Bite-sized TDD plan for the unification spec. Phase 1 (outdoor node) + Phase 2
(outdoor-root flood) are additive + unit-tested (full code/tests in the plan).
Phase 3 is the single visual-gated cutover (wire one path, repoint exit portals,
delete the branch/BuildFromExterior/DrawPortal/OutsideView). Phase 4 cleanup.
Pure-outdoor regression guard keeps open-world rendering byte-identical.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Brainstormed design to collapse acdream's two render paths (OutdoorRoot vs
RetailPViewInside) into one, matching retail SmartBox::RenderNormalMode ->
DrawInside(viewer_cell). Roots the FLAP as the two-branch split toggling on the
viewer cell crossing the indoor/outdoor boundary (pinned 2026-06-07 via live
render-sig); the 2026-06-05 viewer-cell-stability plan (boom + dead-zone + w-clip)
is exhausted. Models the outdoor world as a flood-graph cell node whose shell is
the landscape, so one flood + one draw handle indoor and outdoor uniformly.
Clean cutover, 4-phase plan (phases 1-2 additive, phase 3 the visual-gated cutover).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Checkpoint of the unified retail-faithful indoor render. The two-week HANG/grey is fixed and the
interior seals (live-verified by the user). Commits the session render-rewrite foundation together
with the fixes that made it functional.
- HANG fix: PortalVisibilityBuilder.Build portal flood did not terminate (the faithful ProjectToClip
near-side clip drifts per round, defeating the CellView dedup; the BFS had no bound after U.2a removed
MaxReprocessPerCell). Fix = drift-tolerant snapped/canonical CellView.Add dedup (PortalView.cs) plus
restored MaxReprocessPerCell=16 bounded re-enqueue (PortalVisibilityBuilder.cs). Re-enqueue is kept
(load-bearing for late-slice propagation, Build_ViewGrowthAfterDoneCell_PropagatesNewSlicesToExit);
only its count is capped. CellViewDedupTests added.
- Seal (DrawCells Task 2): RetailPViewRenderer.DrawEnvCellShells draws EVERY visible cell via
IndoorDrawPlan.ShellPass (was gated on the ClipFrameAssembler slot filter, leaving slot-less cells grey).
- Look-in FPS: GameWindow exterior look-in candidates limited to the player landblock +-1 (was all ~81
loaded LBs iterated every outdoor frame). No behaviour change (far cells were >48m, already culled).
Remaining dominant issue = the FLAP at transitions: viewer-cell metastability (render roots at the
camera-eye cell, which oscillates outdoor-indoor as the 3rd-person boom drifts across the doorway,
confirmed in render-sig). SEPARATE fix, NOT the DrawCells port. Full handoff + flap fix plan + tracked
follow-ups (#78 terrain, look-in-from-inside, look-in FPS, L-spotlight):
docs/research/2026-06-07-indoor-render-session-handoff.md.
Baselines: build 0 err; App.Tests 210/210; Core.Tests 1331 pass / 4 fail (pre-existing) / 1 skip.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Pure port of retail DrawCells membership: reverse cell_draw_list, per-slice. Pins the
grey regression — a cell in OrderedVisibleCells is never dropped from the shell pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Task-by-task plan (TDD pin for the grey regression + per-task visual gates) to replace the
indoor-render approximation with a verbatim PView::DrawCells port, sequenced so Task 2 alone
should kill the grey. Pickup handoff for a fresh session: state, baselines, rules, do-not-relitigate.
Local commit only (not pushed).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Design for replacing the indoor render approximation layer with a verbatim
port of retail PView::DrawCells (0x5a4840). Locates the grey/bleed in the
ClipFrameAssembler slot-pool + drawableCells filter (RetailPViewRenderer.cs:52/237):
visible cells without a clip-slot are dropped (grey) and the per-cell trim was
globally disabled (bleed). Plan: draw EVERY OrderedVisibleCells cell, trim shells
per-slice via ClipPlaneSet gl_ClipDistance, draw objects membership+depth gated
(no hard clip → no half-character). Scope A+B (DrawInside + look-in DrawPortal);
keeps the faithful PortalVisibilityBuilder + ProjectToClip/ClipToRegion ported
this session. Local commit only (not pushed).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The user's primary symptom (interior walls + cellar floor render as grey clear-color
with dynamic objects / outdoor slices showing through; flicker at room/cellar
transitions) is the KNOWN R1-completion problem: the PView flood roots at the CAMERA
cell (viewer cell), and when the camera is in a different interior cell than the player
(room 0171 vs cellar 0174), the flood does not seal the player's cell. Decisive
evidence: flap-cam root=0xA9B40171 playerCell=0xA9B40174. This handoff separates the
two problems I conflated, lists the disproven causes, gives the next diagnostic step
(shell/flap/vis probes in the cellar), and a kickoff prompt. HEAD 2b7f5a1.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Retail SmartBox::RenderNormalMode (0x453aa0:92665) branches DrawInside vs the
outdoor LScape::draw on is_player_outside (the PLAYER's cell, 0x451e80), then
roots DrawInside at the VIEWER cell. acdream keyed the whole branch off the
camera cell (clipRoot = visibility.CameraCell), so a 3rd-person chase camera
lagging in a doorway AFTER the player stepped outside took the DrawInside path
rooted at the threshold cell, where the exit-portal flood degenerates: grey
world + entities-through-walls. Now ShouldRenderIndoor(playerCellId,
viewerCellResolved) gates the branch on the player; the DrawInside root stays
the viewer cell (handoff invariant preserved).
SCOPE / HONESTY: this REDUCES the player-OUTSIDE doorway grey (visual-confirmed
reduced a lot) but does NOT fix the deeper symptom: when the player is in one
interior cell (cellar 0174) and the camera is in another (room 0171), the flood
roots at the camera cell and does NOT seal the player's cell, so the cellar
floor / interior walls drop to grey. That is the KNOWN R1-completion problem
(2026-06-05 Residual A handoff + 2026-06-02 design doc section 3: a
SHELL-SEALING / wrong-flood-root bug), not this branch.
