wip: continuous backbone floor profile for the cave planner (L134 design 4, untested)

Written by Luna before the Codex usage limit cut the round: one shared distance/floor profile across backbone rooms
and edges, edge cubic floors derived from it, sloped route-room floors, lower-floor backing cap at overlaps, and
validator checks for joint height/slope continuity. Not yet built or audited.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1

ARCHITECTURE.md

@@ -1522,11 +1522,12 @@ collision ring became walkable at 0.632 s. `voxel.StreamingCoarseFirst=1` with
 residency by about one third on the staged plan, but null-RHI residency does not prove rendered
 enclosure. Keep coarse-first opt-in until that view is checked in a rendered owner-plan run.
 
-The separate `voxel.StreamingStartupApplyBoost` switch stays default-off. At `1`, it temporarily allows
-12 visible mesh submissions / 14 ms while the initial desired set and walkable ring are incomplete,
-with a 30 second ceiling; at `0`, the authored budget is unchanged. On the matched legacy-order test,
-the boost changed full-set readiness from 17.412 s to 17.346 s and removed count-cap frames, but did
-not improve end-to-end residency. Keep the normal 4 / 2 ms guard for play.
+The separate `voxel.StreamingStartupApplyBoost` switch is default-on in the current code. At `1`, it
+temporarily allows up to 24 visible mesh submissions / 14 ms while the initial desired set and
+walkable ring are incomplete, with a 30 second ceiling; at `0`, the authored budget is unchanged.
+The historical matched S6 run changed full-set readiness from 17.412 s to 17.346 s and removed
+count-cap frames, but did not improve end-to-end residency. These current defaults are defined in
+`Source/VoxelForge/Private/VoxelWorld.cpp`; the S6 measurement remains historical context.
 
 The S6 trace found the shared pool was not saturated: its 10-slot idle estimate was about 132
 slot-seconds on legacy order and 73 on gated coarse-first, while game-thread tile applies totaled only
@@ -1556,8 +1557,8 @@ by its RAII guard when the worker lambda exits. This still counts queued and run
 results do not consume an admission slot. `voxel.StreamingLegacyPendingResultCap=1` restores the old
 `PendingTiles` accounting; the test harness's `NearReserveMode=Legacy` also sets
 `voxel.StreamingDynamicNearReserve=0`, reproducing the S6 path. `StreamingDynamicNearReserve` now
-defaults to `1`; `StreamingCoarseFirst` remains `0`, and the apply boost remains `0` in the matched
-measurement.
+defaults to `1`; `StreamingCoarseFirst` remains `0`. That historical S6 matched measurement disabled
+the startup apply boost (`voxel.StreamingStartupApplyBoost=0`); the current default is on (§8.27).
 
 On the 2,578-key owner SurfaceWorld trace, the old-path control reached the full set in 17.527 s,
 walkable ring in 0.650 s, and full LOD0 in 2.281 s. Dynamic admission reached them in 7.780 s,

CLAUDE.md

@@ -10,6 +10,7 @@ then wait for his results / pasted errors. Same for in-editor checks — ask for
 try to run anything.
 
 ## Read first (in order, load only what the task needs)
+- **[CURRENT-STATE.md](CURRENT-STATE.md) — current shipping snapshot and verification entry point.**
 - **[CODEMAP.md](CODEMAP.md) — ALWAYS.** Navigation: what it is, data-flow (§2), symbol→`file:line`
   index (§3), "I want to change X → go here" (§5), conventions & gotchas (§6). Trust **symbol names
   over line numbers** (lines drift).

CODEMAP.md

@@ -388,7 +388,7 @@ casse cette arithmetique *structurellement* — `BeginPlay` loggue une **Error**
 | `AdvanceHeadlessDigMotionTest` / `RecordHeadlessDigMotionFrame` | 3060 / 3217 | Explicit `-voxel.TestDigMotion=1` repro: optional near digs, then repeats underfoot digs and samples `Generator->GetDensityAt` immediately after each edit, on its first subsequent simulated frame, and on every following observation frame. `-voxel.TestDigMotionFieldOverride=1` is a measurement-only prototype that clears `CurrentFloor` and switches CMC to falling when the sampled post-edit density is air; it is inactive by default. The test counts false support, false air and downward velocity beyond one gravity step. `-voxel.TestDigMotionUnderfootDigCount` is clamped to 1–100; `-voxel.TestDigMotionUnderfootIntervalSeconds=0` performs per-frame rapid edits. The accompanying gap audit is scoped to the dig session. |
 | `MaybeFinishHeadlessStreamingTest` | — | The existing `voxel.TestExitSeconds` diagnostic emits `VoxelForgeStreamingExitDiag` at most every 30 elapsed seconds, including threshold and early-return test flags; no output when the timer is unconfigured. |
 | `UpdateVisualStalenessAudit` | 5034 | Test-only camera-frustum coverage audit: tracks up to 24 real pre-skirt isosurface points per visible component, then counts frames where no visible tile component covers a previously visible, still-desired point (excluding edited bounds). `voxel.TestVisualGapInjectEarlyHide=1` hides a valid component for three frames and must be caught. `voxel.TestGapAuditLegacyHandoff=1` restores old handoff culling for A/B baselines. |
-| `ProcessPendingChunks` | 7435 | Drains `CriticalProcessQueue`, `EditedProcessQueue`, `NearProcessQueue`, then `ProcessQueue`; applies under the authored 4-mesh/2 ms budget during normal play and the bounded shared startup budget (12 meshes/14 ms) until the initial desired set and walkable ring are ready. `voxel.StreamingStartupApplyBoost=0` restores the authored budget. `startup_apply_config` traces the actor-read CVar and both budgets; `startup_apply_budget` records activation/end. Reads per-request cancellation before removing its map entry, traces stale/obsolete/aborted discards with epoch and ordinal, then applies via `ApplyTileResult`. Step 1 prequeue still releases here and emits `VFStep1UploadCorpusReleased`. |
+| `ProcessPendingChunks` | 7435 | Drains `CriticalProcessQueue`, `EditedProcessQueue`, `NearProcessQueue`, then `ProcessQueue`; applies under the authored 4-mesh/2 ms budget during normal play and the default-on bounded startup budget (24 meshes/14 ms, at most 30 s) until the initial desired set and walkable ring are ready. `voxel.StreamingStartupApplyBoost=0` restores the authored budget. `startup_apply_config` traces the actor-read CVar and both budgets; `startup_apply_budget` records activation/end. Reads per-request cancellation before removing its map entry, traces stale/obsolete/aborted discards with epoch and ordinal, then applies via `ApplyTileResult`. Step 1 prequeue still releases here and emits `VFStep1UploadCorpusReleased`. |
 | `ApplyTileResult` | — | **Shared game-thread apply** for one `FChunkResult` (async drain + sync carve): discards stale generation epochs and keys no longer desired, marks loaded, ingests capture, EMPTY releases the tile's existing component (a re-gen can flip content→empty on band change — old geometry must not linger), else `ApplyMeshToTile`. Returns true iff a visible mesh uploaded (counts the budget). Doesn't touch `PendingTiles` (caller's). |
 | `FVoxelStackSampler` / `FScopedVoxelStackRegistration` (`VoxelStackSampler.h/.cpp`) | — | **Diagnostic raw stack sampler, off by default.** A sampler thread captures registered generation workers' stacks (`CaptureThreadStackBackTrace`) every N µs and tags each with the tile's LOD. It writes bounded raw PCs to CSV plus a symbolization-free summary under `Saved/`; symbol resolution is deferred until after process exit so the measured process never initializes/calls DbgHelp for this diagnostic. Lifecycle teardown is serialized and registration TLS is session-tagged. Registration is per tile TASK in `LoadTile` (and the export's tasks), never per sample; disabled costs one relaxed atomic load. Game `-voxel.SampleStacks=<us>`, commandlet `-samplestacks[=us]`. |
