway:
the blockset template (every Render::block_check call of one frame — 2,601
resident blocks x 2 exit views) proves retail tests a9c9 at ring slot
25,46 and returns PARTIALLY then OUTSIDE; the blockcheck template dumps
its four corner interval vectors (0 0 0 0 300.4|310.2 and 0 0 0 1001 ...,
slab 75..330). acdream's replay at the same P pose reproduces the sentinel
pattern, the four edge planes and both verdicts, drawing a9c9 once, with
clip heights 298.8/308.5 m — a 0.5 % plane difference (about 0.35 px of
projected door-vertex position). At the fixture pose that margin is what
flips the south-west corner from inside to outside on the fourth edge
plane, so retail's four-corner unanimity test says OUTSIDE where ours
says PartiallyInside. Not fixable bit-exactly short of D3D's x87
transform; the KnownFailure row stays with its comment rewritten, the
register carries AD-118, and the first filtered capture's silent miss
(cdb sign-extends poi() inside .if; use dwo()) is noted in the template.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Campaign OVERHAUL S3 chunk 1 fix round 2 (docs/research/2026-09-01-overhaul/
s3-walk-ownership-map.md §11.6), applied on top of 36be6b598 after the
three-lens re-review. H1-H6, the last round the plan allows.
H1 — the weather OC printed 2-4 times per interior-rooted frame instead of
once: the print (and the `_sky.RenderWeather` mesh draw it sits beside) lived
inside `DrawLandscapeSliceLate`, which `RetailPViewRenderer` calls once per
active `OutsideViewSlices` entry. Moved both into a new
`RetailPViewPassExecutor.DrawWeatherOnce`, called ONCE, unclipped/no-scissor,
after the slice loop in `DrawLandscapeDynamicsPhase` — retail's own
`GameSky::Draw(1)` runs once, after `LScape::draw`'s whole landblock loop.
The gate is extracted as a pure `ShouldDrawWeatherOnce(bool,bool,uint)`
predicate (retail's `SmartBox::is_player_outside` ANDed with the two render
toggles) so `RetailPViewPassExecutorTests` can pin "no OC while the player
stands indoors" without a live GL/DAT `SkyRenderer`; two structural
(CompiledCallGraph) tests prove the call moved out of the per-slice loop and
that the mesh draw + print each fire exactly once per invocation — this
codebase has no existing runtime-construction fixture for
`RetailPViewPassExecutor`, so the pin is structural + a testable pure gate
rather than an end-to-end GL drive. One deliberate deviation from the literal
"gated exactly as today": the print now runs through `_sky?.RenderWeather(...)`
(null-conditional, matching every other `_sky?.RenderSky(...)` call site in
this codebase) instead of an explicit `if (_sky is not null)` wrapper — a
missing sky asset no longer also suppresses the transcript print, since the
print's only job is trace fidelity and retail's own `GameSky` is never null.
H2 — WalkFrameDriver's EC-print comment still claimed "EC and OC counts are
always exactly equal"; replaced with the real citation
(holtburg-doorway-still.walk.log:1126,1131,1134,1137 — four EC prints for one
cell across four look-in DC turns), matching WalkTranscriptDump.
PrintEnvCellShell's own comment (already corrected in round 1).
H3 — the eight-kind signature's ORACLE side
(WalkTraceReplayContext.Signature8(WalkOracleFrame)) derived EC/OC from each
DC's cell list — the SAME derivation the REPLAY side's Recorder already used,
so the comparison could never disagree with itself on EC/OC placement or
content (how G7's SC-ordering regression shipped green with LC/SC). Now reads
its own literally captured EC/OC events, excluding only the trailing
per-frame weather OC — refined beyond the spec's literal "last event + P-cell
id" rule with an additional "P cell itself looks outdoor" check, after
cathedral-arrival and foundry-deep both proved the naive rule also strips an
interior root's OWN real trailing object-list turn when its nearest cell (the
reversed flood loop's last draw) happens to be the camera's own root cell.
All four kit-pose rows still reproduce: three exactly; #458 (this round's new
issue, formerly an inline-only note) re-verified at the SAME token index 165
under the new literal comparison.