Tests: Core 1331p / 4f (documented) / 1s, App 187p, build green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Port retail CameraManager::UpdateCamera's convergence snap (0x00456fcd):
once the per-frame lerp step is below 0.0004 m AND the rotation within
0.000199999995, freeze the damped eye at an exact fixed point instead of
Vector3.Lerp's endless sub-mm asymptote. The drift was walking the 3rd-person
eye across the vestibule/room portal plane at rest, flipping the per-frame
viewer-cell resolve 0170<->0171 -> the indoor grey/texture flicker. The
collided-eye firewall (separate publishedEye local) is already present.
Adds ApplyConvergenceSnap static (TDD: 3 unit tests + 1 integration freeze
test) + SnapEpsilon/RotCloseEpsilon. App suite 183 -> 187, all green.
Plan: docs/superpowers/plans/2026-06-05-indoor-viewer-cell-flicker-fix.md
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Diagnosis session: the indoor bluish void + grey/texture flicker is visibility metastability at cell boundaries, not a missing flood (R1's per-cell DrawInside is built; the cellar seals). Confirmed by named-retail decomp AND a live cdb capture of retail (viewer_cell rock-stable: clean monotonic transitions, zero oscillation across 4916 samples). Retail stays stable via boom stability + a 0.2mm viewer-cell dead-zone + clip-space portal clipping; acdream diverges on all three. Handoff documents the root cause, the cdb evidence, and the prioritized 3-part retail-faithful fix (boom stability -> dead-zone -> w-space clip) with decomp anchors + a planning/implementation kickoff prompt. Adds the reusable retail viewer-cell cdb capture script and the superseding CLAUDE.md banner.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
When the chase camera roots in a thin doorway cell and the eye stands in an interior portal opening (live capture: vestibule->room portal D=0.16m, proj=0), the 2D projection degenerates and the neighbour was culled (cells=1) -> only the thin cell drew -> bluish void / transparent ceiling. Retail's 3D clip imposes no constraint for a portal the eye is inside, so the neighbour is fully visible. When the clipped region is empty but the eye stands in the opening (EyeInsidePortalOpening: within 0.5m of the portal plane AND point-in-opening), flood the neighbour with the current view. Guarded so an off-screen degenerate portal stays culled (no #95 blowup; over-include is mesh-frustum-culled at draw). Visual-verified: cellar ceiling now solid.
Band-aid for thin-cell-root coverage; likely superseded by the boom-stability + viewer-cell dead-zone + w=0 near-plane clip fix next session (reassess / maybe revert). 2 RED->GREEN tests; cyclic/hub termination guards unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ProjectToNdc clipped only the eye half-space (w>MinW, a 2026-06-03 workaround) and left the 4 frustum side planes to the 2D ScreenPolygonClip. When the eye is within a portal's near plane, small-w verts explode under the perspective divide (probe saw NDC (10.2,-67.4)); the 2D clip then collapses to empty -> OutsideView empty -> terrain Skip -> the bluish doorway void. Clip the eye plane + 4 side planes (homogeneous Sutherland-Hodgman) before the divide so NDC is bounded to the screen by construction, matching retail GetClip -> polyClipFinish (clip in clip-space before the divide; pc:432344).
Partial: NOT the full flicker fix. The dominant cause (camera boom drift + viewer-cell flip at boundaries + missing w=0 near-plane clip) is identified and deferred to the next session per the handoff. 2 RED->GREEN tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Residual A (camera collision = verbatim SmartBox::update_viewer) is SHIPPED +
user-kept (0ffc3f5/5177b54/9e70031). Wrap it and hand off to the render session:
- New canonical handoff (docs/research/2026-06-05-render-residual-a-shipped-core-
inside-render-handoff.md): what A shipped, what A EXPOSED (the render roots at the
viewer cell — clipRoot=CameraCell, GameWindow.cs:7322 — and A made that cell
accurate, so the PVS flood from the viewer cell doesn't reach the player's cell →
cellar floor drops), the reframing (the user's "step C" = the CORE inside render /
R1 completion, NOT R2 outside-looking-in), the evidence-first job, KEEP/DON'T, the
kickoff prompt.
- CLAUDE.md banner: A SHIPPED; next = core inside render (R1 completion).
- Render redesign spec: 2026-06-05 sync note (A shipped; R1 is actually incomplete —
the bleed + cellar-floor drop are the unfinished flood/seal; next is R1, not R2).
The visible problems (bleed + the floor A exposed) are the same family: the inside
path still draws the whole outdoor world instead of retail's "inside → DrawInside
only". A faithful DrawInside seals them by construction (render spec 2026-06-02 §2).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Complete Render Residual A's faithful port: PhysicsCameraCollisionProbe.SweepEye
now mirrors SmartBox::update_viewer (acclient_2013_pseudo_c.txt:92761) end-to-end:
- Start cell (pc:92824-92844): indoor (>=0x100) seats the sweep at the head-PIVOT
via PhysicsEngine.AdjustPosition (the cellar-lip case — feet in the low connector,
head up at floor level); outdoor keeps the player cell.
- Sweep pivot -> sought-eye from the seated start cell (unchanged 0x5c viewer flags).
- Success (pc:92870): set_viewer(curr_pos), viewer_cell = curr_cell.
- Fallback 1 (pc:92878): AdjustPosition(sought_eye).
- Fallback 2 / no-cell (pc:92775, 92886): snap to player, viewer_cell = null. This
also makes cellId==0 faithful (was returning the desired eye; retail snaps to
player_pos) and adds the playerPos arg to ICameraCollisionProbe.SweepEye.
Supporting: ResolveResult.Ok surfaces FindTransitionalPosition's return (retail
find_valid_position != 0, pc:273898) so SweepEye knows when to fall back.