 | Room-cache performance (2026-09-13): `VF_FindPlayerFitPointForRoomMemoized` / `GPlayerFitMemo` (`VoxelCaveMorphology.cpp`), `voxel.TileCacheWindow` + `VoxelGenLOD::TileOriginVoxels/TileCellsPerAxis` (`VoxelGenerator.h`), `FChunkSDFSpatialIndex` | — | **Player-fit memo:** exact 97-word key, process-wide, `voxel.PlayerFitMemo`. **Tile cache window:** the room cache covers the requesting tile's footprint plus a fixed collect margin. It is set via `TGuardValue` in `GenerateMesh`/classify, and is window-invariant by construction (the margin does not grow with the tile). **LOD-gated:** fused path at LOD ≥ `voxel.TileCacheWindowMinLOD` (3), op-stack at LOD ≥ `voxel.TileCacheWindowOpStackMinLOD` (4); lower LODs keep the 4-chunk worker window, because the tile window slowed LOD0 evaluation ~6%. Strates whose authored terrain ops exceed the proven 256-voxel extent (or are malformed/unloaded) fall back to the legacy window (`VF_ComputeTerrainOpEnvelope`). `voxel.SpatialIndex` is -1 auto (off at fused LOD0, on elsewhere), 0 or 1. **Spatial index:** ordered XY buckets over cache primitives, exact rejection. The `static_assert` in `VoxelStrateTypes.h` ties `VF_STRATE_PARAM_FIELDS` to `FStrateGenerationParams` (83 fields + 8 B padding). |
@@ -396,7 +396,7 @@ casse cette arithmetique *structurellement* — `BeginPlay` loggue une **Error**
 | `RecordTileHash` / `WriteTileHashDump` | — | Headless `-voxel.TileHashDump=<absolute path>` parity instrument. Records the exact `FVoxelMeshData` used by the streaming result, including empty/classified tiles, then writes sorted absolute tile keys, LOD step/cells, content-band limits, settings digest context, and the shared `VoxelGeometryHash` at `EndPlay`. |
 | `SyncRemeshTile` | — | **INSTANT DIG**: level-0 same-frame re-mesh on the game thread. Uses `PreEditDensityCache` when the tile is resident in the bounded local cache, replays the current DiffLayer over that exact procedural lattice, then meshes/applies inline so a carve is visible THIS frame. A cache miss uses normal generation; it captures the unedited lattice if the tile needs meshing, while uniform-tile proofs remain enabled. |
 | `PreEditDensityCache` / `VoxelForgePreEditDensityCache::BuildEditedGrid` | — | Per-world cache of the pawn's current L0 tile: exact 35³ pre-edit samples, max **171,500 bytes**. It is pruned when the pawn crosses chunk boundaries and cleared on `RegenerateAllChunks`. `BuildEditedGrid` snapshots touched diff chunks once and reproduces `GetDensityAt`'s final `pre-edit - EvaluateMods` operation before the per-call shared-grid mesher path. |
-| `UpdateChunksAroundPosition` / `GetGenerationAdmissionCount` | 10990 / 10971 | Shared Core/legacy scheduler. Builds the desired set and delta-culls leavers; admits the occupied/support tile plus collision-required LOD0 tiles in the pawn's 3×3 XY ring as `CollisionCritical`. S5 coarse-first admission stays opt-in (`voxel.StreamingCoarseFirst=1`). `voxel.StreamingDynamicNearReserve=1` is now the default: it keeps the authored near reserve while moving or until the stationary walkable ring is ready, then makes those slots available to coarse work. Admission counts submitted queued/running tasks, not completed results waiting in the MPSC result queues; `voxel.StreamingLegacyPendingResultCap=1` restores the old `PendingTiles`-based cap for the negative control. `PendingTiles` still lasts through game-thread apply for duplicate suppression and result bookkeeping. The clipmap stays non-overlapping and load-before-unload / `TransitionHold` reconciliation remains authoritative. Standalone has no mirror of this actor scheduler. §8.10 + §8.26–8.28. |
+| `UpdateChunksAroundPosition` / `GetGenerationAdmissionCount` | 10990 / 10971 | Shared Core/legacy scheduler. Builds the desired set and delta-culls leavers; admits the occupied/support tile plus collision-required LOD0 tiles in the pawn's 3×3 XY ring as `CollisionCritical`. S5 coarse-first admission stays opt-in (`voxel.StreamingCoarseFirst=1`). `voxel.StreamingDynamicNearReserve=1` is the default: it keeps the authored near reserve while moving or until the stationary walkable ring is ready, then makes those slots available to coarse work. Admission counts submitted queued/running tasks, not completed results waiting in the MPSC result queues; `voxel.StreamingLegacyPendingResultCap=1` restores the old `PendingTiles`-based cap for the negative control. `voxel.StreamingFarApplySupplement=1` is the default L121 lane: after the visible-apply cap, at most one far result is admitted when under half the apply-time budget and startup boost is inactive. `PendingTiles` still lasts through game-thread apply for duplicate suppression and result bookkeeping. The clipmap stays non-overlapping and load-before-unload / `TransitionHold` reconciliation remains authoritative. Standalone has no mirror of this actor scheduler. §8.10 + §8.26–8.28. |
 | `AdvanceHeadlessCollisionGateStressTest` / `ObserveHeadlessCollisionGateStressTest` / `MaybeFinishHeadlessCollisionGateStressTest` / `MeasureHeadlessCollisionGateStressTerrain` | — | Explicit `-voxel.TestCollisionGateStress=1` harness only. The normal spawn-height cvar offsets from the measured density surface; `voxel.TestCollisionGateStressDeepDrop=1` carves a process-local box shaft from above the top to just above the floor tile, extending along the post-landing route so the pawn does not run into the shaft sidewall, temporarily raises only the transient test DiffLayer brush-radius allowance for that fixture, removes only the local pawn's redundant streaming anchor for this isolated repro, proves every sample in two complete outside-ring level-0 tiles is air before seeding their known-empty status, prewarms the known-empty in-ring column, and asserts that the support walk reaches the in-layout floor outside the aligned level-0 render ring before the pawn drops. The log reports whether the fixed gate retains that otherwise out-of-ring target. Other streaming anchors remain active. The fixture rejects a target or floor outside the layout, a non-air empty-tile proof, or a floor inside the aligned level-0 ring. The `GetDensityAt` column at capsule feet defines the local air/solid surface; the deep-drop column crossing is cached after its first exact scan. Each frame records actor position, velocity, direct foot tile and gate support readiness, gate state, signed feet clearance, and a ≤0.5-voxel sweep between positions. PASS requires no measured clearance below −5 cm; deep-drop gate-on also requires horizontal travel during a hold. Result logs include the terminal density column, first crossing below spawn height, hold count/time, and gated horizontal progress. |
 | `BuildDesiredTiles` | 6482 | Builds `DesiredSorted`+`DesiredStamped` (player clipmap shells), then adds streaming anchors and the pending player's collision-only support column before the leaver sweep. It orders from the true local pawn position and builds `CriticalDesiredTiles` for the unresolved support target, occupied/support and heading tiles. Also rebuilds `DesiredTransitionDescendants` (ancestor → desired descendants, including face/edge-touching replacements) for the reconcile pass. |
 | `CollisionGateDesiredTiles` / `RequestCollisionGateSupportTile` | 5062 | Retains an unresolved support target and up to three level-0 tiles below it across desired-set rebuilds. The additional keys are hidden collision-only tiles, queued ahead of ordinary streaming, and pruned once resolved or outside the active column. |