H4 — OnSortCellTurn ran RequireOpenFrame + two range validations before
testing the flag. Since the hook is print-only (no stream side effect, unlike
OnLandCellTurn's unconditional WalkFrameEvent record), the flag check now
runs FIRST and returns immediately when off — flag-off cost drops to one
interface dispatch per visited land cell.
H5 — launch-options.md's ACDREAM_DUMP_WALK_TRANSCRIPT row: measured
≈1,200-1,400 lines per outdoor frame (terrace-edge 1,384; cathedral-arrival
1,269; doorway 1,187), replacing the earlier "600-800" estimate; documents
H4's residual flag-off interface-dispatch cost.
H6 — filed docs/ISSUES.md #458 for round 1's LOD-boundary land-cell
divergence (previously only an inline test comment); the InstalledDat lane's
known-failure set is now four (two #383 layout tests, TowerAscent, #458) —
confirmed by a clean run.
MUTATION CHECKS (both restored after confirming failure):
- H1: deleting the OC print inside DrawWeatherOnce made
DrawWeatherOnce_DrawsTheWeatherMeshAndPrintsExactlyOnce fail with
"Assert.Single() Failure: The collection did not contain any matching
items".
- H3: reversing AppendFloodTurns (OC before EC) made
Still_fixture_first_frame_reproduces_exactly(cathedral-arrival.walk) fail
("walk diverged from retail (cathedral-arrival.walk)"), diverging at token
index 1241: expected "EC:f4180112" vs actual "OC:f4180112".
Gates: hermetic App suite 6823/6823 passed; InstalledDat lane 244/249 passed
with exactly the four known failures (two #383 layout tests, TowerAscent,
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Campaign OVERHAUL S3 chunk 1 fix round 1 (docs/research/2026-09-01-overhaul/
s3-walk-ownership-map.md §11.5), applied on top of 079483bb6 after the
three-lens review. G1-G10, each verified against source and (where the item
adds a pin) mutation-checked.
G1 — the flag-off integration test was vacuous: Collect_TranscriptFlagOff_
ProducesNoConsoleOutput drove ONE frame, and WalkOracleTrace.Parse
structurally discards the last (open) frame, so it passed whether or not
the flag gate worked. Now drives TWO frames and asserts BOTH that Parse
sees zero frames AND that no captured line starts with any of the ten
transcript-kind prefixes (F /P /LS/LC /SC /BLD /DI /DC /EC /OC ) — robust
to unrelated console noise from parallel test classes.
G2 — LC/SC computed LodCellId before testing the flag (~1,300 land-cell
turns/frame of division/modulo/checked-conversion work paid even when the
flag is off). Both OnLandCellTurn and the new OnSortCellTurn now guard the
LodCellId call itself, not just WalkTranscriptDump's internal check.
G3 — the flag-on round trip never exercised LC/SC/BLD (the T2 fixture's
stub 1x1 landscape has zero blocks). New
OutdoorRoot_TranscriptFlagOn_PrintsLandscapeThenLandCellThenSortCellThen
Building_InOrder drives RetailFrameWalk.DrawLandscape directly (the same
deterministic "CY-only" admission technique the existing outdoor LandCell
test uses) over a landscape with one block + one building, and asserts
LS < LC < SC < BLD by position, with the exact cell/building ids.
G4 — the parts/alpha-depth smoke tests asserted FL/PC (Flushes/
DrawCellsSamples) but never AM/PM (MeshAdds/PortalPolyDraws) for the
alpha-depth log, even though PD/DM were already asserted for the parts
log. Added the missing Assert.Contains for MeshAdds and PortalPolyDraws.
G5 — launch-options.md's side-effects column now states the flag off/on
costs explicitly ("one bool read per print site, nothing else" / "one
console line per walk turn, ≈600-800 lines per outdoor frame").
G6 — removed the "DI f4180108... no, a9b4013f" self-correction in
Oh_doorway_still_first_frame_diff's comment; added a one-line note on why
it and Oh_foundry_deep_... stay bespoke Facts rather than folding into the
Theory (no functional reason, just history).