TDD: 11 new tests (FindVisibleChildCell 4, AdjustPosition 3, ResolveResult.Ok 2,
SweepEye orchestration 2). The seating test's RED proved the sweep does NOT auto-
advance feet->room, so the pivot-seated start cell is genuinely decisive. Core
1326 pass / 4 documented-fail / 1 skip; App 179 pass / 0 fail. No regression.
Per the live-capture finding, the visible payoff is the cellar-corner (point 3);
the cottage-room bluish void stays for residual C. Spec:
docs/superpowers/specs/2026-06-05-residual-a-camera-collision-design.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The two Core physics primitives retail's SmartBox::update_viewer calls down into,
ported verbatim (TDD, 7 new tests):
- CellTransit.FindVisibleChildCell (CEnvCell::find_visible_child_cell, pc:311397):
return the cell whose cell-BSP point_in_cell contains a world point — start cell
first, then (stab-list mode) the start's VisibleCellIds or (portal mode) its
direct portals. Sibling of FindCellList. Mirrors FindCellList's null-CellBSP skip
(CellTransit.cs:518) so a cell lacking hydrated CellBSP doesn't spuriously claim
every point via PointInsideCellBsp's null-node "inside" default.
- PhysicsEngine.AdjustPosition (CPhysicsObj::AdjustPosition, pc:280009): resolve a
point's cell from a seed. Indoor (>=0x100) → FindVisibleChildCell(stab-list);
outdoor → landcell snap (same grid lookup as ResolveCellId). The seen_outside
sub-fallback is deferred (off the cottage/cellar path; spec §6).
Both are unwired into any production path — they land the machinery update_viewer's
start-cell + fallback 1 need (and that residual C also needs). The App SweepEye
orchestration that calls them lands next.
Decomp-faithful per the live-capture finding: A's V1 sweep already contains the eye
(eyeInRoot=Y 99.75%, never void); this completes A as a verbatim port. Spec:
docs/superpowers/specs/2026-06-05-residual-a-camera-collision-design.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Live ACDREAM_PROBE_FLAP capture (Holtburg cottage/cellar) proved the V1 camera
spring-arm already contains the eye (eyeInRoot=Y 99.75%, viewerCell never 0,
indoor collide 97.6% in 0174). The dominant inside-cottage bluish void is the
render-sealing residual C (DrawPortal), NOT the camera.
This spec scopes the FAITHFUL completion of Residual A: port the two missing
update_viewer pieces verbatim — the indoor start-cell seated at the pivot via
CPhysicsObj::AdjustPosition (pc:280009) → CEnvCell::find_visible_child_cell
(pc:311397), plus the two AdjustPosition/snap-to-player fallbacks — and land
FindVisibleChildCell (which residual C also needs).
Faithful layering (mirrors retail SmartBox→CPhysicsObj): primitives in Core
(PhysicsEngine.AdjustPosition + CellTransit.FindVisibleChildCell + ResolveResult.Ok),
orchestration in App PhysicsCameraCollisionProbe.SweepEye. Deterministic crux test
(start-cell resolution) in Core.Tests with the cottage fixtures; SweepEye glue in
App.Tests. Visible payoff is narrow (the cellar-corner, point 3); the cottage-room
void stays for residual C.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Session wrap: cellar-lip wedge fixed + visual-verified (cc4590f/9fdf6a5/41db027).
Next task per the plan = Render Residual A: keep the chase camera eye inside the
player's cell by porting retail SmartBox::update_viewer verbatim (fixes interior
walls going grey/transparent from inside).
- New canonical handoff with copy-paste fresh-session kickoff prompt, the retail
update_viewer decode, the V1 current-state map, the gap to pin (faithful
start-cell + AdjustPosition fallbacks + the no-wall-hit cause), and the
evidence-first plan ([flap-sweep] capture → deterministic SweepEye test → port).
- Key finding recorded: find_valid_position (pc:273890) just calls
find_transitional_position — the sweep function is faithful, NOT the divergence.
- CLAUDE.md banner updated to point at the new state + handoff.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The stale-footCenter fix (cc4590f) is visually confirmed: cellar ascent is
smooth, inn door still blocks, generic step-up still climbs. The residual
9/29 (0,-1,0)-sliding-normal records did NOT manifest in live play —
confirming they were buggy-trajectory artifacts.
Remove the temporary investigation scaffolding added for this trace:
- [fc-dispatch] probe in BSPQuery.FindCollisions
- [step-sphere-down] probe in BSPQuery.StepSphereDown
- CellarLipWedgeTests.Diagnostic_TraceRecordByIndex [Theory]
Kept: the fix, the Fix_StaleFootCenter_* regression guards, and the
DocumentsResidualWedge_* documents-the-bug test. Core suite 1317 pass /
4 fail (documented baseline) / 1 skip.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Root cause of the "blocked at the last cellar step" wedge (the primary,
ramp-climb family — 20/29 captured records). The prior session's pinned
"find_walkable is never called during the step-down" was a probe artifact:
a fresh [fc-dispatch]/[step-sphere-down] trace proves Path-3 StepSphereDown
IS reached for both the carried cell and the iterated other-cell.
The real divergence is in Transition.CheckOtherCells. Retail's
check_other_cells (acclient_2013_pseudo_c.txt:272735 → (*cell+0x88)(this))
re-collides the OTHER cells against the LIVE sphere_path.global_sphere — the
position AFTER the primary insert_into_cell ran. The primary collide can MOVE
the sphere: a Path-5 full-hit dispatches step_sphere_up, and a successful
step-up climbs the foot onto the cottage floor yet still returns OK. acdream
instead reused a footCenter snapshot captured BEFORE the primary collide, so
once the lip-riser step-up climbed the foot onto the floor, check_other_cells
still queried 0171 at the pre-climb (sunk ~0.25 m below the floor) position →
the foot spuriously near-missed the very floor it had climbed onto →
neg_step_up → a doomed second step_up vs the floor normal (0,0,1) whose
step_up_slide unwound the climb → validate_transition reverted → 0% advance.