CURRENT-STATE.md

@@ -0,0 +1,57 @@
+# VoxelForge current state
+
+As of 2026-10-31. L131 is on `assembler/world` as `8107196`; this L132 worktree is on `wip/l132-dispatch-truth`, starting at `ebae71b`. Current-state claims below point to their code or measured report.
+
+## What ships today
+
+- **Core:** evaluator version **227**. Unreal's `AVoxelWorld::bUseVoxelForgeCore` is opt-in and defaults to `false`; the legacy generator remains the default runtime path. [Core version](Standalone/Core/include/VoxelForgeCore/VoxelForgeCore.h), [runtime switch](Source/VoxelForge/Public/VoxelWorld.h), [L130 report](Standalone/Research/WorldAssemblerM3/Linux130/REPORT.md).
+- **Verified owner4 generator dispatch (L132):** SW binds `SurfaceWorld-native` because the owner export contains a matching `SurfaceWorld.Heightfield2D` v2 plan. FI binds `compiled-analytic/FloatingIslands` because its `FloatingIslands.ClusterField` v0 layout has no matching native island plan. TN and VS bind `compiled-analytic/TunnelNetwork` and `compiled-analytic/VerticalShafts`; their legacy recipe IDs provide neither a `NativeCaves`/`NativeLattice` graph recipe nor native field capability. The L132 fast-tier report prints the binding and generated tile-center share at each owner strate. [Dispatch](Standalone/Core/src/ShapeWorldInternal.h), [adapter recipe checks](Standalone/Core/include/VoxelForgeCore/VoxelForgeCore.h), [L132 report](Standalone/Research/WorldAssemblerM3/Linux132/REPORT.md).
+- **Passages:** continuous smooth swept tubes with a continuous floor profile. L125 stamps conservative route slabs and outside sentinels into the tile cache; exact route geometry is still evaluated. All 21,425 route samples passed the open-air walk check; the nine-station floor probe's maximum root offset was 7.884 mm. L130 sets the shared influence reach to **1.0533 m** (0.75 m shell + 0.21 m relief + 0.50 m smooth-max support / 6 + 0.01 m margin), converted once from metres to Core voxels for the proof, feature, fallback, and off-grid bounds. [L125](Standalone/Research/WorldAssemblerM3/Linux125/REPORT.md), [L130](Standalone/Research/WorldAssemblerM3/Linux130/REPORT.md).
+- **Streaming:** arrival uses a bounded startup apply burst: `voxel.StreamingStartupApplyBoost=1` by default, up to 24 visible meshes or 14 ms per frame for at most 30 seconds, until the initial desired set and walkable ring are ready. `voxel.StreamingFarApplySupplement=1` is also default-on: after the normal visible apply cap, at most one far result applies when under half the apply-time budget; it is inactive during the startup burst. Dynamic near reserve defaults on; coarse-first remains off; the legacy pending-result cap defaults off. The far-apply lane is Unreal-only. [CVar source](Source/VoxelForge/Private/VoxelWorld.cpp), [L121 report](Standalone/Research/WorldAssemblerM3/Linux121/REPORT.md), [architecture](ARCHITECTURE.md).
+- **Determinism:** proven on Linux with Clang and GCC at W1/W8/W12. L131's unchanged-output comparison covers owner4's SW/FI/TN/VS views in 24 compiler/worker cases; the canonical owner4/owner2 recipe matrix also passes. Windows determinism is unverified. [L131 checks](Standalone/Research/WorldAssemblerM3/Linux131/REPORT.md), [v227 canonical hashes](Standalone/Tools/verify/baselines/owner-hash/canonical/).
+
+## Current performance references
+
+Headless medians in seconds, five interleaved L80-style measurements per version in each cell. Near-ready means the complete LOD0–2 set; whole-view is the full requested view. Values below are the edited L131 binary; the paired pre-cleanup comparison is in the [L131 report](Standalone/Research/WorldAssemblerM3/Linux131/REPORT.md) and [timing summary](Saved/Linux/l131/timing/summary.json).
+
+| Stratum | Near-ready W8 / W12 | Whole-view W8 / W12 |
+|---|---:|---:|
+| S0 | 0.804 / 0.848 | 1.811 / 1.832 |
+| S1 | 1.014 / 1.002 | 1.993 / 1.886 |
+| S2 | 1.060 / 1.012 | 1.960 / 1.865 |
+| S3 | 0.917 / 0.884 | 1.887 / 1.782 |
+
+Across 40 paired runs per metric, the mean relative change from pre-cleanup L130 was −0.22% near-ready (95% CI −1.02% to +0.57%) and −0.29% whole-view (−1.04% to +0.46%); neither indicates a measurable slowdown. The machine gate was load 0.81 and 99.2% idle. [L80 protocol](Standalone/Research/WorldAssemblerM3/Linux80/REPORT.md), [runner and raw runs](Standalone/Research/WorldAssemblerM3/Linux131/RunL131Timing.py).
+
+The latest recorded Linux Unreal start-to-LOD0–2 values are L127's one run per stratum: S0 2.456 s, S1 2.231 s, S2 2.514 s, S3 2.221 s. L130 did not rerun UE startup or movement. Five L127 runs at 30 m/s, 12 s moving plus 5 s dwell, had median moving near coverage **40.22%**, far holes **3.46%**, tick p99 **10.80 ms**, and worst tick **13.27 ms**. After stopping, near coverage reached 100% and far holes reached 0%. These are the latest UE references, not v227 UE measurements. [L127 report](Standalone/Research/WorldAssemblerM3/Linux127/REPORT.md), [L127 movement summary](Saved/Linux/l127/ue/movement-pairs-summary.json), [L127 start summaries](Saved/Linux/l127/ue/merge/start-summary.json).
+
+## Open gaps
+
+- Multiplayer world lifecycle is unfinished: seed/layout replication, authoritative modification RPCs, and serialized diff replay for late join are still open. See [ARCHITECTURE §9.5 and §9.7](ARCHITECTURE.md).
+- Determinism has not been verified on Windows. Linux Clang/GCC agreement does not establish MSVC agreement.
+- NullRHI measures residency and apply timing, not the first presented pixel or rendered enclosure. L127 also found four low-contrast views among 23 otherwise verified captures. [L121](Standalone/Research/WorldAssemblerM3/Linux121/REPORT.md), [L127](Standalone/Research/WorldAssemblerM3/Linux127/REPORT.md).
+- UE startup/movement numbers above predate L130's v227 reach change; there is no L130/L131 UE movement comparison.
+
+## How to verify
+
+- Build standalone Clang or GCC with [`Standalone/BuildLinux.sh`](Standalone/BuildLinux.sh), for example `bash Standalone/BuildLinux.sh --compiler=clang++-19 --variant=l132-clang --fp-contract=off --jobs=4`. The script pins CPUs 0–11 and applies `prlimit --as=6442450944` itself.
+- Fast tier: `python3 Standalone/Tools/verify/verify.py fast --out Saved/Linux/l132 --overall-timeout 1740`. It includes the L132 four-pose owner density-dispatch report; per-run logs and the summary stay under `Saved/Linux/l132`.
+- Canonical output baselines are the v227 files in `Standalone/Tools/verify/baselines/owner-hash/canonical/`. L131 also provides reproducible pre-cleanup identity matrices for owner4 tile hashes and owner4/owner2 canonical hashes under `Standalone/Research/WorldAssemblerM3/Linux131/`.
+- Timing follows L80: quiet machine, no overlapping build, one warm-up per version, five interleaved runs per stratum and worker count, W8/W12. Each headless run uses `taskset -c 0-11`, `prlimit --as=6442450944`, and a 1,740 s timeout. See [`RunL131Timing.py`](Standalone/Research/WorldAssemblerM3/Linux131/RunL131Timing.py).
+- Linux host build: `/home/jahni/UnrealEngine/Engine/Build/BatchFiles/Linux/Build.sh VFHostEditor Linux Development -Project=/home/jahni/VFHost/VFHost.uproject -NoUBA -MaxParallelActions=4 -Log=/home/jahni/VoxelM/Plugins/VoxelForge/Saved/Linux/l131/ubt.log`. Run Unreal without `prlimit`.
+
+## Closed experiments
+
+These rejected speed/streaming proposals stay closed unless new evidence changes their measured failure. Reports and surviving patches are linked in the [L131 inventory](Standalone/Research/WorldAssemblerM3/Linux131/REPORT.md).
+
+| Line | Result |
+|---|---|
+| L116 tunnel chord bound and z cache | Pair count fell 20.8%, but stayed 1.96× L109 and readiness was 15–30% slower. Patch: [`chord-bound-zcolumn-not-merged.patch`](Standalone/Research/WorldAssemblerM3/Linux116/patches/chord-bound-zcolumn-not-merged.patch), commit `89504ec`; see [report](Standalone/Research/WorldAssemblerM3/Linux116/REPORT.md). |
+| L118 pair pruning / deferred floor work | Pruned 17–18% of candidate pairs but remained 17.2–25.7% behind L109 near-ready. [Report](Standalone/Research/WorldAssemblerM3/Linux118/REPORT.md); patch: [`l118-designs.patch`](Standalone/Research/WorldAssemblerM3/Linux118/l118-designs.patch). |
+| L126 floor-call filters | Cut floor calls 20–23% without reducing pair counts; all eight cells missed L109. Candidate source was restored and no standalone patch was kept; see [report](Standalone/Research/WorldAssemblerM3/Linux126/REPORT.md) and commit `01a5094`. |
+| L128 half-open slab ownership | Ownership audit passed, but readiness was 14.8–28.0% behind L109 and moving coverage dropped. Report and patch `l128-slab-ownership.patch` are in archived commit `7022d02`. |
+| L129 2 m physical reach | v225's metre/Core-voxel bounds mismatch made the correctly bounded candidate do more work; it was slower and failed UE movement gates. Report and patch `l129-reach-2m.patch` are in archived commit `0d1140c`. |
+| L120 priority split / extra worker slots | Neither met coverage and tick gates; 10 slots reached a 16.67 ms worst tick against 15.43 ms baseline. [Report](Standalone/Research/WorldAssemblerM3/Linux120/REPORT.md); no production patch retained. |
+| L122 admission relaxation | Early admission missed p99 or far-hole gates; no behavior retained. [Report](Standalone/Research/WorldAssemblerM3/Linux122/REPORT.md); patch: [`l122-trace.patch`](Standalone/Research/WorldAssemblerM3/Linux122/l122-trace.patch). |
+| L123 lookahead / completion gate | Lookahead failed p99 and worsened near/far coverage; queue-only also failed p99. [Report](Standalone/Research/WorldAssemblerM3/Linux123/REPORT.md); patch: [`l123-queue-lookahead.patch`](Standalone/Research/WorldAssemblerM3/Linux123/l123-queue-lookahead.patch). |
+| L124 early LOD0 pages | Optimistic gain stayed below 100 ms except negative cases; no publication/scheduling implementation was made. [Report](Standalone/Research/WorldAssemblerM3/Linux124/REPORT.md). |

Source/VoxelForge/Private/VoxelStartupTrace.cpp

@@ -423,6 +423,7 @@ void RecordApplyFrame(const FApplyFrameSample& Sample)
              "\"active_tasks\":%d,\"running_workers\":%d,\"task_cap\":%d,"
              "\"pending_tiles\":%d,\"pending_results_estimate\":%d,"
              "\"results_drained\":%d,\"apply_results\":%d,\"meshes_applied\":%d,\"max_applies\":%d,"
+             "\"supplemental_far_applies\":%d,"
              "\"apply_budget_ms\":%s,\"count_limit_reached\":%d,\"time_limit_reached\":%d,"
              "\"pending_by_lod\":%s,\"apply_results_by_lod\":%s,"
              "\"meshes_by_lod\":%s,\"apply_ms_by_lod\":%s"),
@@ -432,7 +433,8 @@ void RecordApplyFrame(const FApplyFrameSample& Sample)
         Sample.RunningWorkerTasks, Sample.TaskCap, Sample.PendingTiles,
         Sample.PendingResultsEstimate, Sample.ResultsDrained, Sample.ApplyResultsByLevelTotal,
         Sample.MeshesApplied,
-        Sample.MaxApplies, *Number(Sample.ApplyBudgetMilliseconds),
+        Sample.MaxApplies, Sample.SupplementalFarApplies,
+        *Number(Sample.ApplyBudgetMilliseconds),
         Sample.bCountLimitReached ? 1 : 0, Sample.bTimeLimitReached ? 1 : 0,
         *PendingByLevel, *ApplyResultsByLevel, *MeshesByLevel, *ApplyMillisecondsByLevel));
 }

Source/VoxelForge/Private/VoxelWorld.cpp

@@ -268,6 +268,12 @@ namespace
         GVoxelForgeStreamingResultBacklogCap,
         TEXT("Maximum queued results plus admitted tile tasks; 0 disables the safety cap. This is separate from the worker-slot cap."));
 
