Resolve the Fix D contradiction with decomp (workflow wf_f660eb88 + adversarial
verify) + 4 live cdb captures. The D3D-FF model was the WRONG oracle: retail has
TWO light systems — STATIC torches BAKE into wall vertices (calc_point_light,
triple-clamped: range gate + per-channel min(scale*color,color) + per-vertex
[0,1] from black), DYNAMIC lights go D3D hardware. The captured intensity=100 is
the purple PORTAL (magenta, dynamic), not a wall torch. Ground truth: 38 static
warm torches (orange (1,0.588,0.314)/cream, intensity=100, falloff 3-5) + 2 dynamic.
acdream over-brightness = two confirmed bugs: D-1 mesh_modern.vert folds
ambient+sun+torches into one UNCLAMPED accumulator (single frag clamp) -> warm
blowout; D-2 EnvCellRenderer never binds SSBO 4/5 so the cell shell reads a leaked
light set. Spec: D-1 in-shader clamp-split (clamp the torch sum on its own before
ambient/sun); D-2 bind the shell's own per-cell light set (mirror WbDrawDispatcher);
LightBake.cs is the C# conformance oracle. Adds the 4 reusable cdb capture scripts.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Session handoff: live-cdb grounding shipped Fix A (point-light shape), Fix B
(per-object selection), Fix C (sun-vector magnitude / ~32% over-bright). Fix D
(outdoor objects too bright near torches) is fully grounded but BLOCKED on one
capture (the building's render path) — the D3D-FF math says it'd make objects
brighter, so not ported. Full cdb cheat-sheet + the contradiction + the next
capture in the doc.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Outdoor lighting was ~32% too bright (washed-out, weak shading). Live cdb on
retail (SmartBox::SetWorldAmbientLight + SkyDesc::GetLighting + LScape::sunlight,
binary matches refs/acclient.pdb) pinned it: at the SAME game time + DayGroup,
acdream's ambient COLOR matched retail exactly (the purple is correct, authored
per-time-of-day in the sky dat) but the LEVEL was 0.607 vs retail's 0.459.
level = AmbBright + 0.2·|sunVec|, both AmbBright=0.40, so acdream's |sunVec|≈1.06
vs retail's ≈0.30. Retail's LScape::sunlight read live = (0.2238, ~0, 0.00352),
magnitude 0.224 = DirBright, y≈0.
RetailSunVector had `y = cos(P)` (≈1) — the raw PRE-transform value SkyDesc::
GetLighting writes to arg5 (0x00500ac9), before LScape::set_sky_position's
world transform. acdream ported the un-transformed vector, so the y=cos(P)≈1
term inflated |sunVec| to ~1.06. That magnitude feeds BOTH the ambient boost
(SkyKeyframe.AmbientColor) AND the sun colour (SkyKeyframe.SunColor =
DirColor×|sunVec|), over-brightening the whole scene (terrain, objects, sky)
~30% and also pointing the sun the wrong way.
Fix: RetailSunVector = DirBright × (cos(P)·sin(H), cos(P)·cos(H), sin(P)) — the
world-space spherical form LScape::sunlight actually holds; |sunVec| == DirBright
for all H/P. After: acdream ambient (0.353,0.176,0.449) vs retail (0.360,0.180,
0.459) — within ~2%, user-confirmed "better outside". Sun direction also corrected
(was pointing ~North from the bad y term).
Tests updated to the cdb-verified values (the prior tests pinned the inflated
magnitude). 18/18 sky tests green. reference-retail-ambient-values memory updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Outdoor objects brightened as the camera approached: lighting selected the
nearest 8 lights to the VIEWER and fed that one global set to everything
(LightManager.Tick), so a building's wall torches only lit it once the camera
got close enough for them to win the global top-8. Probe confirmed the scale of
the problem: a single Holtburg view registers 129 point lights — the global cap
of 8 was hopeless.
Retail selects up to 8 lights PER OBJECT by the object's own position
(minimize_object_lighting 0x0054d480), so a torch always lights the wall it
sits on, camera-independent. Ported faithfully:
- LightManager.SelectForObject (pure, TDD, 8 new tests): candidacy
(light.pos − center)² < (Range + radius)², nearest-8 among those. Plus
BuildPointLightSnapshot for the per-frame stable-indexed light list.
- mesh_modern.vert: two SSBOs — binding=4 GLOBAL point-light array (the
snapshot), binding=5 per-instance light SET (8 int indices into it, -1 =
unused), parallel to the binding=0 instance buffer (mirrors the U.3 clip-slot
mechanism). accumulateLights keeps ambient + sun from the SceneLighting UBO
(cleared as faithful by the lighting audit) and loops THIS instance's point
lights. pointContribution factored out (same calc_point_light wrap+norm shape).
- WbDrawDispatcher: per-entity light set computed ONCE at the isNewEntity site
(constant across the entity's parts), by the entity's AABB sphere; threaded
into grp.LightSets parallel to grp.Matrices; global + per-instance buffers
uploaded in Phase 5. Camera-independent ⇒ stable for static buildings.
- GameWindow: BuildPointLightSnapshot + dispatcher.SetSceneLights each frame.
Tests: 17/17 LightManager + 36/36 dispatcher clip-slot/clip-frame green
(parallel-array lockstep preserved). Visually gated: the meeting hall now holds
steady as the camera approaches (was the popping symptom).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The torch/point-light look was wrong two ways, both now fixed against the
named retail decomp (calc_point_light 0x0059c8b0) via our verified
LightBake.PointContribution port:
1. Per-PIXEL → per-VERTEX. accumulateLights moved from mesh_modern.frag to
mesh_modern.vert so point lights Gouraud-interpolate across each triangle
the way retail's fixed-function T&L does. The per-pixel eval made a tight,
hard-edged "spotlight" pool on flat walls; per-vertex is a soft, broad
gradient. frag now just consumes the interpolated vLit (+ fog + flash).