G7 — SC printed from OnLandscapeCellTurn, which RetailFrameWalk.
DrawLandscape calls AFTER DrawBuilding — retail's own order is
LC -> SC -> BLD (terrace-edge.walk.log:2331-2333), not LC -> BLD -> SC.
Fixed by adding IWalkEventSink.OnSortCellTurn, fired at the DrawSortCell
entry point (same alwaysDrawObjects||cellInView gate, before DrawBuilding);
the driver moves its SC print there and OnLandscapeCellTurn no longer
prints one.
G8 — the trailing "OC <viewer land cell>" retail prints once per complete
outdoor frame is GameSky::Draw's weather branch calling
DrawObjCellForDummies(after_sky_cell) @0x005070da, not a walk turn.
RetailPViewPassExecutor.DrawLandscapeSliceLate now prints it at the point
it actually calls RenderWeather; PrintObjectCellTurn's doc, which denied
this line existed, is corrected. New
Oh_terrace_edge_outdoor_frame_ends_with_the_weather_ObjectCellTurn pins
this directly against the OH capture: every complete outdoor frame's last
event is an OC whose cell id equals that same frame's own P-line cell id.
G9 — "EC and OC counts are always exactly equal per pose" was false
(terrace-edge: 12 EC vs 16 OC, the extra 4 being G8's weather OC, which
has no EC counterpart). The visit-scoped/stamp-dedupe conclusion was
right; only the count claim needed fixing. Both doc sites corrected.
G10 — the conformance signature dropped LC/SC/EC/OC entirely, which is
how G7 shipped green: no test ever compared their placement. Recorder now
tracks LC/SC via the new OnLandCellTurn/OnSortCellTurn hooks and derives
EC/OC per DC from PView::DrawCells's own two reverse loops (shell
far-to-near, then object-list far-to-near) — a look-in flood's DC derives
them immediately, the interior root's OWN flood defers to
OnInteriorFloodDrawTurn (verified against cathedral-leak.walk.log:4,1328,
where the root DC's own EC/OC print last). WalkTraceReplayContext.
Signature8 applies the identical derivation to oracle frames and never
produces the trailing weather OC (G8's own dedicated pin covers that
separately, matching the spec's "exclude from the comparison, assert
separately" instruction). Used ONLY for the four OH-capture-rooted
conformance rows (the older FW0 fixtures predate the LC/SC/EC/OC
breakpoints and carry none of those lines).
Applying the eight-kind check surfaced a real, separate bug: WalkLandscape
DatBuilder never set WalkLandBlock.LandblockId, so every LC/SC token any
DAT-built world ever produced read landblockId=0 — latent because the old
four-kind signature filtered LC/SC out. Fixed (one field, test-fixture-
only, no production behavior change). With that fixed, three of the four
OH rows (terrace-edge, cathedral-arrival, foundry-deep) reproduce EXACTLY
at the eight-kind level. Oh_doorway_still_first_frame_diff still diverges
by one extra LC/SC pair (token index 165: replay draws LC:a9c90001|
SC:a9c90001 that retail never does) — a real WalkLandscape.CheckBlocks
visibility question at a ring-2/ring-3 LOD boundary, out of this round's
print-only scope, needing the retail oracle to resolve. Marked
[Trait("Status","KnownFailure")] with the exact position recorded in its
own doc comment, matching TowerAscent's existing pattern — the InstalledDat
lane now shows FOUR known failures (the pre-existing two #383 layout tests
+ TowerAscent, plus this new finding), not three; never weakened the pin
itself.
Mutation checks (all four restored after observing the failure):
- G1: forcing WalkTranscriptDump.Enabled true made
Collect_TranscriptFlagOff_ProducesNoConsoleOutput fail with
"Assert.DoesNotContain() Failure: Filter matched in collection ... Collection:
[\"DI 00000100\", \"DC pv=00000000 ov=1 n=2: 00000100 00000101\", \"LS\", ...]".
- G3: deleting PrintLandCell's Console.WriteLine made the new outdoor test
fail with "expected an LC line".
- G4: making MeshAddPattern unmatchable made all five
Alpha_depth_log_parses_with_nonempty_samples rows fail with
"Assert.Contains() Failure: Filter not matched in collection".