Fix: re-read footCenter = sp.GlobalSphere[0].Origin at the top of
RunCheckOtherCellsAndAdvance (one line). Pre-fix 0/29 wedge records advanced;
post-fix 20/29 climb onto Z≈94.
No regression: full Core suite 1321 pass / 4 fail (the documented baseline:
Apparatus_Grounded_50cmOffCenter, 2× DoorBugTrajectoryReplay LiveCompare_*,
BSPStepUpTests.D4) / 1 skip. The 2 door LiveCompare divergences are
byte-identical with/without the fix (the door's step_up FAILS → sphere
restored → position unchanged → footCenter == live).
Tests: CellarLipWedgeTests.Fix_StaleFootCenter_RampRecordClimbsCottageFloor +
Fix_StaleFootCenter_MajorityOfWedgeRecordsAdvance (new, GREEN).
DocumentsResidualWedge_LiveFloorCp_SlidingNormalKillsPlusY documents the
remaining 9/29 (0,-1,0)-sliding-normal +Y-kill family (slide territory,
deferred to the visual gate).
Apparatus retained (gated on ACDREAM_PROBE_INDOOR_BSP): [fc-dispatch] in
BSPQuery.FindCollisions + [step-sphere-down] in BSPQuery.StepSphereDown +
CellarLipWedgeTests.Diagnostic_TraceRecordByIndex — strip once the residual
is resolved.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P2 / M1.5 "blocked at the last step" cellar-lip wedge. This session built a faithful
deterministic reproduction and peeled the cause through six evidence-disproven framings
to one bounded question. NO fix landed — the last layers were each disproven by evidence,
and guessing at the load-bearing collision code is the saga's failure mode.
Apparatus:
- CellarLipWedgeTests.cs + Fixtures/cellar-lip/ (3 real cell dumps + wedge-records.jsonl =
29 captured ACDREAM_CAPTURE_RESOLVE wedge calls). Replays the exact calls + body-before
through the lip-cell engine: all 29 reproduce at 0% advance in <200 ms. Tests are
documents-the-bug / diagnostics (GREEN while the wedge exists).
- TEMP probes ([path5-wall]/[fw-enter]/[find-walkable] in BSPQuery; [neg-poly]/[stepsphereup]/
[stepdown-decide]/CheckOtherCells cn/sn/negHit in TransitionTypes), gated on
ACDREAM_PROBE_INDOOR_BSP, marked STRIP. TransitionTypes neg-poly shortcut has a reverted-fix
comment (slide attempt didn't clear the wedge).
- tools/cdb/retail-*-trace.cdb (retail cdb traces).
Findings (handoff: docs/research/2026-06-04-p2-cellar-lip-flatfloor-cp-handoff.md, see the
"NEXT-SESSION KICKOFF" at top):
- Flat-floor contact plane is retail-faithful (v1 trace, full-file correlation). NOT the bug.
- PosHitsSphere cull sign is retail-faithful (cdb -z verified; the Binary Ninja `test ah,N; jp`
parity-jump reads inverted — caught + reverted a wrong fix from that mis-read).
- Sphere radius correct (0.48 player / 0.30 camera probe).
- Retail connector cell 0xA9B40175 never blocks (CEnvCell::find_collisions trace: 0 Collided/Slid).
- PINNED: during the step-up's step-down, BSPQuery.FindWalkableInternal is never called for cell
0171, so the cottage floor (poly 0x0023, Z=94) is never tested as walkable -> no contact plane
-> step-up fails -> StepUpSlide=Collided -> wedge.
Next: trace FindEnvCollisions -> FindCollisions path dispatch for 0171 during StepDown=true (why
StepSphereDown/find_walkable is skipped), port retail, validate via CellarLipWedgeTests, regress
DoorBugTrajectoryReplayTests + visual gate.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Appends the copy-paste kickoff prompt for the next session: pursue the principled
P1 fix for the cellar-lip cell-resolver ping-pong (demote ResolveCellId / make the
swept curr_cell the per-frame membership authority), NOT a stickiness band-aid.
Captures the evidence, apparatus, retail anchors, do-not list, and test baseline.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Instrumented acdream at the cellar lip (ACDREAM_DUMP_STEPUP=1): step-up WORKS
(518 attempts, 220 SUCCESS landing the candidate on the cottage floor Z=94.0,
matching retail's landings), but the committed CurPos never advances -- every
success is reverted, and [cell-transit] shows ResolveCellId ping-ponging every
tick at the 3-cell junction (0xA9B40175<->0174<->0171, reason=resolver). So the
wedge is a MEMBERSHIP cell-resolution instability reverting a working step-up --
NOT a collision/step-up bug, NOT edge-slide.
Notably this contradicts the master-plan P1 claim that ResolveCellId was demoted
out of the per-frame path: it is STILL driving per-frame cell changes here and is
unstable. Fix direction = the parked, approval-gated (a) ResolveCellId
demotion/stickiness (membership), now justified as a real bug by live evidence.
Collision-side fixes (abbd761 B1, 0935a31 slide_sphere) are correct + kept.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Live retail cdb trace (tools/cdb/cellar-corner-escape.cdb) of the Holtburg
cottage cellar-top corner decodes the ground truth: retail escapes by
step_sphere_up->step_up (196x vs 38 near-misses), transitioning the contact
plane from the ramp (N.z=0.78) onto the flat cottage floor (N.z=1.0, 76
landings). acdream slides at the lip and never makes that ramp->floor
transition -> the intermittent cellar wedge.
So the remaining cellar bug is the #98-core step-up-onto-cottage-floor
(DoStepDown / step_sphere_down / find_walkable), which the shipped B1 (abbd761)
+ slide_sphere (0935a31) fixes got close to but didn't finish. Door still
blocks; generic step-up climbs; cellar went always-stuck -> works-mostly.