2. Simplified ramp → faithful calc_point_light shape. The live point/spot
branch was max(0,N·L) × linear(1−d/range) × cap — missing two terms our
LightBake.cs port already has:
• half-Lambert WRAP (1/1.5)·(N·D + 0.5·d), D un-normalised — a face
angled away from a torch still catches light (retail's soft terminator)
instead of snapping to black.
• distance-cube NORM branch norm = distsq>1 ? distsq·d : d — inverse-
square-ish soft far halo + punchy near field, vs the flat linear ramp.
Per-channel no-blowout cap (min(scale·color, color)) retained.
The per-channel cap was also added to the legacy mesh.frag for consistency.
A read-only retail-vs-acdream lighting audit (11-agent workflow) confirmed
these two as the cause of the "better but a bit off" look and cleared the
ambient/sun/terrain/color-space chain as already faithful. Remaining
confirmed divergences (per-object light selection; dungeon static vertex
bake) are filed as the next fixes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- #137: dungeon collision wrong at doors / wall openings (EnvCell collision; needs repro).
- #138: teleport OUT of a dungeon loads the outdoor world incompletely (missing trees/
scenery, broken collision) + a position desync (avatar moves but player position doesn't)
— hypothesised as the dungeon-streaming collapse→EXPAND gap (same machinery as #135).
- #135 marked DONE (user-verified FPS-steady dungeon login); #136 closed (editor-marker hide).
- CLAUDE.md current-state refreshed: #135/#136 shipped, A7 lighting + #137/#138 remaining.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Rewrite the #136 entry with the definitive root cause (editor-only dat placement
marker hidden by retail's distance degrade, inherited as visible from the WB-derived
render path) replacing the earlier refuted texture-pipeline hypothesis; mark FIXED.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The "red cone" (+ green floor petals) in the 0x0007 Town Network dungeon is a dat
EnvCell static object (Setup 0x02000C39 / GfxObj 0x010028CA) using pure red/green
MARKER textures (0x08000109 / 0x0800010A). It is an EDITOR-ONLY placement marker:
its DIDDegrade table 0x11000118 is {slot0 Id=mesh MaxDist=0, slot1 Id=0 MaxDist=FLT_MAX},
i.e. visible ONLY at distance 0 (the WorldBuilder editor origin) and degraded to
GfxObj id 0 (nothing) at any real distance. retail's distance-based degrade
(CPhysicsPart::UpdateViewerDistance 0x0050E030 -> Draw 0x0050D7A0) therefore never
draws it in the live client.
acdream's render pipeline is extracted from WorldBuilder, which (being an editor)
renders every cell static's base mesh directly and has NO degrade handling at all
(zero DIDDegrade references in references/WorldBuilder) — so acdream inherited the
"show the marker" behavior and drew it forever. It only became visible now because
the #135 login-into-dungeon fix drops the player at the exact saved spawn next to it.
Fix: GfxObjDegradeResolver.IsRuntimeHiddenMarker() detects the editor-marker pattern
(HasDIDDegrade + Degrades[0].MaxDist==0 + a degrade entry with Id==0). The EnvCell
static-object hydration (GameWindow ~5793) skips such GfxObjs — whole-stab for bare
GfxObj stabs, per-part for Setup stabs (an all-marker Setup then drops via
meshRefs.Count==0). This is the faithful equivalent of retail's runtime degrade for
static geometry (always viewed at distance > 0); real LOD objects (slot0.MaxDist>0)
and degrade-to-real-mesh objects are untouched.
Diagnosis was extensive (geometry-not-VFX via particle-off; texture-not-lighting via
flat-ambient frame dumps; per-surface runtime decode pinned the red/green marker
surfaces; a draw-time probe pinned the dat-static entity id; a dat dump of the Setup +
degrade table confirmed the editor-marker pattern). Verified live via a frame dump:
the red cone + green petals are gone, all real dungeon decorations still render.
4 new GfxObjDegradeResolver unit tests cover the marker / normal-LOD / no-table /
degrades-to-real-mesh cases.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Investigated the user-reported divergence (a solid-red cone in the 0x0007 dungeon
that retail doesn't draw). Narrowed by elimination:
- geometry, not VFX (survives particles-off)
- object 0x70007055 / Setup 0x020019F0, physState=0x1C — NOT NoDraw/Hidden
- its distinguishing texture 0x06006D65 (DXT1 256x128) DECODES tan/opaque offline,
identical to a neighbour decoration (0x020019EE / tex 0x06006D63) that renders fine
- not a per-instance tint (hook dropped)
=> the red is introduced at runtime in the WB bindless texture-array upload/sampling
path (a #105-class "samples undefined until flushed" / layer-handle misassignment),
possibly lighting. Both WB-render-migration and sky/lighting are FROZEN phases, so the
fix awaits explicit sign-off. Full diagnosis + reusable diagnostic approach in the issue.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two regressions from the pre-collapse (712f17f), found by live gate + a runtime
probe:
1) Login-into-dungeon stopped loading the dungeon. The login-hold streaming
observer fell through to the OFFLINE fly-camera branch once
_lastLivePlayerLandblockId was filtered to the player guid (a dungeon-local
NPC used to keep it pinned). A camera-derived observer far from the
pre-collapsed dungeon tripped ExitDungeonExpand and unloaded it. Fix: a LIVE
in-world session never uses the fly camera for the observer — it follows the
player's server landblock, falling back to the recentered spawn center
(_liveCenterX/Y). The fly camera is the OFFLINE observer only.