+    int32 GVoxelForgeStreamingFarApplySupplement = 1;
+    FAutoConsoleVariableRef CVarVoxelForgeStreamingFarApplySupplement(
+        TEXT("voxel.StreamingFarApplySupplement"),
+        GVoxelForgeStreamingFarApplySupplement,
+        TEXT("After the normal visible-apply count is reached, allow one far mesh apply when half of the frame time budget remains; 0 disables the supplemental far lane."));
+
     int32 GVoxelForgeStreamingReadyQueueDepth = 64;
     FAutoConsoleVariableRef CVarVoxelForgeStreamingReadyQueueDepth(
         TEXT("voxel.StreamingReadyQueueDepth"),
@@ -2223,6 +2229,12 @@ void AVoxelWorld::BeginPlay()
     }
     FParse::Value(FCommandLine::Get(), TEXT("voxel.StreamingStartupApplyBoost="),
                   GVoxelForgeStreamingStartupApplyBoost);
+    if (FParse::Value(FCommandLine::Get(), TEXT("voxel.StreamingFarApplySupplement="),
+                      GVoxelForgeStreamingFarApplySupplement))
+    {
+        GVoxelForgeStreamingFarApplySupplement =
+            GVoxelForgeStreamingFarApplySupplement != 0 ? 1 : 0;
+    }
     // Headless BuildHost does not reliably apply arbitrary console-variable command-line tokens
     // before BeginPlay; parse the test switches explicitly so A/B runs exercise the requested path.
     // BuildHost sans interface n'applique pas toujours les CVar arbitraires avant BeginPlay ;
@@ -2737,6 +2749,7 @@ void AVoxelWorld::BeginPlay()
                  "\"streaming_collision_lookahead\":%d,\"core_region_concurrent_admission\":%d,"
                  "\"collision_gate\":%d,"
                  "\"streaming_result_backlog_cap\":%d,\"streaming_startup_apply_boost\":%d,"
+                 "\"streaming_far_apply_supplement\":%d,"
                  "\"streaming_background_worker_reserve\":%d,\"core_parallel_for_background_priority\":%d,"
                  "\"streaming_tile_background_low\":%d,\"streaming_tile_background_normal\":%d,"
                  "\"core_column_cache_direct_mapped\":%d,"
@@ -2758,6 +2771,7 @@ void AVoxelWorld::BeginPlay()
             GVoxelForgeStreamingCollisionLookahead, GVoxelForgeCoreRegionConcurrentAdmission,
             GVoxelForgeCollisionGate,
             GVoxelForgeStreamingResultBacklogCap, GVoxelForgeStreamingStartupApplyBoost,
+            GVoxelForgeStreamingFarApplySupplement,
             GVoxelForgeStreamingBackgroundWorkerReserve,
             GVoxelForgeCoreParallelForBackgroundPriority,
             GVoxelForgeStreamingTileBackgroundLow,
@@ -9881,6 +9895,7 @@ void AVoxelWorld::ProcessPendingChunks()
     const double ApplyBudgetSeconds =
         static_cast<double>(ApplyBudgetMilliseconds) * 0.001;
     int32 MeshesApplied = 0;
+    int32 SupplementalFarApplies = 0;
     int32 ResultsDrained = 0;
     int32 ApplyResultsByLevelTotal = 0;
     double ApplySeconds = 0.0;
@@ -9888,6 +9903,8 @@ void AVoxelWorld::ProcessPendingChunks()
     TStaticArray<int32, 9> MeshesAppliedByLevel{};
     TStaticArray<double, 9> ApplySecondsByLevel{};
     bool bTimeBudgetHit = false;
+    bool bForceSupplementalFarDequeue = false;
+    bool bDequeuedSupplementalFar = false;
     const auto RecordApplyFrame = [&]()
     {
         if (!VoxelForgeStartupTrace::IsActive()) return;
@@ -9911,6 +9928,7 @@ void AVoxelWorld::ProcessPendingChunks()
         Sample.ResultsDrained = ResultsDrained;
         Sample.ApplyResultsByLevelTotal = ApplyResultsByLevelTotal;
         Sample.MeshesApplied = MeshesApplied;
+        Sample.SupplementalFarApplies = SupplementalFarApplies;
         Sample.MaxApplies = MaxApplies;
         Sample.ApplyBudgetMilliseconds = ApplyBudgetMilliseconds;
         Sample.bCountLimitReached = MeshesApplied >= MaxApplies;
@@ -9939,6 +9957,7 @@ void AVoxelWorld::ProcessPendingChunks()
     FChunkResult DequeuedChunk;
     const auto DequeueNextResult = [&]() -> bool
     {
+        bDequeuedSupplementalFar = false;
         if (bStep1UploadCorpusReleased
             && Step1HeldApplyCursor < Step1HeldApplyResults.Num())
         {
@@ -9954,8 +9973,19 @@ void AVoxelWorld::ProcessPendingChunks()
         // Mirror the admission order. A horizon result that completed earlier must not consume
         // the game-thread apply budget ahead of collision support, an edit, or ordinary LOD0.
         if (CriticalProcessQueue.Dequeue(DequeuedChunk)
-            || EditedProcessQueue.Dequeue(DequeuedChunk)
-            || NearProcessQueue.Dequeue(DequeuedChunk)
+            || EditedProcessQueue.Dequeue(DequeuedChunk))
+        {
+            RecordStreamingResultDequeued(DequeuedChunk);
+            return true;
+        }
+        if (bForceSupplementalFarDequeue)
+        {
+            if (!ProcessQueue.Dequeue(DequeuedChunk)) return false;
+            bDequeuedSupplementalFar = true;
+            RecordStreamingResultDequeued(DequeuedChunk);
+            return true;
+        }
+        if (NearProcessQueue.Dequeue(DequeuedChunk)
             || ProcessQueue.Dequeue(DequeuedChunk))
         {
             RecordStreamingResultDequeued(DequeuedChunk);
@@ -10104,6 +10134,10 @@ void AVoxelWorld::ProcessPendingChunks()
         if (bMeshApplied)
         {
             ++MeshesApplied;
+            if (bDequeuedSupplementalFar && ResultLevel >= 3)
+            {
+                ++SupplementalFarApplies;
+            }
             if (ResultLevel >= 0 && ResultLevel < 9)
             {
                 ++MeshesAppliedByLevel[ResultLevel];
@@ -10122,6 +10156,19 @@ void AVoxelWorld::ProcessPendingChunks()
         }
         if (MeshesApplied >= MaxApplies)
         {
+            const double DrainElapsedSeconds = FPlatformTime::ToSeconds64(
+                FPlatformTime::Cycles64() - DrainStartCycles);
+            const bool bCanSupplementFar = MeshesApplied == MaxApplies
+                && SupplementalFarApplies == 0
+                && !bUseStartupApplyBoost
+                && GVoxelForgeStreamingFarApplySupplement != 0
+                && ApplyBudgetSeconds > 0.0
+                && DrainElapsedSeconds < ApplyBudgetSeconds * 0.5;
+            if (bCanSupplementFar)
+            {
+                bForceSupplementalFarDequeue = true;
+                continue;
+            }
             break;
         }
     }
@@ -15360,6 +15407,8 @@ bool AVoxelWorld::ApplyMeshToTile(FChunkResult& Result)
     const int32 NumIndices = Result.NumTriangles * 3;
     const int32 FirstCapIndex = FMath::Clamp(
         NumIndices - Result.NumCeilingTriangles * 3, 0, NumIndices);
+    TFuture<ERealtimeMeshProxyUpdateStatus> ProxyPublicationFuture;
+    bool bHasProxyPublicationFuture = false;
     {
         TRACE_CPUPROFILER_EVENT_SCOPE(VoxelForge_RMC_CreateSections);
         if (bHasGroundTris)
@@ -15369,9 +15418,10 @@ bool AVoxelWorld::ApplyMeshToTile(FChunkResult& Result)
             const FRealtimeMeshStreamRange GroundRange =
                 URealtimeMeshBlueprintFunctionLibrary::MakeStreamRange(
                     0, Result.NumVertices, 0, FirstCapIndex);
-            RTMesh->CreateSection(
+            ProxyPublicationFuture = RTMesh->CreateSection(
                 FRealtimeMeshSectionKey::CreateForPolyGroup(GroupKey, 0),
                 GroundConfig, GroundRange, /*bShouldCreateCollision*/ bWantsCollision);
+            bHasProxyPublicationFuture = true;
         }
         if (bHasCeilingTris)
         {
@@ -15380,11 +15430,48 @@ bool AVoxelWorld::ApplyMeshToTile(FChunkResult& Result)
             const FRealtimeMeshStreamRange CapRange =
                 URealtimeMeshBlueprintFunctionLibrary::MakeStreamRange(
                     0, Result.NumVertices, FirstCapIndex, NumIndices);
-            RTMesh->CreateSection(
+            ProxyPublicationFuture = RTMesh->CreateSection(
                 FRealtimeMeshSectionKey::CreateForPolyGroup(GroupKey, 1),
                 CapConfig, CapRange, /*bShouldCreateCollision*/ bWantsCollision);
+            bHasProxyPublicationFuture = true;
         }
     }
+    const uint64 ProxyPublicationSubmitCycles = FPlatformTime::Cycles64();
+    if (bTraceRmcTiming && bHasProxyPublicationFuture)
+    {
+        const TWeakObjectPtr<AVoxelWorld> WeakWorld(this);
+        const TWeakObjectPtr<URealtimeMeshComponent> WeakComponent(MeshComp);
+        const FVoxelTileKey PublishedTile = Tile;
+        const uint64 RequestOrdinal = Result.RequestOrdinal;
+        const uint64 ApplyStartedCycles = ApplyStartCycles;
+        ProxyPublicationFuture.Next(
+            [WeakWorld, WeakComponent, PublishedTile, RequestOrdinal,
+             ApplyStartedCycles, ProxyPublicationSubmitCycles](
+                ERealtimeMeshProxyUpdateStatus Status)
+            {
+                if (!VoxelForgeStartupTrace::IsActive()) return;
+                const uint64 PublishedCycles = FPlatformTime::Cycles64();
+                const URealtimeMeshComponent* Component = WeakComponent.Get();
+                VoxelForgeStartupTrace::RecordEvent(TEXT("rmc_proxy_published"), FString::Printf(
+                    TEXT("\"tile\":[%d,%d,%d],\"level\":%d,\"request_ordinal\":%llu,"
+                         "\"proxy_update_status\":%d,\"apply_to_proxy_ms\":%.3f,"
+                         "\"section_submit_to_proxy_ms\":%.3f,\"component_visible\":%d,"
+                         "\"world_valid\":%d"),
+                    PublishedTile.Coord.X, PublishedTile.Coord.Y, PublishedTile.Coord.Z,
+                    PublishedTile.Level, static_cast<unsigned long long>(RequestOrdinal),
+                    static_cast<int32>(Status),
+                    ApplyStartedCycles != 0 && PublishedCycles >= ApplyStartedCycles
+                        ? FPlatformTime::ToSeconds64(PublishedCycles - ApplyStartedCycles) * 1000.0
+                        : 0.0,
+                    ProxyPublicationSubmitCycles != 0
+                            && PublishedCycles >= ProxyPublicationSubmitCycles
+                        ? FPlatformTime::ToSeconds64(
+                              PublishedCycles - ProxyPublicationSubmitCycles) * 1000.0
+                        : 0.0,
+                    Component && Component->IsVisible() ? 1 : 0,
+                    WeakWorld.IsValid() ? 1 : 0));
+            });
+    }
     if (bTraceRmcTiming)
     {
         TraceSectionConfigMilliseconds = FPlatformTime::ToSeconds64(

Source/VoxelForge/Public/VoxelStartupTrace.h

@@ -59,6 +59,7 @@ namespace VoxelForgeStartupTrace
         int32 ResultsDrained = 0;
         int32 ApplyResultsByLevelTotal = 0;
         int32 MeshesApplied = 0;
+        int32 SupplementalFarApplies = 0;
         int32 MaxApplies = 0;
         double ApplyBudgetMilliseconds = 0.0;
         bool bCountLimitReached = false;