- G7/G10: moving OnSortCellTurn back after DrawBuilding made the
terrace-edge conformance row fail with "walk diverged from retail
(terrace-edge.walk)", first divergence exactly at its first building
(f518002e): expected ...SC:f518002e|BLD:f518002e..., actual
...BLD:f518002e|SC:f518002e....
Gates: hermetic lane 6,816/0 (baseline 6,814 + G3's new test + G8's new
test); InstalledDat lane 244/4 known failures (two pre-existing #383
layout tests + TowerAscent + the new Oh_doorway_still_first_frame_diff
finding, documented above).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Campaign OVERHAUL S3 chunk 1 (docs/research/2026-09-01-overhaul/s3-walk-ownership-map.md
§11): the transcript-kinds/fixtures/print-only-emitter half of the walk work,
built AFTER chunk 3 landed LC/SC (the per-land-cell interleave).
B1 — the emitter (print-only, never gates admission/depth/order):
- ACDREAM_DUMP_WALK_TRANSCRIPT=1 is read once into RuntimeOptions.DumpWalkTranscript
(rule 4) and handed to RenderingDiagnostics.DumpWalkTranscriptEnabled (rule 5,
a settable static, not a second env read) once at GameWindow construction — the
deep walk call sites have no reachable RuntimeOptions reference.
- WalkTranscriptDump (new) prints the OH line kinds — F/P/LS/LC/SC/BLD/DI/DC/EC/OC —
to Console at the exact points retail's cdb breakpoints sit
(tools/walk-oracle/oh/oh-capture-walk.cdb.template), gated internally so every
method bails out before any string work when the flag is off.
- Every call site lives in WalkFrameDriver.cs, at the point the driver already
processes that turn: Collect (F/P, after BeginFrame), Emit's DI/LS/DC/BLD cases,
OnLandCellTurn/OnLandscapeCellTurn (LC/SC, at LOD resolution via the new
WalkTranscriptDump.LodCellId helper, before the 8x8-bucket expansion), and the
EmitFloodTurns/EmitCellContentsTurn loops (EC/OC — both UNCONDITIONAL per flood
visit, matching the OH captures' always-equal EC/OC counts; retail's own
DrawEnvCell stamp dedupe sits past the breakpoint, inside the function).
- DC's "pv=" field encodes interior(0)/outdoor(1) as an 8-hex-digit 0/1 so it
satisfies the same pv=[0-9a-f]{8} regex real captures use; derived from
_currentDcStage at the DC event (CellStatic = interior pview, else outdoor).
- The frame-root pose (origin, quaternion) is a reasonable orthonormal basis built
from the walk's own CyPlane.Normal forward vector and WalkLandscape's own
ViewerWorldOriginX/Y block origin — self-consistent for the round-trip parser,
not a byte-exact reproduction of retail's Frame (B4's diff never compares P).
B2 — WalkOracleTrace learns EC/OC event kinds (LC/SC already existed from chunk 3).
New WalkOraclePartsTrace.cs holds two small read-only parsers for the parts log
(PD/DM) and the alpha-depth log (AM/FL/PM/PC) — records only, no validator, no
canonical JSONL, no new tool.
B3 — fixtures: the five OH walk captures join WalkOracleTraceTests.AllFixtures
(now (root, name) pairs — FW0's own root plus the OH capture directory) for
parse + complete-frame pins. The four kit poses join WalkTraceConformanceTests'
still-fixture coverage as NEW rows (the OH cathedral-arrival root is f4180108,
not FW0's f4180106): terrace-edge/cathedral-arrival extend the existing theory
(now (root, fixture) parameterized); holtburg-doorway-still and foundry-deep get
dedicated tests mirroring their FW0 siblings' own structure. Finding: the OH
foundry-deep capture's own retail transcript draws 12 real town buildings through
its exit chain (unlike the FW0 capture, which apparently reached none at that
pose) — the FW0 test's stub 1x1 landscape undershoots it (first divergence:
nothing after "LS" vs retail's real BLD content); fixed by building the full
landscape/building world via WalkLandscapeDatBuilder.Build, matching the shared
theory's own approach, not by skipping or weakening the row.