Next (handoff doc): instrument acdream's OWN corner path (does step_up fire at
the lip and fail to land on the cottage floor?) before porting the lip-climb --
no guessing (#98 saga rule).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replace the A6.P4 Collided shortcut in transitional_insert's neg_step_up==0
branch with the faithful CSphere::slide_sphere. Retail
(acclient_2013_pseudo_c.txt:273350-273351, CTransition::transitional_insert):
neg_step_up==0 (HEAD-sphere near-miss) -> slide_sphere -> continue the insert
loop; neg_step_up==1 (foot) -> step_up_slide. The acdream foot branch already
did that; only the head branch took the shortcut (SetCollisionNormal + return
Collided = dead hard-stop). The slide itself is the existing
SlideSphereInternal (Sphere.SlideSphere port): it strips the into-wall
component and keeps the tangential crease (collisionNormal x contactPlane.N).
Surfaced by the B1 near-miss-gate fix (abbd761): once the grounded mover climbs
onto the cottage floor, its head sphere brushes the cellar stairwell walls and
the old hard-stop wedged it (2026-06-04 live capture: 274 (0,-1,0) + 78
(1,0,0) hits, out==current, dead oscillation). Post-fix capture shows 96
hit-and-advanced frames (the body slides along the walls).
Visual-verified 2026-06-04: closed cottage door still BLOCKS (no walkthrough --
drifts sideways along it, retail-faithful); cellar ascent now works (was always
stuck). An intermittent corner-wedge (slide into the -Y/+X wall corner) remains
-- separate finer issue, under investigation.
Core 1310 pass / 4 fail (pre-existing: 3 door documents-the-bug + D4 airborne).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The prior localization (step-up CLIMB) was disproven by an ITestOutputHelper
capture. Records the real root cause (A6.P4 near-miss missing retail's
num_sphere>1 gate, fixed in abbd761), that the door blocks faithfully with a
real floor, and that the remaining red tests are separate (apparatus
synthetic-floor artifact, LiveCompare buggy-captures, D4 airborne) — not
simple "flip to green" targets. Next is the user visual gate.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The 2026-06-03 handoff localized the failing Core tests to the BSP Path 5
step-up CLIMB (find_walkable/step_sphere_down). An ITestOutputHelper capture
of B1 disproved that: the climb code is correct (matches ACE
Polygon.adjust_sphere_to_plane / BSPTree.step_sphere_down exactly). The real
bug is the A6.P4 near-miss dispatch in FindCollisions' Path 5 (Contact
branch), which diverged from retail three ways:
1. Recorded a near-miss NegPolyHit UNCONDITIONALLY. Retail gates both
set_neg_poly_hit calls behind `if (num_sphere > 1)`
(acclient_2013_pseudo_c.txt:323852).
2. Checked the foot sphere's near-miss before the head's. Retail checks
the head (sphere1) first.
3. Mapped foot->neg_step_up=false / head->true. Retail maps head(index 0)
->false (slide), foot(index 1)->true (step-up), per
SPHEREPATH::set_neg_poly_hit (:323279, neg_step_up = arg2).
For B1's single foot sphere, the spurious near-miss -> outer loop
`!NegStepUp -> SetCollisionNormal + Collided` -> revert: the grounded mover
wedged at x=0.1 and never advanced to the wall to step up. With the verbatim
gate, a single-sphere near-miss records nothing, the sphere advances,
full-hits the wall, and step_sphere_up climbs the 0.25 m step (verified via
probe capture: foot ends at (0.6, 0, 0.25)).
The Holtburg cottage door still blocks faithfully (door slab (0,-1,0) normal,
stops in front of the door) when the scenario has a real floor — confirmed
this change does not regress the door.
The two BSPQueryTests Path5 near-miss tests used a single sphere (the very
non-retail assumption that caused this wedge); converted to the production
2-sphere shape where the head sphere records the near-miss, matching retail.
Core 1312 pass / 4 fail (the 4 pre-existing: 3 door documents-the-bug + D4
airborne, none regressed here); App 177 green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P1 membership is DONE (proven to already match retail; the 0/11 was a cdb capture artifact;
merged + pushed). P2 root cause localized to BSP Path 5 grounded step-up: the Path 5 wrappers
(DoStepUp=retail step_up, DoStepDown=retail step_down) are verified faithful + reached; the
divergence is in the step-up CLIMB (find_walkable/step_sphere_down up-adjust when sp.StepUp=true).
- docs/research/2026-06-03-p2-door-stepup-handoff.md: canonical P2 pickup + fresh-session prompt +
DO-NOT-RETRY (the wrappers) + the tooling note (xunit swallows Console.WriteLine).
- master-plan §3: P1 marked DONE + the (a)-(d) deletes/unifications re-scoped to approval-gated
refactors of working code; P2 localization recorded.
- CLAUDE.md M1.5: dated 2026-06-03 pointer (P1 done, P2 active, render seam in P3/P4, pickup doc).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Standing/pacing the Agent of Arcanum doorway in the acdream client with ACDREAM_PROBE_CELL=1:
the player [cell-transit] sequence is clean + monotonic and crosses at the same Y thresholds as
retail's aligned golden, firing only on character movement (never camera-only). So P1 membership
is correct LIVE, matching the conformance proof.
The visible flap + purple void are the RENDER half, not membership: the flap is the camera-collision
residual (chase eye drifts out of the player cell -> viewer-cell flips; master-plan P3,
SmartBox::update_viewer), the void is the unported PView seal (master-plan P4). The user's intuition
"transition feels tied to the camera" is retail-faithful: retail keys the render on the viewer
(camera) cell, physics+lighting on the player cell.
Per user direction, P2 door collision is next; the render half (P3 camera -> P4 PView) follows.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The P1 "doorway membership lags retail" premise is FALSIFIED. acdream's swept
ResolveWithTransition already matches retail's true per-frame curr_cell: the
production gate ProductionPath_IndoorCrossings reads 9/9 on the indoor 0170<->0171
crossings with NO code change, once fed an aligned retail golden.