Standalone/Core/include/VoxelForgeCore/PassageSpline.h

@@ -0,0 +1,255 @@
+#pragma once
+
+#include <algorithm>
+#include <array>
+#include <cmath>
+#include <cstddef>
+#include <limits>
+#include <span>
+#include <vector>
+
+namespace VoxelForgeCore::PassageSpline
+{
+    using FPoint = std::array<double,4>;
+    inline constexpr std::size_t ArcTableSteps = 32;
+    inline constexpr double UpperMouthFloorBlendRadiusMetres = 3.5;
+
+    struct FSegment
+    {
+        FPoint a{},b{};
+        double alongMetres=0.0,lengthMetres=0.0;
+        std::size_t controlSpan=0;
+        double tA=0.0,tB=1.0;
+        double plannedFloorA=0.0,plannedFloorB=0.0;
+    };
+
+    inline double KnotSpan(const FPoint& a,const FPoint& b,double unitsPerMetre)
+    {
+        const double dx=(b[0]-a[0])/unitsPerMetre;
+        const double dy=(b[1]-a[1])/unitsPerMetre;
+        return std::max(1.0e-6,std::sqrt(std::sqrt(dx*dx+dy*dy)));
+    }
+
+    inline FPoint EvaluateRaw(std::span<const FPoint> controls,std::size_t span,
+        double t,double unitsPerMetre)
+    {
+        const auto& p1=controls[span];
+        const auto& p2=controls[span+1];
+        const auto& p0=controls[span==0?0:span-1];
+        const auto& p3=controls[std::min(span+2,controls.size()-1)];
+        const double dt0=KnotSpan(p0,p1,unitsPerMetre);
+        const double dt1=KnotSpan(p1,p2,unitsPerMetre);
+        const double dt2=KnotSpan(p2,p3,unitsPerMetre);
+        const double t2=t*t,t3=t2*t;
+        FPoint result{};
+        for(const std::size_t axis:{0u,1u,3u})
+        {
+            const double m1=dt1*((p1[axis]-p0[axis])/dt0
+                -(p2[axis]-p0[axis])/(dt0+dt1)
+                +(p2[axis]-p1[axis])/dt1);
+            const double m2=dt1*((p2[axis]-p1[axis])/dt1
+                -(p3[axis]-p1[axis])/(dt1+dt2)
+                +(p3[axis]-p2[axis])/dt2);
+            result[axis]=(2.0*t3-3.0*t2+1.0)*p1[axis]
+                +(-2.0*t3+3.0*t2)*p2[axis]
+                +(t3-2.0*t2+t)*m1+(t3-t2)*m2;
+        }
+        result[2]=std::lerp(p1[2],p2[2],t);
+        result[3]=std::clamp(result[3],std::min(p1[3],p2[3]),
+            std::max(p1[3],p2[3]));
+        return result;
+    }
+
+    inline bool ClosestRouteHeightAtXY(std::span<const std::array<double,3>> route,
+        double x,double y,double& height,double& distanceSquared)
+    {
+        if(route.empty())return false;
+        distanceSquared=std::numeric_limits<double>::infinity();
+        const auto consider=[&](const std::array<double,3>& point)
+        {
+            const double dx=x-point[0],dy=y-point[1];
+            const double candidate=dx*dx+dy*dy;
+            if(candidate<distanceSquared)
+            {distanceSquared=candidate;height=point[2];}
+        };
+        if(route.size()==1)
+        {
+            consider(route.front());
+            return true;
+        }
+        for(std::size_t i=1;i<route.size();++i)
+        {
+            const auto& a=route[i-1];
+            const auto& b=route[i];
+            const double dx=b[0]-a[0],dy=b[1]-a[1];
+            const double lengthSquared=dx*dx+dy*dy;
+            const double t=lengthSquared>1.0e-12
+                ?std::clamp(((x-a[0])*dx+(y-a[1])*dy)/lengthSquared,0.0,1.0):0.0;
+            const std::array<double,3> point={a[0]+t*dx,a[1]+t*dy,
+                std::lerp(a[2],b[2],t)};
+            consider(point);
+        }
+        return std::isfinite(distanceSquared);
+    }
+
+    inline double BlendUpperMouthFloorAtXY(
+        std::span<const std::array<double,3>> receivingRoute,double x,double y,
+        double plannedFloor,double mouthX,double mouthY,double mouthFloor,
+        double tangentX,double tangentY,
+        double blendRadiusMetres,double playerCenterAboveFloorMetres,
+        double unitsPerMetre=1.0)
+    {
+        if(receivingRoute.empty()||!(blendRadiusMetres>0.0)
+            ||!(unitsPerMetre>0.0))return plannedFloor;
+        const double dx=(x-mouthX)/unitsPerMetre;
+        const double dy=(y-mouthY)/unitsPerMetre;
+        if(dx*tangentX+dy*tangentY>=0.0)return plannedFloor;
+        const double dz=(plannedFloor-mouthFloor)/unitsPerMetre;
+        const double distance=std::sqrt(dx*dx+dy*dy+dz*dz);
+        double routeHeight=0.0,routeDistanceSquared=0.0;
+        if(distance>blendRadiusMetres
+            ||!ClosestRouteHeightAtXY(receivingRoute,x,y,routeHeight,
+                routeDistanceSquared)
+            ||routeDistanceSquared>std::pow(2.0*blendRadiusMetres*unitsPerMetre,2.0))
+            return plannedFloor;
+        const double t=std::clamp(distance/blendRadiusMetres,0.0,1.0);
+        const double blend=t*t*(3.0-2.0*t);
+        const double receivingFloor=routeHeight
+            -playerCenterAboveFloorMetres*unitsPerMetre;
+        return std::lerp(plannedFloor,receivingFloor,blend);
+    }
+
+    inline std::vector<FSegment> BuildSegments(std::span<const FPoint> controls,
+        std::size_t spanCount,double spacingMetres,double unitsPerMetre=1.0,
+        std::span<const double> plannedFloorProfile={},
+        double maximumChordErrorMetres=0.0)
+    {
+        std::vector<FSegment> result;
+        if(controls.size()<2||!(spacingMetres>0.0)||!(unitsPerMetre>0.0))return result;
+        spanCount=std::min(spanCount,controls.size()-1);
+        double alongMetres=0.0;
+        for(std::size_t span=0;span<spanCount;++span)
+        {
+            std::array<double,ArcTableSteps+1> horizontalArc{};
+            auto previous=EvaluateRaw(controls,span,0.0,unitsPerMetre);
+            for(std::size_t step=1;step<=ArcTableSteps;++step)
+            {
+                const auto current=EvaluateRaw(controls,span,
+                    static_cast<double>(step)/ArcTableSteps,unitsPerMetre);
+                const double dx=(current[0]-previous[0])/unitsPerMetre;
+                const double dy=(current[1]-previous[1])/unitsPerMetre;
+                horizontalArc[step]=horizontalArc[step-1]+std::sqrt(dx*dx+dy*dy);
+                previous=current;
+            }
+            const double spanLength=horizontalArc.back();
+            const std::size_t baseSegmentCount=std::max<std::size_t>(1,
+                static_cast<std::size_t>(std::ceil(spanLength/spacingMetres)));
+            std::vector<FPoint> points;
+            std::vector<double> parameters;
+            points.reserve(baseSegmentCount+1);
+            parameters.reserve(baseSegmentCount+1);
+            const auto parameterAtArc=[&](double targetArc)
+            {
+                const auto upper=std::lower_bound(horizontalArc.begin()+1,
+                    horizontalArc.end(),targetArc);
+                const std::size_t upperIndex=static_cast<std::size_t>(
+                    upper-horizontalArc.begin());
+                const std::size_t lowerIndex=upperIndex-1;
+                const double arcSpan=horizontalArc[upperIndex]-horizontalArc[lowerIndex];
+                const double arcT=arcSpan>1.0e-12
+                    ?std::clamp((targetArc-horizontalArc[lowerIndex])/arcSpan,0.0,1.0):0.0;
+                return (static_cast<double>(lowerIndex)+arcT)/ArcTableSteps;
+            };
+            parameters.push_back(0.0);
+            const double chordError=std::max(0.0,maximumChordErrorMetres);
+            const auto appendAdaptive=[&](auto&& self,double tA,double tB,
+                const FPoint& a,const FPoint& b,unsigned depth)->void
+            {
+                constexpr std::array<double,3> CheckFractions={0.25,0.5,0.75};
+                double maximumError=0.0,polylineLength=0.0;
+                FPoint previous=a;
+                for(const double fraction:CheckFractions)
+                {
+                    const FPoint current=EvaluateRaw(controls,span,
+                        std::lerp(tA,tB,fraction),unitsPerMetre);
+                    const double chordX=std::lerp(a[0],b[0],fraction);
+                    const double chordY=std::lerp(a[1],b[1],fraction);
+                    const double dx=(current[0]-chordX)/unitsPerMetre;
+                    const double dy=(current[1]-chordY)/unitsPerMetre;
+                    const double dr=(current[3]-std::lerp(a[3],b[3],fraction))
+                        /unitsPerMetre;
+                    maximumError=std::max(maximumError,
+                        std::sqrt(dx*dx+dy*dy+dr*dr));
+                    const double segmentX=(current[0]-previous[0])/unitsPerMetre;
+                    const double segmentY=(current[1]-previous[1])/unitsPerMetre;
+                    polylineLength+=std::sqrt(segmentX*segmentX+segmentY*segmentY);
+                    previous=current;
+                }
+                const double finalX=(b[0]-previous[0])/unitsPerMetre;
+                const double finalY=(b[1]-previous[1])/unitsPerMetre;
+                polylineLength+=std::sqrt(finalX*finalX+finalY*finalY);
+                const bool split=(polylineLength>spacingMetres*(1.0+1.0e-9)
+                    ||(chordError>0.0&&maximumError>chordError))&&depth<12;
+                if(split)
+                {
+                    const double tMid=(tA+tB)*0.5;
+                    const FPoint middle=EvaluateRaw(controls,span,tMid,unitsPerMetre);
+                    self(self,tA,tMid,a,middle,depth+1);
+                    self(self,tMid,tB,middle,b,depth+1);
+                    return;
+                }
+                parameters.push_back(tB);
+            };
+            for(std::size_t segment=1;segment<=baseSegmentCount;++segment)
+            {
+                const double targetA=spanLength*static_cast<double>(segment-1)
+                    /baseSegmentCount;
+                const double targetB=spanLength*static_cast<double>(segment)
+                    /baseSegmentCount;
+                const double tA=parameterAtArc(targetA);
+                const double tB=parameterAtArc(targetB);
+                const FPoint a=EvaluateRaw(controls,span,tA,unitsPerMetre);
+                const FPoint b=EvaluateRaw(controls,span,tB,unitsPerMetre);
+                appendAdaptive(appendAdaptive,tA,tB,a,b,0);
+            }
+            points.reserve(parameters.size());
+            for(const double t:parameters)
+                points.push_back(EvaluateRaw(controls,span,t,unitsPerMetre));
+            std::vector<double> sampledHorizontalArc(points.size(),0.0);
+            for(std::size_t i=1;i<points.size();++i)
+            {
+                const double dx=(points[i][0]-points[i-1][0])/unitsPerMetre;
+                const double dy=(points[i][1]-points[i-1][1])/unitsPerMetre;
+                sampledHorizontalArc[i]=sampledHorizontalArc[i-1]
+                    +std::sqrt(dx*dx+dy*dy);
+            }
+            const double sampledSpanLength=sampledHorizontalArc.back();
+            for(std::size_t i=0;i<points.size();++i)
+            {
+                const double progress=sampledSpanLength>1.0e-12
+                    ?sampledHorizontalArc[i]/sampledSpanLength:parameters[i];
+                points[i][2]=std::lerp(controls[span][2],controls[span+1][2],progress);
+            }
+            for(std::size_t i=1;i<points.size();++i)
+            {
+                const double dx=(points[i][0]-points[i-1][0])/unitsPerMetre;
+                const double dy=(points[i][1]-points[i-1][1])/unitsPerMetre;
+                const double dz=(points[i][2]-points[i-1][2])/unitsPerMetre;
+                const double length=std::sqrt(dx*dx+dy*dy+dz*dz);
+                const auto floorAt=[&](std::size_t index)
+                {
+                    if(plannedFloorProfile.size()!=controls.size())
+                        return points[index][2];
+                    const double progress=sampledSpanLength>1.0e-12
+                        ?sampledHorizontalArc[index]/sampledSpanLength:parameters[index];
+                    return std::lerp(plannedFloorProfile[span],
+                        plannedFloorProfile[span+1],progress);
+                };
+                result.push_back({points[i-1],points[i],alongMetres,length,span,
+                    parameters[i-1],parameters[i],floorAt(i-1),floorAt(i)});
+                alongMetres+=length;
+            }
+        }
+        return result;
+    }
+}