WalkTraceReplayContext.Signature(WalkOracleFrame) now filters to the DI/DC/BLD/LS
kinds (LC/SC/EC/OC never had a WalkEvent analogue in RetailFrameWalk's own
four-kind vocabulary) instead of mapping them to a "?" placeholder, so the still-
fixture comparison stays apples-to-apples on both sides.
B4 — WalkTranscriptSignatureDiff (test-side only, no runner/tool): diffs two
transcripts (raw lines or parsed frames) at the full DI/DC/BLD/LS/LC/SC/EC/OC
level, reporting the first divergent event and position per frame. Proven over a
synthetic self-vs-self-minus-one-event pair (SignatureDiff_ReportsTheExactRemovedEvent).
Tests: T1 (flag off) is split into a unit-level zero-allocation/zero-output check
on WalkTranscriptDump itself (the walk's pre-existing allocation, e.g.
RetailFrameWalk.EmitDrawCells's per-call array, is untouched by this chunk and not
independently zero-alloc) and an integration-level Collect() check; both assert via
WalkOracleTrace.Parse returning zero frames rather than raw string equality, which
is robust to unrelated Console.WriteLine noise from other test classes running in
parallel (a real, observed hazard — WalkFrameDriverTests joins
CameraDiagnosticsCollection for the same reason CornerFloodReplayTests/
Issue181WallPressEquilibriumTests already do, issue #251). T2 proves the
emitter/parser round trip on a synthetic interior frame. T3's InstalledDat rows all
pass. T4: LaunchOptionsDocumentationTests green with the ACDREAM_DUMP_WALK_TRANSCRIPT
row (both directions).
Gates: hermetic lane 6,814/0 (was 6,795 baseline + new tests), three consecutive
clean runs; InstalledDat lane 245/3 known-failures (the two pre-existing #383
layout tests + TowerAscent) unchanged from baseline.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Render::deg_mul is auto-tuned by frame load and SWUNG between oracle
captures: doorway-still ran right after the heavy terrace-edge capture
with the multiplier depressed (mul <= 0 puts thresholds at/below ideal,
so 001e/0026/002f select the portless level 1 - retail's zero look-in
floods, reproduced exactly at mul = 0), while every other fixture pins
~ +0.99 (thresholds at max). The recon session's "-0.99" live dump was
real - taken under the same cdb load. RetailFrameWalk now exposes the
multiplier; the doorway test pins 0, the rest use the default.
foundry-entry: frames 1-66 reproduce exactly; the F67-F79 standing
segment diverges ONLY in building 0036's intra-building DC order
(retail 116,118,11d vs replay 11d,116,118). The BSP-traversal microscope
pins the flip to 0036's ROOT plane (N=(0,0,-1) D=2.8): replay eye z
2.33 (d=+0.47, NEG-first) vs retail behaving as d<0 (POS-first) - a
structural ~0.5 m frame question (positionPush(2)/part-scale), only
adjudicable live. Turnkey probe:
tools/walk-oracle/fw1-f67-viewpoint-probe.cdb.
Fixture status: nine of ten fully conformant; foundry-entry exact
through F66 with the 13-frame order segment parked on the probe.
Suites: Walk 127/1 skip; hermetic 6,687/0.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The moving-fixture divergence was the building degrade ladder:
GfxObjDegradeInfo::get_degrade @0x0051e4b0 (Ghidra-verified - BN's
FPU-flag pseudo-C misread BOTH arm selection and one formula) slides
each level's threshold from ideal toward MAX as the multiplier
approaches 1, and the live client runs the positive arm at ~0.99, so
level 0's portal-bearing BSP survives to ~max_dist (48 for the Holtburg
cottages), not ideal (24). The recon note's "deg_mul = -0.99" was a
sign misread; the negative arm's threshold slides toward MIN and
contradicts the fixtures from both directions. With the two-arm port,
holtburg-walkout, holtburg-transitions, and holtburg-walkabout pass
every pairable frame - the walkout-F2 microscope's prediction
(103,100 | 100x3 | 124 through the cottage exit views) landed exactly.