2) Even with the dungeon resident, auto-entry hung: the #106 "ground ready" gate
required SampleTerrainZ under the spawn, but a dungeon's negative-offset cells
place the spawn's WORLD position in a NEIGHBOUR terrain landblock the #135
collapse deliberately doesn't load (probe: cellReady=True, terrReady=False
forever). The terrain gate is wrong for an indoor spawn — the player lands on
the EnvCell FLOOR. Fix: gate an indoor (hydratable) spawn/teleport on
IsSpawnCellReady, not the terrain heightmap; outdoor (and unhydratable→demote)
spawns still hold on terrain. Applied to both isSpawnGroundReady (login auto-
entry) and TeleportArrivalReadiness (teleport). This is the faithful equivalent
of retail's synchronous cell load + place-on-floor; the pre-#135 terrain hold
only passed because the 25x25 window streamed the neighbour terrain.
Verified live: login into 0x0007 → auto-entered player mode, snapped to
0x00070145, dungeon renders, FPS steady. Register AD-2 amended.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
On login (or teleport) into a dungeon, FPS started ~10 and climbed over ~30 s.
Root cause: the dungeon "collapse" (which shrinks the 25x25 streaming window to
the player's single dungeon landblock — AC dungeons have no neighbours) only
fires once the per-frame `insideDungeon` gate reads true, and that gate keys on
the physics CurrCell, which isn't set until the player is PLACED, which waits for
the dungeon landblock to hydrate. So during the whole hydration window NormalTick
bootstraps the full window — ~24 unrelated ocean-grid neighbour dungeons + their
~19k entities each — and the collapse only mops them up afterward. That mop-up is
the ramp.
Fix: trigger the SAME collapse early, the instant we recenter the streaming center
onto a sealed dungeon cell, before the first NormalTick.
- StreamingController.PreCollapseToDungeon(cx,cy): fires EnterDungeonCollapse
early (idempotent). The expensive neighbour window is never enqueued.
- GameWindow.IsSealedDungeonCell(cellId): reads the EnvCell dat SeenOutside flag
(CurrCell is null pre-placement) — the same flag ObjCell.SeenOutside and the
per-frame gate use, so the early decision matches the eventual one. Distinguishes
a real dungeon from a cottage/inn interior (SeenOutside → keeps its outdoor
surround). Excludes the 0xFFFE/0xFFFF structural shell ids so an outdoor spawn id
can't type-confuse a LandBlock record as an EnvCell.
- Hooks: OnLiveEntitySpawnedLocked (login) + OnLivePositionUpdated (teleport).
- Observer robustness: during a teleport PortalSpace hold the streaming observer
follows the recentered destination, not the frozen pre-teleport position (which
could drift >=2 landblocks off and trip ExitDungeonExpand). And
_lastLivePlayerLandblockId is now filtered to the player guid (resolves the
Phase A.1 TODO) so a stray NPC UpdatePosition can't drift the login-hold observer
off the dungeon.
Faithful EARLY trigger of the existing AP-36 collapse mechanism, not a new
workaround — AP-36 amended in the same commit. Adversarially reviewed across
timing / threading / faithfulness lenses; 5 new tests including the real runtime
ordering (Tick bootstraps, then PreCollapse cancels). Core suite green (1463).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Wrap-up bookkeeping for the dungeon work this session:
- #135 — login FPS ramp (~10 fps -> high over ~30 s): the streaming
collapse only fires once CurrCell resolves to a sealed cell, so the
first-frame bootstrap loads ~24 neighbour ocean-grid dungeons (+ ~19k
entities each) then unloads them. Residual of the dungeon collapse;
clean fix = pre-collapse at login when the spawn cell is a sealed
dungeon cell.
- #134 — ramp slide-response feel ("lags downward" instead of gliding
along the slope). SURFACED (not caused) by 3e006d3 caching the ramp
connector cell in the physics graph; the slope-walk/edge-slide is now
exercised. Port the retail slide-response; no band-aid.
- #133 — progress note: dungeon FPS FIXED (streaming collapse to the
single dungeon landblock, 14-30 -> ~1000+ fps) + grey barrier FIXED
(register portals-only connector cells for BOTH visibility and the
physics graph even when they build 0 sub-meshes; d90c538 + 3e006d3).
A7 per-vertex lighting bake (LightBake Core 3b93f91) is the remaining
"lighting off" work; revised diagnosis (intensity=100 is the real dat
value; the divergence is no-static-light-burnin, not a mis-read).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The faithful fix for the spotty dungeon/house/outdoor lighting is retail's per-vertex
static-light bake (D3DPolyRender::SetStaticLightingVertexColors 0x0059cfe0), NOT a
per-pixel ramp. This lands the GL-free Core: LightBake.PointContribution /
ComputeVertexColor port calc_point_light (0x0059c8b0) VERBATIM — verified against a
clean Ghidra decompile (the BN pseudo-C is x87-mangled): half-Lambert wrap with
LIGHT_POINT_RANGE=0.75 (0x007e5430), the distsq>1 norm branch, the per-channel
min-to-color clamp, and the final [0,1] clamp. static_light_factor=1.3 (0x00820e24)
is already folded into LightSource.Range by LightInfoLoader.
7 conformance tests (hand-derived golden values) green. NOT wired yet — the
integration (a per-vertex colour attribute on the cell mesh + the bake driver keyed
on envCellId + the shader consumption) is the remaining A7 work; see ISSUES.md A7.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Remove the throwaway probes added to diagnose the dungeon FPS/grey issues now that
they're fixed: the ACDREAM_LOG_FPS headless line + [cellreg] registration line
(GameWindow), and the [pv-trace] 0x0007 gate-widen + raw-NDC bbox addition to the
flap probe (PortalVisibilityBuilder, reverted to the pre-#133 form). The permanent
Phase-U.4c [flap]/[pv-trace] probes (ACDREAM_PROBE_FLAP) are kept as-is.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
After registering portals-only connector cells for VISIBILITY (d90c538), an
angle-dependent residual grey remained when the camera crossed a ramp: the
camera-collision sweep (SmartBox::update_viewer -> sphere_path.curr_cell, pc:92870)
could not transit INTO the connector cell because it had no physics cell to sweep
into — CacheCellStruct was still gated on drawable sub-meshes. So the viewer cell
stalled one cell behind the eye (confirmed live: [flap-sweep] transited every cached
neighbour but NEVER the un-cached connector 0x014D, viewerCell stuck at 0x00070103
while the eye sat 1.32 m past the connector's portal plane), and the side test
correctly culled the on-screen connector portal -> grey.