Root cause of the false 0/11: CPhysicsObj::SetPositionInternal calls change_cell
(acclient_2013_pseudo_c.txt:283456) BEFORE set_frame writes m_position (:283458),
so the original golden (find-cell-list-capture.cdb, read at the change_cell BP)
paired each frame's NEW cell with the PREVIOUS frame's position — a one-frame skew.
Verified 3 ways: the decomp ordering; golden_picked[i] == geom(golden_position[i+1])
for all 22 rows; acdream's static pick == golden_picked[i-1] for all rows. Both
retail and acdream pick with center-only point_in_cell on global_sphere[0] (no XY
lead; cache_global_sphere @ pc:274196). curr_cell commits via validate_transition
(@ pc:272608, curr_cell = check_cell) = the find_cell_list pick, structurally
identical to acdream's RunCheckOtherCellsAndAdvance -> FindCellSet -> SetCheckPos.
There was nothing to port; a swept advance would make membership LEAD by a frame.
- tools/cdb/find-cell-list-capture-aligned.cdb: re-capture reads the committed
position from the set_frame that follows change_cell (cell+position same instant).
- Fixtures/find-cell-list-threshold.log: replaced with the aligned capture.
- ThresholdPortalCrossingReplayTests / FindCellListConformanceTests: rewritten from
documents-the-bug to assert retail truth (per-segment / per-indoor-pick equality).
- handoff + notes + README + memory: banners correcting the disproven premise.
Still open (NOT indoor membership, which is DONE): outdoor->indoor 0031<->0170 entry
conformance (needs landcell + building stab in the gate cache); master-plan cleanups
(delete CheckBuildingTransit, unify find_env_collisions, demote ResolveCellId) refactor
working retail-faithful code -> need explicit user approval.
Conformance 60 pass / 1 skip / 0 fail; full Core 1309 pass / 5 fail (pre-existing
2 BSPStepUp + 3 door-collision = P2) / 1 skip.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Single pickup document for the next session: state both altitudes, the 8
session commits, the confirmed finding (production membership diverges 0/11 —
swept move completes but curr_cell never advances across the portal), the
DO-NOT-RETRY list (3 falsified hypotheses), the apparatus inventory + run
commands, the P1 fix scope (two decomp questions + the acdream code map + the
RED gate), the test baseline, and a copy-paste pickup prompt.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replays the golden indoor 0170<->0171 segments through the real
PhysicsEngine.ResolveWithTransition (engine builds the global sphere + sweeps;
cells loaded from dats with real BSP). Result: 0/11 match retail. Every segment
restPos==target (the sweep completes the move) but CellId stays on the SOURCE
cell — acdream moves the body across the doorway yet NEVER advances curr_cell.
So the 'probe artifact' hypothesis is FALSIFIED: production membership genuinely
lags retail.
Refined mechanism: both retail and acdream PICK with center-only point_in_cell
(architect's radius-aware-pick hypothesis falsified, confirmed by reading
CEnvCell::point_in_cell -> BSPTREE::point_inside_cell_bsp). The gap is retail's
curr_cell ADVANCES across the portal mid-sweep (swept crossing / leading sphere
point) while acdream's swept advance keeps the source cell. P1 ports that advance.
ProductionPath_IndoorCrossings_DivergeFromRetail_PendingP1 is the RED gate the P1
fix must turn GREEN. Conformance 60 pass / 1 skip / 0 fail.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Read CEnvCell::point_in_cell @ 0x52c300 -> CCellStruct::point_in_cell @ 0x5338f0
-> BSPTREE::point_inside_cell_bsp: the find_cell_list PICK (pc:308810) is
CENTER-ONLY, at global_sphere[0].center (the swept sphere center), NOT
radius-aware and NOT the foot origin. So acdream's PointInsideCellBsp pick
criterion ALREADY matches retail. The architect's 'use SphereIntersectsCellBsp
in the pick' hypothesis is FALSIFIED. The P0 FindCellList_DoorwayThreshold probe
fed the foot origin (captured m_position) through no sweep -> its 'all 22
diverge' is a PROBE ARTIFACT, not a confirmed production divergence (the data's
own tell: retail commits the cell AHEAD of motion while the foot is behind = the
swept sphere center crossing the portal).
P1's decisive first step is the PRODUCTION-PATH trajectory conformance (replay
the golden through ResolveWithTransition, which uses sp.GlobalSphere + the sweep)
BEFORE designing any fix. Do not port a portal-crossing/radius pick on the probe.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
From reading CEnvCell::find_transit_cells @ pc:309968: P1 is mostly REWIRING
curr_cell advancement (RunCheckOtherCellsAndAdvance/SetCheckPos) to use the
portal-crossing candidate, not FindCellSet's point-in-cell pick. The P1
conformance must replay the golden positions through ResolveWithTransition (a
trajectory incl. outdoor landcell 0031), not bare FindCellList.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P0 Task 6 complete. Captured live retail membership at the 0031<->0170<->0171
doorway via cdb on CPhysicsObj::change_cell (symbol-driven; offsets verified by
discover-types.cdb; PDB MATCH). 22 transitions, clean monotonic sequence, NO
ping-pong (retail is correct-by-construction). Golden:
Conformance/Fixtures/find-cell-list-threshold.log.
ROOT-CAUSE FINDING (the central P1 work): retail transitions membership at the
PORTAL CROSSING (CEnvCell::find_transit_cells @ 0x52c820 pc:309968 — sphere crosses
the doorway polygon plane), while acdream's FindCellList re-picks by POINT-IN-CELL
containment at the foot. Retail commits room 0171 while the foot is STILL inside
vestibule 0170's BSP (in_0171=0); acdream lags. ALL 22 transitions diverge for this
one criterion mismatch — not a per-cell hysteresis or a building-entry-only split.
This is master-plan §0 'hysteresis gap' confirmed against the real client.