Standalone/Core/include/VoxelForgeCore/VoxelForgeCore.h

@@ -12,6 +12,7 @@
 #include <vector>
 
 #include "VoxelForgeCore/Linux77Profiler.h"
+#include "VoxelForgeCore/PassageSpline.h"
 
 // UBT supplies this export/import macro for the runtime module. The standalone target has no
 // module boundary, so leave it empty there. UBT fournit la macro; le viewer autonome reste neutre.
@@ -137,46 +138,44 @@ namespace VoxelForgeCore
         bool passageComposerProfiled = false;
         std::uint64_t passageTileFeatureCount = 0;
         std::uint64_t passageTileRouteSegmentCandidates = 0;
-        std::uint64_t passageTileMouthSegmentCandidates = 0;
         std::uint64_t resolvedBoundaryBandQueries = 0;
         std::uint64_t passageComposerQueries = 0;
         std::uint64_t passageIterations = 0;
-        std::uint64_t passageTileBoundChecks = 0;
-        std::uint64_t passageTileBoundRejects = 0;
-        std::uint64_t passageTileBoundSkippedIterations = 0;
-        std::uint64_t passageUpperMouthFilterChecks = 0;
-        std::uint64_t passageColumnCandidates = 0;
-        std::uint64_t passageColumnDistanceCalls = 0;
-        std::uint64_t passageNegativeBoundChecks = 0;
-        std::uint64_t passageNegativeBoundRejects = 0;
-        std::uint64_t passagePositiveBoundChecks = 0;
-        std::uint64_t passagePositiveBoundRejects = 0;
         std::uint64_t passageCarveSegmentVisits = 0;
-        std::uint64_t passageSmoothCarveJoins = 0;
-        std::uint64_t passageSupportSegmentVisits = 0;
-        std::uint64_t passageSmoothSupportJoins = 0;
-        std::uint64_t passageSupportCapsuleChecks = 0;
-        std::uint64_t passageLaterBoreChecks = 0;
-        std::uint64_t passageOtherCapsuleChecks = 0;
-        std::uint64_t passageMouthSupportCalls = 0;
-        std::uint64_t passageMouthRouteSegmentChecks = 0;
         std::uint64_t passageReliefFieldCalls = 0;
-        std::uint64_t passageSmoothstepEvaluations = 0;
         // L99 tile-local swept-tube base and detail passes. These are populated only
         // by the opt-in passage composer profile, except the mesh composition itself.
         std::uint64_t passageSweepRouteSegmentCandidates = 0;
         std::uint64_t passageSweepSegmentSamplePairs = 0;
+        std::uint64_t passageSweepPrepareNanoseconds = 0;
         std::uint64_t passageSweepOutsideInfluencePairs = 0;
+        std::uint64_t passageSweepShellSupportSkips = 0;
         std::uint64_t passageSweepMinimumUpdates = 0;
         std::uint64_t passageSweepBaseInfluenceSamples = 0;
         std::uint64_t passageSweepDetailShellSamples = 0;
         std::uint64_t passageSweepGridSamples = 0;
         std::uint64_t passageSweepNoInfluenceGridSamples = 0;
+        std::uint64_t passageSweepTouchedSamples = 0;
+        std::uint64_t passageSweepStampedOutsideSamples = 0;
+        std::uint64_t passageSweepCacheValidSamples = 0;
+        std::uint64_t passageSweepCacheValidUniqueSamples = 0;
+        std::uint64_t passageSweepUniqueTouchedSamples = 0;
+        std::uint64_t passageSweepBodyBandSamples = 0;
+        std::uint64_t passageSweepShellBandSamples = 0;
+        std::uint64_t passageSweepUniqueBandSamples = 0;
+        std::uint64_t passageSweepOverlapPairs = 0;
+        std::uint64_t passageSweepContributingPairs = 0;
+        std::uint64_t passageSweepCrossPassageOverlaps = 0;
+        std::uint64_t passageSweepFloorProfileCalls = 0;
+        std::uint64_t passageSweepAllocatedBytes = 0;
+        std::uint64_t passageSweepPrepareScratchBytes = 0;
         std::uint64_t passageSweepLookups = 0;
         std::uint64_t passageSweepCacheHits = 0;
         std::uint64_t passageSweepBaseInfluenceHits = 0;
         std::uint64_t passageSweepOutOfInfluenceHits = 0;
         std::uint64_t passageSweepMissingSamples = 0;
+        std::uint64_t passageSweepOffGridMisses = 0;
+        std::uint64_t passageSweepUnpreparedMisses = 0;
         std::array<std::uint64_t,12> passageSweepMissingLookupsByCaller{};
         std::array<std::uint64_t,12> passageSweepExactRouteFallbacksByCaller{};
         std::array<std::uint64_t,12> passageSweepFeatureBoundSkipsByCaller{};
@@ -488,7 +487,19 @@ namespace VoxelForgeCore
     // Version 210 keeps bulk native-cave sampling on the scalar field path at strata edges.
     // Version 211 makes coordinate biasing in the shared cell hash unsigned, removing signed-overflow UB.
     // Version 212 balances generated entry routes against existing walk lanes and preserves their shared treads.
-    inline constexpr std::uint32_t WorldAssemblyEvaluatorVersion = 216;
+    // Version 218 sweeps passages around a dense smooth spline and limits exposed shells to a trough.
+    // Version 219 smooths passage XY while mapping each planned height chord over its sampled horizontal arc.
+    // Version 220 adaptively samples passage curves and rounds open-air wall crests.
+    // Version 221 ranks passage segments before floor shaping and preserves their signed exterior distance.
+    // Version 223 projects the floor onto one continuous arc-length profile and groups
+    // dense passage samples into coarse tile candidates. Version 224 keeps the same
+    // continuous floor while relaxing the spline chord tolerance and bounding shell work.
+    // Version 226 unifies the passage influence reach in metres and converts it to core
+    // voxels for every broadphase and proof bound. Version 227 uses the exact smooth-max
+    // support width/6 when deriving that reach. Version 228 composes deterministic cave
+    // graphs from fixed passage mouths and the owner Tunnel Network settings. Version 229
+    // budgets cave floor support; 230 records a shared backbone floor profile and joint cap.
+    inline constexpr std::uint32_t WorldAssemblyEvaluatorVersion = 230;
 
     enum class FieldChannel : std::uint32_t
     {
@@ -883,6 +894,7 @@ namespace VoxelForgeCore
     inline constexpr std::uint32_t ResolvedLayoutStampVocabularyVersionV2 = 2;
     inline constexpr std::uint32_t ResolvedLayoutStampVocabularyVersionV3 = 3;
     inline constexpr std::uint32_t ResolvedLayoutCavernGraphVocabularyVersionV1 = 1;
+    inline constexpr std::uint32_t ResolvedLayoutCavernGraphVocabularyVersionV2 = 2;
     inline constexpr std::uint32_t ResolvedLayoutOreVeinVocabularyVersionV1 = 1;
     enum class ResolvedLayoutValidationModeV3 : std::uint8_t { compatibility, strict };
 
@@ -947,6 +959,11 @@ namespace VoxelForgeCore
         std::array<std::int64_t, 3> centerMicrometres{};
         std::array<std::int64_t, 3> halfExtentsMicrometres{};
         std::int64_t floorMicrometres = 0;
+        // Optional shared backbone profile reference. The room floor plane is
+        // floorMicrometres plus this Q16 XY gradient from centerMicrometres.
+        std::string floorProfileStableId;
+        std::int64_t floorProfileDistanceMicrometres = -1;
+        std::array<std::int32_t, 2> floorGradientQ16{};
         std::int64_t smoothWidthMicrometres = 0;
         // Optional feature-to-feature union width. Zero preserves the historical
         // behavior where smoothWidthMicrometres also controls the graph union.
@@ -966,6 +983,11 @@ namespace VoxelForgeCore
         std::int32_t strateIndex = -1;
         // A single cubic Bezier in XY. Its four Z values are the cubic floor profile.
         std::array<std::array<std::int64_t, 3>, 4> controlPointsMicrometres{};
+        // Edges with the same profile ID are adjacent pieces of one shared h(s).
+        // Distances are signed along that profile, so a terminal edge may be reversed.
+        std::string floorProfileStableId;
+        std::int64_t floorProfileStartDistanceMicrometres = -1;
+        std::int64_t floorProfileEndDistanceMicrometres = -1;
         // Resolve-time, error-bounded approximation of the cubic centerline. Each
         // consecutive pair is a physical segment used by the native cave field.
         std::vector<std::array<std::int64_t, 3>> polylineMicrometres;
@@ -982,6 +1004,37 @@ namespace VoxelForgeCore
         std::array<std::int64_t, 3> boundsMaximumMicrometres{};
     };
 