Fix: move CacheCellStruct out of the `cellSubMeshes.Count > 0` gate, next to
BuildLoadedCell — cache EVERY cell with a valid cellStruct for physics too. Retail
keeps the whole landblock cell array resident for the sweep; a portals-only
connector has an empty collision BSP but its portals drive the transit. User-gated:
"I see no grey background any longer."
Build green; 12 flood-gate tests + 677 physics/cell/transit tests green (no collision
or membership regression). TEMP render probes still retained (strip after).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The grey "barrier" at a dungeon ramp was a one-cell registration gap. The ramp's
connector cell (0x0007014D) is a portals-only pass-through — CellMesh.Build yields
0 drawable sub-meshes for it (you walk through it on adjacent floors). But the whole
registration block — including the portal-VISIBILITY registration (BuildLoadedCell ->
_cellVisibility) — was gated behind `if (cellSubMeshes.Count > 0)`. So that cell was
never added to the visibility graph; the flood lookup-missed it (PortalVisibilityBuilder
:369), couldn't traverse it to the room below, and the grey clear color showed through.
Confirmed live via two added probes: [cellreg] registered=204/205 (only 0x014D missing)
+ [pv-trace] p4->0x0007014D skip=lookup-miss. After the fix: registered=205,
hasRamp=True, skip=lookup-miss gone, the room below renders.
Fix: compute the cell transforms and call BuildLoadedCell (visibility) for EVERY cell
with a valid cellStruct, regardless of drawable sub-meshes — matching retail, which
keeps the whole landblock cell array resident before the flood runs. Drawing
(RegisterCell, _pendingCellMeshes) and the physics BSP (CacheCellStruct) stay gated on
drawable geometry (a portals-only connector has nothing to draw and no collision
surface). Not a regression from the FPS-collapse work — a pre-existing gate the
now-navigable dungeon exposed (every ramp/stair/cellar mouth would show it).
TEMP diagnostics retained for the residual angle-grey investigation (strip after):
[cellreg] (GameWindow), the 0x0007 [pv-trace] gate widen + raw-NDC bbox (PortalVisibility-
Builder). Three earlier render-math theories (portal_side, on-screen clip, near-eye
projection) were each refuted by apparatus/probe before shipping — this is the verified one.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The dungeon-streaming gate read SeenOutside from the render registry
(_cellVisibility.TryGetCell), which only succeeds AFTER the landblock FINALIZES —
~tens of seconds for a 205-cell dungeon. So the collapse fired late and the full
25x25 neighbor window churned in first ("~30s to stabilize at high FPS").
EnvCell extends ObjCell, which already carries SeenOutside (set from the EnvCell
dat flags at construction), so CurrCell.SeenOutside is available the moment the
player is placed (the snap). Read it directly instead of the registry. Collapse now
engages ~3s in (snap) instead of ~30s (finalize); residual is the ~24 neighbors the
bootstrap loads before the snap, which then unload. Also simplifies the predicate.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
After the dungeon-collapse fix the local player avatar stopped rendering: the
per-frame RelocateEntity moved the player entity to its position-derived landblock
floor(pp/192), which for a dungeon's negative-local-Y cell is the off-by-one (0,6)
— the very landblock the collapse unloads. So the player entity sat in an unloaded
landblock and was never drawn (the dungeon itself, in 0x0007, rendered fine).
Fix: when the player is in an indoor cell (CellId low word >= 0x0100), relocate to
the cell's OWN landblock (CellId >> 16), matching the streaming-collapse pin. The
cell id is authoritative for ocean-placed dungeon geometry. Outdoor entities keep
the position-derived path.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
"The dungeon is broken" — the collapse was unloading the REAL dungeon. A dungeon's
EnvCells sit at arbitrary "ocean" world coords with negative cell-local Y (snap
showed pos=(58.9,-69.6) in cell 0x00070133), so the observer landblock
_liveCenterY + floor(pp.Y/192) = 7 + floor(-69.6/192) = 7 + (-1) = 6 lands one row
off. The collapse pinned to 0x0006 and unloaded 0x0007 — the real dungeon — which
nulled CurrCell (the cell no longer existed) and left the player floating in
outdoor-lit empty space (lb 1/1 @ ~1585 fps, but the wrong landblock). This is the
Bug-A negative-local-coordinate class.
Fix: when inside a dungeon, pin the collapse to the cell's OWN landblock
(CurrCell.Id >> 16), never the position-derived observer landblock — the cell id is
the authoritative landblock for ocean-placed dungeon geometry.
Also hardened the hysteresis so a transient CurrCell flicker can't thrash:
- Re-collapse when insideDungeon at a DIFFERENT landblock (multi-landblock dungeon).
- Expand only on a DISTANT move (Chebyshev > 1) — a real exit teleports far from the
ocean-grid block; the off-by-one flicker is always an ADJACENT (±1) landblock, so
it now HOLDS the collapse instead of expanding.
- SweepCollapsed always preserves _collapsedCenter (the true dungeon landblock),
never the per-frame observer landblock.
Build green; 59 streaming tests green (flicker regression test updated to the
realistic adjacent off-by-one).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The first cut of the dungeon gate keyed expand on the per-frame insideDungeon
signal (CurrCell is a sealed EnvCell). Live, CurrCell momentarily resolves to
null mid-frame while the player stays put in the dungeon landblock, so the gate
flipped collapse→expand→collapse every few frames. Each expand re-streamed the
full 25×25 window; the unloads couldn't keep up (MaxCompletionsPerFrame=4), so
registered lights leaked to 212k and FPS spiked to single digits between the
~199 fps collapsed frames.