FindCellList_DoorwayThreshold_DivergesFromRetail_PendingP1 (documents-the-bug, GREEN)
+ ThresholdDivergenceDiagnosticTests (per-transition containment print) pin it; both
flip when P1 ports the directed portal crossing. Conformance 59 pass / 1 skip / 0 fail;
full Core 1308 pass / 5 fail (baseline) / 1 skip — no new failures.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P0 Tasks 6 (autonomous half) + 7. FindCellList_DoorwayThreshold_MatchesRetailTrace
asserts acdream's pick == each captured retail pick; skips until the capture
fixture lands. PvsConformanceTests scaffolds the render visible-set golden
(skipped; filled in P4). ConformanceDats.FixturesDir resolves fixtures from the
source tree (issue98 pattern). Notes record: existing retail traces are
collision-only (no membership) so the strict P1 gate needs the one live capture;
plus the P1 re-scope finding (Stage-1 membership already on this branch).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P0 Task 5. RetailTrace parses the [fcl] golden format (seed/pos/picked,
RetailCellPick); 4 TDD tests green. find-cell-list-capture.cdb targets
CPhysicsObj::change_cell (commit-on-diff) to capture retail's accepted
membership sequence at the doorway; README is the operator runbook
(dt offset verification + decode_retail_hex float decode). The live run
is P0's one user-gated step (Task 6 mines existing traces first).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P0 Task 4. FindCellList resolves a sphere deep inside room 0171 -> 0171,
deep inside vestibule 0170 -> 0170, and re-picks 0170 from a stale 0171
seed (membership re-picks by containment, not the seed). Retail-faithful
by construction (candidate cells loaded from the real dats). The subtle
doorway-threshold pick is the trace-backed golden (Task 6).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P0 (verbatim-spatial-pipeline-port) Tasks 1+2. ConformanceDats loads the
cottage-doorway cells from the real dats with their real ContainmentBsp;
CottageDoorwayCharacterizationTests maps the Holtburg 0140..017F indoor
neighborhood and pins the master-plan threshold building (origin
161.93,7.50,94.00): 0xA9B40170 vestibule (exit portal 0xFFFF + portal to
0171), 0xA9B40171 room. Grid math confirms the outdoor side is landcell
0xA9B40031 -> the 0031<->0170<->0171 ping-pong is verified real. Verified
interior points recorded for the point_in_cell/find_cell_list goldens.
Plan: docs/superpowers/plans/2026-06-03-p0-conformance-apparatus.md
Notes: docs/research/2026-06-03-p0-conformance-apparatus-notes.md
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The doorway saga (void -> transparent walls -> flaps) proved patching the hybrid is hopeless:
retail does membership + collision + camera + render as ONE coupled pipeline; acdream
reimplemented pieces with mismatched criteria at the seams. Master plan to port ALL of it
verbatim: A membership (find_cell_list/find_transit_cells/find_building_transit_cells intrinsic,
no bridge), B uniform collision (no indoor/outdoor fork) + door collision, C camera
(update_viewer + find_visible_child_cell), D the full PView render (ConstructView/InitCell/
ClipPortals/GetClip/DrawCells/DrawPortal + the update_count watermark). KEEP/REPLACE/DELETE
lists, decomp anchors per function, P0-P6 sequence (apparatus-first, foundation-up, visual gate
each), and the kickoff prompt. Supersedes the render-only redesign's scope.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The cottage doorway 'void' (dark cell + floating entities while the chase camera looks
through the opening): PortalProjection.ProjectToNdc clipped portals on w+z>=0 — the GL
[-1,1] near-plane test — but acdream's camera builds its projection with D3D-convention
Matrix4x4.CreatePerspectiveFieldOfView and a 1.0 m near plane (RetailChaseCamera). Against
that matrix w+z>=0 discards everything within ~0.5 m of the eye, so when the camera orbits
to ~0.1 m from a doorway portal the near edge is clipped, the far edge projects off-screen
([flap] showed p->0171 D=0.10 proj=4 clip=0 ndc Y=-3.5..-6.6), the room behind is culled
(vis=1) and only the tiny vestibule shell draws -> dark void. Rotating away moved the eye
off the portal -> vis=5 -> room rendered.
Fix: clip against the EYE (w > MinW, MinW=0.05 m), near-INDEPENDENT — a portal you're
standing in still projects (covers the screen) so the cell behind stays visible. We only
use the projected x/y for the visibility clip region, so keeping vertices in front of the
near plane is correct. Matches retail PView::GetClip near-clipping the portal before project.
RED->GREEN regression test (doorway 0.1 m from a near=1.0 eye); 177 App tests green; the
existing straddling ±50 bound still holds.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Phase W single-viewpoint V1 (un-split). The render mode decision, indoor root, and portal
side-test now key on the collided-camera viewer cell + eye (RetailChaseCamera.ViewerCellId +
camPos) — retail RenderNormalMode -> DrawInside(viewer_cell) @92675; InitCell side-test vs
viewer.viewpoint @432991. Lighting / seen_outside / playerInsideCell stay on the PLAYER cell
(CurrCell), retail CellManager::ChangePosition @4559B0. The old per-render player-root +
eye-projection split (U.4c) is removed; the flap is avoided by the robust graph-tracked viewer
cell (no AABB, no grace). [flap-cam] probe extended with viewerCell vs playerCell. CurrCell
stays player-only (blue-hole fix intact). App 176 green; Core 1295/5 baseline (no new fails).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Update() now always sets ViewerCellId: the camera-collision sweep's swept cell when collision
is on (retail viewer_cell = sphere_path.curr_cell), else the passed player cell. This is the
robust, per-frame, graph-tracked 'which cell is the camera in?' answer that V1 roots the render
on — no AABB, no grace frames (the U.4c flap source). 176 App tests green (2 new).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The camera spring-arm sweep already resolves the collided eye's cell (ResolveResult.CellId
= sp.CurCellId = retail viewer_cell = sphere_path.curr_cell, update_viewer pc:92871).