+    // Inputs to the reusable, plan-time cave graph composer. All distances are
+    // integer micrometres and unit-interval values are Q16 so the resolved graph
+    // does not depend on worker count or floating-point random distributions.
+    struct ResolvedCavernPlannerSettingsV1
+    {
+        std::int64_t roomSpacingMicrometres = 16000000;
+        std::int64_t minimumRoomRadiusMicrometres = 2000000;
+        std::int64_t maximumRoomRadiusMicrometres = 5000000;
+        std::int64_t destinationRadiusMicrometres = 6000000;
+        // Solid floor backing below the planned edge floor, preserved after graph
+        // unions and passage composition. This is separate from bore height.
+        std::int64_t floorSupportDepthMicrometres = 750000;
+        std::int64_t minimumTunnelRadiusMicrometres = 750000;
+        std::int64_t maximumTunnelRadiusMicrometres = 1000000;
+        std::int64_t maximumTunnelLengthMicrometres = 45000000;
+        std::int64_t tunnelWarpMicrometres = 0;
+        std::int64_t caveWarpMicrometres = 0;
+        std::int64_t smoothWidthMicrometres = 500000;
+        std::int64_t irregularityAmplitudeMicrometres = 0;
+        std::int64_t irregularityWavelengthMicrometres = 0;
+        std::int32_t roomDensityQ16 = 22938;
+        std::int32_t tunnelDensityQ16 = 26214;
+        std::int32_t roomShapeVarietyQ16 = 32768;
+        std::int32_t roomHeightRatioQ16 = 55706;
+        std::int32_t roomFloorCutMinimumQ16 = 45875;
+        std::int32_t roomFloorCutMaximumQ16 = 58982;
+        std::int32_t tunnelHorizontalBiasQ16 = 32768;
+        std::int32_t originRoomMaximumConnections = 4;
+        bool tunnelsFlowTowardOrigin = true;
+    };
+
     // Immutable material-only deposits. The capsule and its conservative bounds are in
     // micrometres; bounds include irregularity and the blend envelope.
     struct ResolvedLayoutOreVeinRequestV1
@@ -1041,7 +1094,6 @@ namespace VoxelForgeCore
     {
         bool hasSurfaceLanding = false;
         bool hasFloatingIslandLanding = false;
-        bool usePassageAnchorForColumns = false;
         bool hasNativeCaveField = false;
         bool hasNativeLatticeField = false;
     };
@@ -1087,7 +1139,7 @@ namespace VoxelForgeCore
 
     inline bool IsNativeCaveFieldStratum(const ResolvedLayoutStratumV2& stratum) noexcept
     {
-        return stratum.generatorKind=="Caves"
+        return (stratum.generatorKind=="Caves"||stratum.generatorKind=="TunnelNetwork")
             &&stratum.recipeId=="Core.Caves.PrimitiveField"
             &&stratum.recipeVersion==1;
     }
@@ -1145,28 +1197,6 @@ namespace VoxelForgeCore
         std::vector<ResolvedLayoutPassagePointV1> controlPoints;
     };
 
-    struct PassageVoidCrossingAuditV1
-    {
-        std::size_t passageIndex = 0;
-        std::size_t firstSegment = 0;
-        std::size_t lastSegment = 0;
-        std::uint64_t unsupportedSamples = 0;
-        double unsupportedLengthMetres = 0.0;
-        std::array<double, 4> startPointSubvoxels{};
-        std::array<double, 4> endPointSubvoxels{};
-        bool needsColumn = false;
-        bool connectedAtStart = false;
-        bool connectedAtEnd = false;
-        std::array<std::int64_t, 2> columnAxisSubvoxels{};
-        std::int64_t columnTopZSubvoxels = 0;
-        std::int64_t columnBottomZSubvoxels = 0;
-        std::int64_t columnMinimumRadiusSubvoxels = 0;
-        std::int64_t columnNominalRadiusSubvoxels = 0;
-        std::int64_t columnMaximumRadiusSubvoxels = 0;
-        double maximumRouteOffsetMetres = 0.0;
-        double maximumBoreRadiusMetres = 0.0;
-    };
-
     struct ResolvedLayoutPassageConfigV1
     {
         std::int32_t upperStrateIndex = -1;
@@ -1270,6 +1300,17 @@ namespace VoxelForgeCore
         std::span<const ResolvedCavernRoomV1> rooms,
         std::span<const ResolvedCavernEdgeV1> edges,
         std::string& error);
+    // Build a deterministic, walkable room-and-tunnel graph between the two
+    // passage mouths that terminate in strateIndex. Compound rooms are emitted
+    // as overlapping room primitives; the route is planned below a 15 degree
+    // walking grade and every corridor carries a 4.5 m vertical clearance.
+    VFCORE_API bool PlanResolvedCavernNetworkV1(
+        const ResolvedLayoutSnapshotV3& layout, std::int32_t seed,
+        std::int32_t strateIndex,
+        const ResolvedCavernPlannerSettingsV1& settings,
+        std::vector<ResolvedCavernRoomV1>& rooms,
+        std::vector<ResolvedCavernEdgeV1>& edges,
+        std::string& error);
     VFCORE_API bool ResolveResolvedLayoutOreVeinsV1(
         ResolvedLayoutSnapshotV3& layout, std::int32_t seed,
         std::span<const ResolvedLayoutOreVeinRequestV1> requests,
@@ -1836,10 +1877,6 @@ namespace VoxelForgeCore
         bool profileMesherStages = false;
         // Research-only: count passage-composer work per generated tile.
         bool profilePassageComposer = false;
-        // Research-only field comparison. A nonnegative value clamps the source density
-        // entering passage composition to at least -N voxel units for this tile.
-        // A negative value leaves the process-level diagnostic override in control.
-        double diagnosticPassageSourceFloorVoxels = -1.0;
         // Research-only field comparison. Zero keeps the unbounded passage carve;
         // a positive value bounds its signed-distance reach in metres. A negative
         // value leaves the process-level L93 setting in control.
@@ -1932,12 +1969,6 @@ namespace VoxelForgeCore
         OreMiningYieldV1 measureOreMiningYieldForSphere(
             Vec3d centerVoxel, double radiusMetres,
             double sampleSpacingMetres = 0.25) const;
-        // Audit-only view of density-field columns and short spans supporting passage voids.
-        std::vector<PassageVoidCrossingAuditV1> passageVoidCrossings() const;
-        bool isPassageColumnSample(std::size_t passageIndex, std::size_t segmentIndex,
-                                  Vec3d voxelPosition) const;
-        bool isPassageShortSpanSample(std::size_t passageIndex, std::size_t segmentIndex,
-                                     Vec3d voxelPosition) const;
         // Audit-only selector provenance for the same point evaluator; never used by meshing.
         std::string densityDispatchPathAt(Vec3d voxelPosition) const;
         // Analytic source distance without the occupancy-preserving near-zero clamp or packed proxy.