Fix: once collapsed, key the gate on the STABLE observer landblock, not CurrCell.
Stay collapsed while the player remains in the dungeon landblock (_collapsedCenter);
expand only when the observer actually moves to a different landblock (portal/
teleport out). CurrCell flicker no longer thrashes.
Regression test added (Collapsed_CurrCellFlickersToNull_SameLandblock_DoesNotExpand).
Build green; 60 streaming tests green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Dungeon FPS sat at ~30 (frame ~33ms) because the 25x25 streaming window around
the dungeon landblock pulled in ~129 NEIGHBORING landblocks + their thousands of
torch/particle emitters, all drawn though never visible. In AC all dungeons are
packed adjacent in the unused "ocean" map grid, so those neighbors are unrelated
dungeons. The FPS timeline proved it: 247 fps at login (lb 0/0, ~10K entities) →
17 → 30 as landblocks streamed in (lb 0→129) — the cost tracked LANDBLOCK count,
not entities.
Retail-faithful: ACE LandblockManager.GetAdjacentIDs returns ZERO adjacents for a
dungeon (`if (landblock.IsDungeon) return adjacents;`, Landblock.cs:577-582) —
every dungeon is a self-contained landblock you never see out of.
Fix: when the player stands in a sealed indoor cell (CurrCell.IsEnv &&
!SeenOutside — the same predicate that kills the sun/sky), collapse streaming to
just the player's dungeon landblock and unload the neighbors. Building interiors
(cottage/inn) have SeenOutside cells, so they are NOT gated and keep their
surrounding terrain (the frozen building/cellar demo is unaffected). Unloading the
neighbors also tears down their lights (removeTerrain → UnregisterOwner), shrinking
LightManager._all from ~2227 toward retail's ≤40 — which directly helps the A7
lighting bake landing next.
Mechanics (StreamingController):
- Edge IN: ClearPendingLoads() cancels the in-flight 25x25 window (new streamer
ClearLoads control job — worker drops queued Loads, keeps Unloads), unload every
resident neighbor, pin a radius-0 StreamingRegion, (re)load the dungeon block if
needed.
- Stay collapsed: sweep any straggler that finished loading after the edge (a Load
the worker had already dequeued before ClearLoads).
- Edge OUT (portal/teleport to outdoors): rebuild the full two-tier window at the
new center, unload anything stale.
AP-36 added to the divergence register (the gate uses the cheap SeenOutside cell
predicate as an approximation of ACE's full landblock IsDungeon classification).
GameWindow also carries a TEMP ACDREAM_LOG_FPS=1 headless FPS line (strip after
the A7 FPS+lighting verification).
Build green; 58 streaming tests green (6 new dungeon-gate tests).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The dungeon/house/outdoor lights read as hard-edged blown discs ("spotlights")
because our point/spot shader used `atten = 1.0` flat inside a hard `d < range`
cutoff. The mesh.frag comment claimed this was retail-faithful ("no attenuation
inside Range... the bubble-of-light look relies on crisp boundaries", citing
r13 10.2) — that was a misread and the literal cause of the symptom.
Verified against the decomp (not guessed): calc_point_light (0x0059c8b0, the
PER-VERTEX point-light path that lights static walls) scales each light's
contribution by (1 - dist/falloff_eff) — a LINEAR ramp that fades to exactly 0
at the edge, eliminating the hard disc. falloff_eff = Falloff * static_light_factor,
and static_light_factor = 1.3 (0x00820e24), NOT the 1.5 config_hardware_light
rangeAdjust (that 1.5 is the D3D-dynamic path for moving objects, a different
path). The Ghidra port (acclient.c:808639) is more garbled — BN pseudo-C is the
oracle here; the exact normalization factor + a half-Lambert wrap (0.5*dist+N*L)
are x87-obscured (same artifact class as GetPowerBarLevel) and left unported.
Changes:
- mesh_modern.frag + mesh.frag: replace flat atten with clamp(1 - d/range, 0, 1);
Range now carries falloff_eff so the ramp fades to 0 at the cutoff. Fix the
false "no attenuation / crisp bubble" comment in mesh.frag.
- LightInfoLoader: Range = Falloff * 1.3 (static_light_factor), was * 1.5.
- LightManager: correct the stale class doc comment (Tick is now nearest-8
allocation-free partial-select with NO viewer-range slack filter).
- divergence register: AP-16 updated (slack filter removed), AP-35 added
(per-pixel vs per-vertex Gouraud; dropped half-Lambert wrap + normalization).
- test: LightingHookSinkTests Range 8*1.3 = 10.4.
Build + 20 lighting tests green. Visual gate pending (game-wide lighting change:
dungeon torches, house candles, outdoor braziers).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Tick built a new List<>(N) and ran an O(N log N) Sort every frame; in a dungeon
N is thousands of torches, so it allocated a large list per frame (GC pressure ->
FPS). Replace with an insertion partial-select that keeps the nearest maxPoint
directly in the _active window — O(N * maxPoint), maxPoint<=8, zero allocation.
Same selection result (nearest 8); lighting suite 20/20 green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Retail PrimD3DRender::config_hardware_light (0x0059ad30) sets the hardware light
Range = Falloff * rangeAdjust (1.5, global 0x00820cc4). We used Range = Falloff, so
torches reached only 2/3 of retail -> tight 'candle/spotlight' bubbles in dungeons.