Return it from SweepEye so the render can root on the viewer cell (Phase W single-viewpoint
V1, Task 1). Pure plumbing — behavior unchanged; callers extract .Eye. 174 App tests green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The inside/outside render currently splits viewpoints: the player cell roots
visibility + the portal side-test, the eye only projects. Retail uses ONE
viewpoint — the collided camera (viewer) — for the mode decision, indoor root,
side-test, AND projection (RenderNormalMode -> DrawInside(viewer_cell) @92675;
InitCell side-test vs viewer.viewpoint @432991; viewer_cell = sphere_path.curr_cell
@92871). The split makes the render mode follow the player while the screen comes
from the camera -> doorway-straddle void + see-through transition (user evidence
2026-06-03). Spec unifies on the viewer: V1 un-split (robust viewer cell from the
camera sweep, no AABB/grace -> no U.4c flap; lighting stays on the player cell),
V2 DrawPortal (outside-looking-in), V3 floor seal. Supersedes residual-A; merges
A+C. Keeps the blue-hole fix (CurrCell player-only).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Canonical handoff: docs/research/2026-06-03-membership-and-bluehole-shipped-handoff.md
(what shipped: membership Stage 1 ordered-CELLARRAY port + the blue-hole render-root
clobbering fix; the full remaining-issues list — A camera-collision, B R1b particles,
C R2 outside-looking-in, Stage 2 membership, #7 stairs, the 5-test baseline; KEEP/
DON'T-REDO; key files + decomp anchors; copy-paste pickup prompt for next session).
- ISSUES.md: recorded the cottage doorway flap DONE (both causes) in Recently closed.
- render design spec §7: R1 + flap marked DONE; A/B/C mapped to the next render phases.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
THE doorway flap root cause, found via [flap-cam]/[shell]/[cell-transit] (2026-06-03):
the player spawned + stood still in the room (cell 0171, NO [cell-transit] after teleport),
yet the render rooted at the vestibule (0170) for all 77,951 frames — drawing only 0170's
~8-triangle shell, the rest = GL clear color = the bluish void.
CellGraph.CurrCell IS "the player's cell" (the render root), but it was written by
SetCurrAndReturn inside the PER-ENTITY ResolveWithTransition + ResolveCellId — so EVERY NPC
wrote it. A Holtburg NPC (0x000F4240) jump-looping near the doorway clobbered the player's
render root every tick. Standing still (player makes no resolve calls) the NPC's write wins
→ stuck blue void; moving, player/NPC writes fight → the flap. This is why the membership
pick fix (correct, kept) didn't change the visual — the render root was clobbered regardless.
Fix: CurrCell is now written ONLY by the player. New PhysicsEngine.UpdatePlayerCurrCell is
called from PlayerMovementController.UpdateCellId — the single player-only chokepoint for
CellId (teleport / server snap @ SetPosition + per-frame resolver). Removed the CurrCell
write from SetCurrAndReturn (inlined the 2 resolve call sites to sp.CurCellId) and the 4
ResolveCellId sites. NPCs no longer touch the render root. Teleport→UpdateCellId also covers
spawn/standing-still (CurrCell = the player's spawn cell immediately).
CellGraphMembershipTests rewritten to the new contract (3 tests): UpdatePlayerCurrCell writes
the render root; ResolveCellId does NOT (the blue-hole guard); stale-beats-null preserved.
Full Core suite: 1295 pass / 5 fail = the documented §10 baseline, zero new breakage.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The ordered pick alone did not stop the cottage doorway flap: the live [cell-transit]
log showed the cell faithfully following a position that cleanly oscillated between two
values at constant Z — the signature of a bistable membership<->collision feedback loop
(user's §4.4 #3, the forked collision).
Root cause: FindEnvCollisions RE-PICKED the cell from the TARGET position (old line 1958)
BEFORE running the primary collision, so the collision geometry (which cell BSP / terrain)
swapped the instant the pick flipped -> position shifts -> pick flips back.
Retail does NOT do this. CEnvCell::find_env_collisions (acclient_2013_pseudo_c.txt:309573)
collides against the cell it was called ON (sphere_path.check_cell, the carried seed) and
picks the NEW containing cell AFTER, in CTransition::check_other_cells (272717->272761:
check_cell = var_4c). Collide-then-pick.
This commit ports that order:
- remove the pre-pick (production); collide against the carried cell (indoor BSP block /
terrain block unchanged);
- new shared RunCheckOtherCellsAndAdvance() runs the ordered FindCellSet pick +
multi-valued CheckOtherCells + the carried-cell advance AFTER the primary collision, for
BOTH indoor and outdoor seeds;
- the outdoor-seed post-step replaces the removed pre-pick's outdoor->indoor re-entry
promotion (CheckBuildingTransit interior cell + its wall collision on the entry frame).
Cache-null unit-test fallback (ResolveCellId) kept. Full Core suite: 1293 pass / 5 fail =
the documented §10 baseline exactly (2 step-up + 3 door-collision), zero new breakage.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Port CObjCell::find_cell_list (acclient_2013_pseudo_c.txt:308742) faithfully:
- build candidates into an ordered CellArray with the CURRENT cell at index 0
(add_cell @308766);
- EXPAND via a single forward walk over the growing array, mirroring retail's
for(i=0;i<num_cells;i++) cells[i].find_transit_cells loop (308775-308785),
replacing the order-losing Queue/visited BFS;
- PICK in array order with interior-wins-break (308788-308825): current cell at
index 0 wins a boundary straddle, so membership no longer ping-pongs.
Deletes the 5ca2f44 current-first pre-check (the ordered array subsumes it for every
seed). Keeps its guard test (TwoOverlappingCells_CurrentCellWinsTheStraddle) + adds
two conformance tests (current-cell-first ordering; interior-wins over outdoor
fallback). Membership net: 45 pass. Decomp finding: retail stability is emergent from
the ordered pick + carried seed, not a separate portal-crossing detector — see
docs/research/2026-06-03-cell-membership-ordered-cellarray-pseudocode.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