Also this round, falsified and reverted: a BN-driven swap of the
portal walker's negative/in-plane arms (Ghidra shows side 1 = negative
EMITS, side 2 = in-plane does not - the original port was correct; the
swap broke four fixtures). The walker docs and unit tests now pin the
Ghidra-verified truth table.
Parked with findings: foundry-entry F67 (right flood set, one
plane-side classification at the +/-eps boundary orders 11d before
116/118) and doorway-still (retail shows zero floods at a pose one
meter from walkout-F2's flooding pose; multi-portal clip boundary).
Suites: Walk 124/3 skips; hermetic 6,687/0.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The moving driver gains adjacent-pose tolerance (the marker-lag capture
artifact) and stays parked: the four moving fixtures diverge at
punch-edge frames under BOTH adjacent poses. FALSIFIED this round and
reverted: the raw-decode + flipped-gates convention (it broke three
still fixtures - the inverted-decode convention stands, six still
fixtures frame-exact). Remaining instruments in the Skip note: znear,
per-view punch ordering inside DrawMesh, exit-view precision.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Three final pins complete the still-fixture set: (1) the CELL portal
side decode is the INVERSE of the 0x2 bit (uniform with the building
convention; the doorway-still flood proved it - ov=2 n=3 exact, and
foundry-deep stays green); (2) interior frames key the landscape order
off the OUTSIDE-projected landcell (get_outside_cell_id - derived from
the camera origin); (3) the outdoor pview has draw_landscape=FALSE so
look-in floods discard exit portals - the ov=0 pattern of every traced
look-in. CONFORMANT: foundry-deep (every frame), doorway-still,
street-outdoor, terrace-center, terrace-edge (the #456 acceptance
pose), cathedral-arrival - full frames identical to retail. The moving
four diverge only at punch-edge frames (walkabout F9, foundry-entry
F67) - pose-timing sensitivity parked in the driver Skip note.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The moving-fixture sweep pinned the pose decode (quat storage w,x,y,z;
facing = rotated +X; the consistent triple right=-rot(Y), fwd=rot(X),
up=rot(Z) - mean 20.8 deg vs motion, alternatives >=52). With it the
street-outdoor BLD roster and order match retail EXACTLY except one
ring-1 boundary pair (aab50002 extra / a9b3003c missing). Look-in
punches fire at the wrong buildings under BOTH GfxObj plane-winding
conventions (retail punched only 001a and 0022 into the a9b4016x
cells) - the next adjudication targets the PortalRef.PortalIndex join
and the BuildingPortal side-flag decode via a per-building dump. The
driving diff test carries the state in its Skip note; the flip toggle
stays for the next arm.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
WalkLandscapeDatBuilder assembles the replay grid with the LIVE-dumped
retail landscape model (LScape mid_radius=25 -> 51x51; the resolution
pyramid observed on the capture client: side_cell_count 8 in the 3x3
core, 4 at ring 2, 2 at rings 3-4, 1 beyond; buildings attach only to
full-res blocks - matching the traces roster), retail z-slabs
(heightTable[max]+200 / [min]-1 per CLandBlock unpack @0052f297), and
camera-block-local coordinates. The replay context gains the building
half (placements, active-view install, building-polygon clip). The
street-outdoor conformance diff now reaches real adjudication: the walk
over-culls six traced buildings (jagged boundary - not a clean frustum
edge) and the look-in floods differ; the driving test is parked Skip
while the visibility map is instrumented.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The replay harness reconstructs the camera from a pose-stamped oracle
frame (Frame quaternion w,x,y,z storage order; +Y forward / +Z up;
landblock-local origin) and drives the ported walk over adapter-built
cells (cell transforms from EnvCell.Position now populated). The
foundry-deep fixture - the pure-interior frame shape - reproduces
retail EXACTLY on every complete frame: DI + DC(ov=0, [cell]) with no
landscape, 39/39. Conventions are now pinned by live retail output; the
remaining nine fixtures need the outdoor world build-out (landscape
blocks, terrain z-slabs, building transforms + active-view clip) and
join the same gate.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Erik