Match retail's reach. Ambient 0.20 confirmed retail-faithful (the 0.30 was CreatureMode,
not world cells). Lighting suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The active-light selection dropped any point light whose range didn't reach the
VIEWER (DistSq > Range^2*slack -> skip). Retail's D3D-style fixed pipeline picks
the 8 NEAREST lights and applies the hard range cutoff PER SURFACE in the shader
(mesh_modern.frag: if (d < range)). The viewer-range candidacy filter suppressed
a torch whenever the player stood outside its range, so a dungeon room with 2227
registered torches lit only the ~1 the player was standing in (activeLights ~= 1,
rest of the room at flat 0.2 ambient = the "lighting off" report). Drop the filter;
take the nearest 8 regardless of viewer range. Removed the now-unused RangeSlack
const; updated the two tests that codified the old filter. Core lighting suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Autonomous /loop verification: a live launch into the 0x0007 dungeon renders with a
sane budget (WB-DIAG instances ~39,000, meshMissing=0; was 9.1M pre-Bug-A), correct
membership (no ACE failed-transition spam), navigable. The chain: G.3a teleport
hold+place + Bug A (2ce5e5c, validated-claim landblock prefix) + login-into-dungeon
recenter (47ae237). A headless diagnostic (Issue95DungeonFloodDiagnosticTests, 95d9dab)
proved the portal flood is already bounded (1-17 cells vs the stab_list's 120-204), so
#95's "port grab_visible_cells stab_list bounding" was the WRONG fix and is NOT pursued.
ISSUES #95 -> RESOLVED, #133 -> renders + login-into-dungeon fixed; CLAUDE.md current
state + render digest updated. Remaining for M1.5: A7 dungeon torch/point-lighting.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A character saved inside a far dungeon hung at the #107 auto-entry hold because
the streaming center was fixed at the startup default and the login spawn never
recentered it, so the dungeon never streamed. Mirror the teleport-arrival
recenter on the login player-spawn path: when the player's spawn landblock
differs from the current center, recenter before translating the spawn position
(landblock-local -> new-center frame). No-op for a same-landblock (normal
Holtburg) login.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Re-gate of Bug A revealed: logging in with the character saved inside a far
dungeon hangs at the #107 auto-entry hold (frozen, no [snap]). The streaming
center is set once at startup to the default and the login spawn never recenters
it, so the dungeon never streams and IsSpawnCellReady never goes true. The
teleport-arrival path recenters (G.3a); the login path doesn't. Filed under #133
with the fix shape (recenter onto the spawn landblock at login) + the ACE-reset
workaround.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The "consumed by nobody (zero behavior change)" / "INERT in Stage 1 (no writer)"
comments predate the UCG becoming load-bearing. Verified against the call sites:
CellGraph is populated unconditionally in CacheCellStruct (before the idempotency
+ null-BSP guards, so BSP-less cells are included) and consumed for the player
render/lighting root (CurrCell, written at the PhysicsEngine.UpdatePlayerCurrCell
player chokepoint; read by GameWindow:7502/7717), the universal id->cell resolver
(GetVisible), the 3rd-person camera cell (FindVisibleChildCell), and the
block-local terrain origin (TryGetTerrainOrigin, read by CellTransit:484/736).
Comments only — no behavior change. Core suite 1445 passed / 2 skipped.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The G.3a visual gate ran a real PlayerTeleport into the 0x0007 dungeon. The core
hold+place worked (grounded on the dungeon floor, no ocean) and Bug A (landblock-
prefix mis-stamp) is fixed (2ce5e5c). But the gate proved #95 (portal-graph
visibility blowup, ~9.1M instances/frame) is LIVE under the current pipeline — my
plan's "likely superseded / conditional G.3b" premise was wrong. Spec §2.5/§3.2 +
ISSUES #133/#95 updated: G.3b (grab_visible_cells stab_list bounding) is REQUIRED,
needs its own grounding/brainstorm. Also noted: the render-only hydration decouple
was reverted (e7058ca) for making the player invisible at Holtburg.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The #111 validated-claim branch returned lbPrefix | (cellId & 0xFFFF), where
lbPrefix is found by searching resident landblocks for one containing the
candidate position. A dungeon EnvCell's local Y can be negative, so the dungeon
landblock fails the [0,192) bounds test and the loop matches a neighbouring
(e.g. Holtburg) resident block -> the validated claim 0x00070143 got re-stamped
0xA9B30143, making the client mis-resolve the player to the wrong landblock and
spam ACE with rejected moves. The validated claim's full id is authoritative;
return it directly. Byte-identical for the login case (position in the claim's
own landblock); fixes the far-teleport dungeon case.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Code-review follow-up: the hydration decouple's safety rests not only on
CacheCellStruct self-gating its BSP cache, but on the fact that a geometry-less
cell — though now added to the UCG CellGraph unconditionally — never enters the
_cellStruct BSP dictionary membership/placement resolve through, so the player
can never be rooted in one. Document that load-bearing invariant at the hoist.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
BuildLoadedCell + CacheCellStruct were gated behind cellSubMeshes.Count > 0, so a
geometry-less collision cell got no collision (fall-through) and no visibility
node. Retail couples neither to visible geometry; CacheCellStruct self-gates on a
null PhysicsBSP, so this is safe. Render registration stays behind the submesh
guard.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replaces the unconditional OnLivePositionUpdated snap (which resolved against
the resident old landblocks before the destination streamed in -> ocean) with a
recenter + deferred BeginArrival; per-frame Tick places via the unchanged #111
validated-claim Resolve once SampleTerrainZ + IsSpawnCellReady report ready, or
force-snaps loudly on an impossible claim / ~10s timeout.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
TDD plan for the gated G.3a core: a pure TeleportArrivalController state machine
(hold-until-hydration + force-snap on impossible/timeout) + its GameWindow wiring
(replace the unconditional arrival snap with recenter + deferred BeginArrival;
per-frame Tick; readiness predicate reusing the #107 login triplet) + the EnvCell
physics/visibility hydration decouple + the visual acceptance gate. G.3b/c/d get
their own plans after the gate.
Also syncs the spec: the readiness predicate reuses SampleTerrainZ + IsSpawnCellReady
+ IsSpawnClaimUnhydratable (the validated #107 login gate) rather than a new
IsLandblockApplied query — strictly more faithful, less new surface.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Brainstorm outcome for #133/G.3. Grounds the corrected root cause (dungeon
landblock = flat terrain + EnvCells, streams via the existing pipeline; the
blocker is the teleport-arrival snap firing BEFORE the dest landblock hydrates)
against the current code (5 verified seams) and lays out Approach C:
G.3a core teleport-into-dungeon: hold-until-hydration on the arrival path
(reuse #107 IsSpawnCellReady + IsSpawnClaimUnhydratable) + #111
validated-claim EnvCell placement + dest-ready streaming query +
dest-coord validation + timeout safety + decouple EnvCell
physics/visibility hydration from the render-mesh guard. -> VISUAL GATE
G.3b #95 stab_list bounding — CONDITIONAL on the gate showing the blowup
(its repro is stale, from the T4-deleted WB path; the current flood is
landblock-confined + enqueue-once, so #95 is likely superseded).
G.3c faithful TeleportAnimState portal-tunnel FSM (decomp 004d6300 /
219405-219774); the TAS_TUNNEL hold-exit gates on G.3a's same readiness
predicate (the tunnel IS the hold's visual form).
G.3d recall game-actions (/ls etc.) — same arrival flow; doubles as the test
lever.
Supersedes the §12 port-plan of r09 (most of it already shipped); r09 stays the
wire/format/recall contract reference. Resolves the handoff's 4 open questions.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds the 2026-06-13 dungeon-G.3 handoff doc + a dat-probe test that
RESOLVED the pivotal design ambiguity. A research agent assumed dungeon
landblocks are terrain-less (LandblockLoader.Load returns null ->
"rewrite the pipeline for terrain-less landblocks", 13 seams). The dat
probe refutes it: dungeon landblock 0x0125 has a flat (all-zero-height)
LandBlock record PLUS 71 EnvCells and no buildings/objects -> it streams
fine via the existing pipeline as a flat-terrain landblock.
The real blocker (#133) is narrow: the teleport-arrival handler
(GameWindow.cs:4928) snaps the player via physics.Resolve BEFORE the
dungeon landblock streams in -> Resolve falls back to the resident
Holtburg landblocks -> places the player at A9B3 ocean. Fix shape:
hold-until-hydration (reuse the #107 IsSpawnCellReady gate for the
teleport-arrival path) + place into the EnvCell + the retail
TeleportAnimState portal-space FSM for the full-G.3 loading screen.
ACE confirms dungeons are single-landblock, so "multi-landblock LOD"
is moot.
The handoff captures: this session's closes (#108-residual/#127/#125
gated, #116 partial), the M1.5 re-open decision, the corrected root
cause, the 5-way reference grounding (holtburger/ACE/retail decomp +
the dat probe), the design direction, and the open brainstorm questions.
Next session: resume the brainstorm at "propose approaches" -> spec ->
writing-plans -> implement.
Suites green: App 264+1skip / Core 1445+2skip / UI 420 / Net 294.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Correction to 1bf037a. The user reverted the M1.5 landing: the indoor
world isn't done while dungeons are completely broken. Attempting the
dungeon demo revealed it's not a single bug (#133) but a whole-feature
gap — terrain-less indoor-only dungeon landblocks aren't supported
anywhere in the streaming/load/render/physics pipeline:
- LandblockLoader.Load returns null when there's no LandBlock terrain
record (dungeons have none) -> the dungeon never loads.
- LandblockStreamer fails when the terrain mesh build returns null
(dungeons have no terrain mesh).
- The teleport-arrival snap Resolves before the dungeon hydrates ->
places the player in the old Holtburg frame over ocean.
The user chose the FULL Phase G.3 scope (dungeon streaming + portal-space
loading screen + multi-landblock LOD + PlayerTeleport handling) and
pulled it into M1.5. M1.5 lands only when BOTH the building/cellar demo
(done) and the dungeon demo (enter via portal, navigate 3-5 rooms, walls
block, smooth transitions) pass. M2 (CombatMath) re-deferred.
Currently brainstorming the dungeon-support design (spec ->
docs/superpowers/specs/). Docs corrected: milestones (M1.5 ACTIVE +
extended, M2 DEFERRED, currently-working-toward -> M1.5), CLAUDE.md
current-state, ISSUES.md #133 (G.3 pulled into M1.5).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
M1.5 "Indoor world feels right" lands on its primary building/cellar
demo, user-gated across the 2026-06 sessions: multi-floor inn navigation
without sling-out/wall-clip, cottage cellar descend+ascend without
falling through, walls block everywhere, smooth cell transitions. The
holistic Option-A render port (one DrawInside(viewer_cell), BR-2..BR-7 /
T1..T6) and the A6.P4 per-cell shadow physics shipped and were gated; the
doorway-flap family is closed (#119/#128, #112, #113, #124,
#129/#130/#131/#132, #108-residual, #127) and the #90/synthesis
workarounds removed.
The dungeon half is the one piece NOT landed: attempting the dungeon demo
(meeting-hall portal) surfaced issue #133 - teleport-into-a-dungeon snaps
the player BEFORE the dungeon landblock streams in (GameWindow.cs:4928
Resolve falls back to the resident Holtburg landblocks -> snaps to an
outdoor cell over ocean). That is Phase G.3 (dungeon streaming +
PlayerTeleport handling, M4), not a render bug (#95 died with the Option
A rewrite). Per the milestones doc's pre-flagged choice, the dungeon demo
is promoted to G.3 and M1.5 lands on the building/cellar demo (user
decision 2026-06-13).
Start M2 "Kill a drudge" - first port target CombatMath.ComputeDamage
(port-ready per the combat-math research memo; ACE oracle). Drudges spawn
outdoors for the demo, so M2 does not depend on #133/G.3.
Files: milestones doc (M1.5 -> LANDED, M2 -> ACTIVE, currently-working-
toward flipped), CLAUDE.md current-state -> M2, ISSUES.md #133 filed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The GL root cause was fixed in fcade06 (the gpu_us query-ring stale
errors). This closes the remaining design debt: a genuinely-failed
UploadMeshData was dropped permanently.
Exact mechanism (traced this session): UploadMeshData's catch returns
null, the staged item is already consumed, and _renderData stays empty -
but the prepared data lingers in _cpuMeshCache, so the #128 EnsureLoaded
re-arm hits PrepareMeshDataAsync's CPU-cache short-circuit
(ObjectMeshManager.cs:448-453) which returns the cached data WITHOUT
re-staging it for upload. The mesh stays invisible until CPU-cache
eviction - session-sticky under low cache pressure (the in-tower
scenario).
Fix: the per-frame Tick drain (WbMeshAdapter) now re-stages a failed
upload for the NEXT frame via ObjectMeshManager.UploadOrRequeue, bounded
by MaxUploadRetries (3). The attempt counter lives on the ObjectMeshData
object so it resets to 0 naturally on re-prepare. Re-stages are
collected and re-enqueued AFTER the drain loop, never inside it, so a
deterministic failure cannot spin the queue within a single frame; past
the cap it gives up with a loud [up-retry] ... giving up line - a
genuine GL defect now surfaces instead of the old silent permanent drop
or an unbounded retry storm. Retail loads content synchronously and has
no such failure mode; this converges the async pipeline toward that
guarantee.
The uncaught GenerateMipmaps path (open-question c) is INTENTIONALLY
left to surface errors - a blanket catch there would mask future real
defects (no-workarounds rule), and its trigger (fcade06) is retired.
No visual gate (robustness). Build green; App.Tests 264 + WbMeshAdapter
tests green. No GL-context test seam exists for the upload path, so the
bounded retry is verified by construction + the regression suite.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The user brought up Ghidra; its decompiler (patchmem.gpr, full PDB)
resolved the Binary-Ninja `test ah,5` x87 branch-sign ambiguity that
blocked the desk read. CSphere::slide_sphere (0x00537440) decompiles
cleanly to:
fVar3 = |cross(collisionNormal, contactPlane.N)|²;
if (::F_EPSILON <= fVar3) { // crease exists
... offset = cross * dot(cross,gDelta)/fVar3;
if (|offset|² < ::F_EPSILON) return COLLIDED_TS; // degenerate guard
... add_offset_to_check_pos -> SLID_TS
}
Retail compares the SQUARED magnitudes against F_EPSILON
(0.000199999995 ~= 0.0002 = PhysicsGlobals.EPSILON). Our port compared
against EpsilonSq (0.0002^2 = 4e-8) - a ~5000x too-tight threshold (the
BN pseudo-C rendered the comparison as `test ah,5` after an x87 FCMP,
which is sign-ambiguous; agent reads disagreed). Fixed both comparisons
at TransitionTypes.cs:3098,3105 to EPSILON.
Effect: crease-exists now needs >=0.81 deg between the wall and contact
normals (was 0.011 deg - which routed near-parallel pairs through the
numerically unstable projection); the degenerate guard now hard-stops
slides under ~1.41 cm like retail (was 0.2 mm). Branch POLARITY was
already correct - no change there.
No regression: full physics suite (612) + full Core (1443) green. Not a
register deviation (no row existed; this is an undocumented porting
error corrected to match retail).
This does NOT close#116 - it fixes a tangential constant, not either
reported shape. Ghidra also settled the two shapes' diagnosis (recorded
in ISSUES.md #116 + physics digest):
- Shape-1: our cn=UnitZ default IS retail-faithful (validate_transition
0x0050aa70 has the identical `if (collision_normal_valid==0)
set_collision_normal(UnitZ)`). The real divergence is upstream -
tick-22760 our collision_normal_valid was false where retail's was
true (it recorded the door-face normal). Needs the instrumented
tick-22760 replay.
- Shape-2 (D4 stays skipped, note sharpened): slide_sphere slides
in-frame (SLID_TS) so Z=1.92 is faithful and the D4 Z=2.0 hard-stop
pin is the suspect half; the threshold fix didn't move D4 (real slide,
not degenerate). Needs a cdb trace of an airborne wall hit.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Read both sides quote-for-quote (verified against source): our
CSphere::slide_sphere port (TransitionTypes.cs:3054-3133) vs retail
0x00537440 (decomp 321403-321532). Findings:
1. Shape-1 (tick-22760 lost 3.57cm slide) is NOT the degenerate-offset
guard - retail's guard only kills slides under ~1.4cm. The real
divergence is the collision-normal SOURCE: our harness cn=(0,0,1) vs
live cn=(0,+1,0). Strong lead: TransitionTypes.cs:3701-3702 defaults
cn=Vector3.UnitZ when no valid normal.
2. Shape-2: retail's slide_sphere applies the slide IN-FRAME
(add_offset_to_check_pos @0x53777e, return SLID) - our in-frame slide
to Z=1.92 is likely retail-faithful and the D4 frame-1 hard-stop pin
is the stale one (pending the first-airborne-frame plane state).
3. Candidate epsilon-squaring divergence: retail compares SQUARED
quantities against 0.000199999995 (non-squared) while we compare
against EpsilonSq=0.0002^2 - possibly 1e4x too small. Explains
neither shape; DO NOT change without cdb (the test ah,5 branch
polarity is the undecodable BN construct from the PosHitsSphere saga).
No code change (oracle-first; the issue title calls for cdb and the BN
decomp is provably ambiguous on the branch signs). ISSUES.md #116 +
physics digest carry the leads + the scoped cdb plan.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