Standalone/Core/src/ResolvedLayoutInternal.h

@@ -621,6 +621,10 @@ namespace
                     <<"],\"half_extents_um\":["<<room.halfExtentsMicrometres[0]<<','
                     <<room.halfExtentsMicrometres[1]<<','<<room.halfExtentsMicrometres[2]
                     <<"],\"floor_um\":"<<room.floorMicrometres
+                    <<",\"floor_profile_id\":"<<LayoutEscape(room.floorProfileStableId)
+                    <<",\"floor_profile_distance_um\":"<<room.floorProfileDistanceMicrometres
+                    <<",\"floor_gradient_q16\":["<<room.floorGradientQ16[0]<<','
+                    <<room.floorGradientQ16[1]<<']'
                     <<",\"smooth_width_um\":"<<room.smoothWidthMicrometres
                     ;
                 if(room.unionWidthMicrometres>0)
@@ -647,6 +651,11 @@ namespace
                     out<<'['<<p[0]<<','<<p[1]<<','<<p[2]<<']';
                 }
                 out<<']';
+                out<<",\"floor_profile_id\":"<<LayoutEscape(edge.floorProfileStableId)
+                    <<",\"floor_profile_start_distance_um\":"
+                    <<edge.floorProfileStartDistanceMicrometres
+                    <<",\"floor_profile_end_distance_um\":"
+                    <<edge.floorProfileEndDistanceMicrometres;
                 if(!edge.polylineMicrometres.empty())
                 {
                     out<<",\"polyline_um\":[";
@@ -888,10 +897,7 @@ bool ResolvedLayoutSnapshotV3::finalize(std::string& error)
         else if(IsNativeLatticeFieldStratum(x))
             x.passageCapabilities.hasNativeLatticeField=true;
         else if(x.generatorKind=="FloatingIslands")
-        {
             x.passageCapabilities.hasFloatingIslandLanding=true;
-            x.passageCapabilities.usePassageAnchorForColumns=true;
-        }
         if(i+1<strates.size())
         {
             const auto& lower=strates[i+1];
@@ -1238,7 +1244,8 @@ bool ResolvedLayoutSnapshotV3::validate(ResolvedLayoutValidationModeV3 mode,
             add("cavern_graph_vocabulary_version",
                 "cavern graph vocabulary version is set without graph records");
     }
-    else if(cavernGraphVocabularyVersion!=ResolvedLayoutCavernGraphVocabularyVersionV1)
+    else if(cavernGraphVocabularyVersion!=ResolvedLayoutCavernGraphVocabularyVersionV1
+        &&cavernGraphVocabularyVersion!=ResolvedLayoutCavernGraphVocabularyVersionV2)
         add("cavern_graph_vocabulary_version",
             "cavern graph records require vocabulary version 1");
     if(cavernGraphVocabularyVersion!=0)
@@ -1271,6 +1278,16 @@ bool ResolvedLayoutSnapshotV3::validate(ResolvedLayoutValidationModeV3 mode,
                 ||(room.irregularityAmplitudeMicrometres>0
                     &&room.irregularityWavelengthMicrometres<=0))
                 add("cavern_room_parameters",room.stableId+" has invalid or unbounded parameters");
+            const std::int64_t gradientSquared=static_cast<std::int64_t>(room.floorGradientQ16[0])
+                *room.floorGradientQ16[0]+static_cast<std::int64_t>(room.floorGradientQ16[1])
+                *room.floorGradientQ16[1];
+            if((room.floorProfileStableId.empty()
+                    &&room.floorProfileDistanceMicrometres!=-1)
+                ||(!room.floorProfileStableId.empty()
+                    &&room.floorProfileDistanceMicrometres<0)
+                ||gradientSquared>static_cast<std::int64_t>(13108)*13108)
+                add("cavern_room_floor_profile",room.stableId+
+                    " has an invalid shared profile reference or floor grade");
             const std::int64_t pad=room.smoothWidthMicrometres
                 +room.unionWidthMicrometres+room.irregularityAmplitudeMicrometres;
             for(std::size_t axis=0;axis<3;++axis)
@@ -1311,6 +1328,15 @@ bool ResolvedLayoutSnapshotV3::validate(ResolvedLayoutValidationModeV3 mode,
                 ||(edge.irregularityAmplitudeMicrometres>0
                     &&edge.irregularityWavelengthMicrometres<=0))
                 add("cavern_edge_parameters",edge.stableId+" has invalid or unbounded parameters");
+            const bool hasFloorProfile=!edge.floorProfileStableId.empty();
+            if((hasFloorProfile&&(edge.floorProfileStartDistanceMicrometres<0
+                    ||edge.floorProfileEndDistanceMicrometres<0
+                    ||edge.floorProfileStartDistanceMicrometres
+                        ==edge.floorProfileEndDistanceMicrometres))
+                ||(!hasFloorProfile&&(edge.floorProfileStartDistanceMicrometres!=-1
+                    ||edge.floorProfileEndDistanceMicrometres!=-1)))
+                add("cavern_edge_floor_profile",edge.stableId+
+                    " has an invalid shared profile distance interval");
             std::array<std::int64_t,3> low=edge.controlPointsMicrometres[0];
             std::array<std::int64_t,3> high=low;
             if(nativeLattice&&!edge.polylineMicrometres.empty())
@@ -1366,6 +1392,114 @@ bool ResolvedLayoutSnapshotV3::validate(ResolvedLayoutValidationModeV3 mode,
                 add("cavern_edge_polyline",edge.stableId+
                     " has no bounded resolve-time polyline with matching endpoints");
         }
+        std::vector<std::string> floorProfileIds;
+        for(const auto& edge:cavernEdges)
+            if(!edge.floorProfileStableId.empty()
+                &&std::find(floorProfileIds.begin(),floorProfileIds.end(),
+                    edge.floorProfileStableId)==floorProfileIds.end())
+                floorProfileIds.push_back(edge.floorProfileStableId);
+        for(const std::string& profileId:floorProfileIds)
+        {
+            std::vector<const ResolvedCavernEdgeV1*> profileEdges;
+            for(const auto& edge:cavernEdges)
+                if(edge.floorProfileStableId==profileId)profileEdges.push_back(&edge);
+            std::sort(profileEdges.begin(),profileEdges.end(),[](const auto* a,const auto* b)
+            {
+                return std::min(a->floorProfileStartDistanceMicrometres,
+                    a->floorProfileEndDistanceMicrometres)
+                    <std::min(b->floorProfileStartDistanceMicrometres,
+                        b->floorProfileEndDistanceMicrometres);
+            });
+            const auto endpointZ=[](const ResolvedCavernEdgeV1& edge,bool atMinimum)
+            {
+                const bool forward=edge.floorProfileStartDistanceMicrometres
+                    <edge.floorProfileEndDistanceMicrometres;
+                const std::size_t point=(atMinimum==forward)?0:3;
+                return edge.controlPointsMicrometres[point][2];
+            };
+            const auto endpointSlope=[](const ResolvedCavernEdgeV1& edge,bool atMinimum)
+            {
+                const double span=static_cast<double>(edge.floorProfileEndDistanceMicrometres
+                    -edge.floorProfileStartDistanceMicrometres);
+                if(span==0.0)return std::numeric_limits<double>::infinity();
+                const bool forward=edge.floorProfileStartDistanceMicrometres
+                    <edge.floorProfileEndDistanceMicrometres;
+                const std::size_t control=atMinimum==forward?0:2;
+                return 3.0*(edge.controlPointsMicrometres[control+1][2]
+                    -edge.controlPointsMicrometres[control][2])/span;
+            };
+            const ResolvedCavernEdgeV1* previous=nullptr;
+            for(const auto* edge:profileEdges)
+            {
+                const std::int64_t low=std::min(edge->floorProfileStartDistanceMicrometres,
+                    edge->floorProfileEndDistanceMicrometres);
+                const std::int64_t high=std::max(edge->floorProfileStartDistanceMicrometres,
+                    edge->floorProfileEndDistanceMicrometres);
+                const double slope0=endpointSlope(*edge,true);
+                const double slope1=endpointSlope(*edge,false);
+                if(std::max(std::abs(slope0),std::abs(slope1))>0.200001)
+                    add("cavern_floor_profile_grade",edge->stableId+
+                        " exceeds the shared backbone 1:5 grade limit");
+                if(previous)
+                {
+                    const std::int64_t previousHigh=std::max(
+                        previous->floorProfileStartDistanceMicrometres,
+                        previous->floorProfileEndDistanceMicrometres);
+                    if(low!=previousHigh
+                        ||std::abs(endpointZ(*previous,false)-endpointZ(*edge,true))>1000
+                        ||std::abs(endpointSlope(*previous,false)-slope0)>0.000001)
+                        add("cavern_floor_profile_joint",profileId+
+                            " has a gap, height step, or slope step between adjacent edges");
+                }
+                previous=edge;
+                (void)high;
+            }
+            for(const auto& room:cavernRooms)
+            {
+                if(room.floorProfileStableId!=profileId)continue;
+                const ResolvedCavernEdgeV1* containing=nullptr;
+                for(const auto* edge:profileEdges)
+                {
+                    const auto low=std::min(edge->floorProfileStartDistanceMicrometres,
+                        edge->floorProfileEndDistanceMicrometres);
+                    const auto high=std::max(edge->floorProfileStartDistanceMicrometres,
+                        edge->floorProfileEndDistanceMicrometres);
+                    if(room.floorProfileDistanceMicrometres>=low
+                        &&room.floorProfileDistanceMicrometres<=high)
+                    {containing=edge;break;}
+                }
+                if(!containing)
+                {
+                    add("cavern_room_floor_profile",room.stableId+
+                        " does not lie on its referenced continuous floor profile");
+                    continue;
+                }
+                const auto& edge=*containing;
+                const double t=static_cast<double>(room.floorProfileDistanceMicrometres
+                    -edge.floorProfileStartDistanceMicrometres)
+                    /(edge.floorProfileEndDistanceMicrometres
+                        -edge.floorProfileStartDistanceMicrometres);
+                const double one=1.0-t;
+                const double expected=one*one*one*edge.controlPointsMicrometres[0][2]
+                    +3.0*one*one*t*edge.controlPointsMicrometres[1][2]
+                    +3.0*one*t*t*edge.controlPointsMicrometres[2][2]
+                    +t*t*t*edge.controlPointsMicrometres[3][2];
+                std::array<double,2> routeXY{};
+                for(std::size_t axis=0;axis<2;++axis)
+                    routeXY[axis]=one*one*one*edge.controlPointsMicrometres[0][axis]
+                        +3.0*one*one*t*edge.controlPointsMicrometres[1][axis]
+                        +3.0*one*t*t*edge.controlPointsMicrometres[2][axis]
+                        +t*t*t*edge.controlPointsMicrometres[3][axis];
+                const double roomFloorAtProfile=room.floorMicrometres
+                    +room.floorGradientQ16[0]/65536.0
+                        *(routeXY[0]-room.centerMicrometres[0])
+                    +room.floorGradientQ16[1]/65536.0
+                        *(routeXY[1]-room.centerMicrometres[1]);
+                if(std::abs(expected-roomFloorAtProfile)>1000.0)
+                    add("cavern_room_floor_profile",room.stableId+
+                        " floor does not match its referenced h(s)");
+            }
+        }
     }
     if(oreVeins.empty())
     {
@@ -1529,6 +1663,12 @@ std::optional<ResolvedLayoutSnapshotV3> ResolvedLayoutSnapshotV3::FromJson(
         readTriplet("bounds_min_um",room.boundsMinimumMicrometres);
         readTriplet("bounds_max_um",room.boundsMaximumMicrometres);
         room.floorMicrometres=LayoutInt(v,"floor_um");
+        room.floorProfileStableId=LayoutString(v,"floor_profile_id");
+        if(v.Get("floor_profile_distance_um").IsNumber())
+            room.floorProfileDistanceMicrometres=LayoutInt(v,"floor_profile_distance_um");
+        const auto& floorGradient=v.Get("floor_gradient_q16").AsArray();
+        for(std::size_t axis=0;axis<2&&axis<floorGradient.size();++axis)
+            room.floorGradientQ16[axis]=static_cast<std::int32_t>(floorGradient[axis].Integer());
         room.smoothWidthMicrometres=LayoutInt(v,"smooth_width_um");
         room.unionWidthMicrometres=LayoutInt(v,"union_width_um");
         room.irregularityAmplitudeMicrometres=LayoutInt(v,"irregularity_amplitude_um");
@@ -1542,6 +1682,13 @@ std::optional<ResolvedLayoutSnapshotV3> ResolvedLayoutSnapshotV3::FromJson(
         edge.startAnchorStableId=LayoutString(v,"start_anchor_stable_id");
         edge.endRoomStableId=LayoutString(v,"end_room_stable_id");
         edge.strateIndex=static_cast<std::int32_t>(LayoutInt(v,"strate_index"));
+        edge.floorProfileStableId=LayoutString(v,"floor_profile_id");
+        if(v.Get("floor_profile_start_distance_um").IsNumber())
+            edge.floorProfileStartDistanceMicrometres=
+                LayoutInt(v,"floor_profile_start_distance_um");
+        if(v.Get("floor_profile_end_distance_um").IsNumber())
+            edge.floorProfileEndDistanceMicrometres=
+                LayoutInt(v,"floor_profile_end_distance_um");
         const auto& points=v.Get("control_points_um").AsArray();
         for(std::size_t i=0;i<edge.controlPointsMicrometres.size()&&i<points.size();++i)
         {
@@ -1882,7 +2029,7 @@ bool ResolveResolvedCavernGraphV1(ResolvedLayoutSnapshotV3& layout,std::int32_t
         if(!candidate.finalize(error))return false;
         layout=std::move(candidate);return true;
     }
-    candidate.cavernGraphVocabularyVersion=ResolvedLayoutCavernGraphVocabularyVersionV1;
+    candidate.cavernGraphVocabularyVersion=ResolvedLayoutCavernGraphVocabularyVersionV2;
     candidate.cavernRooms.assign(rooms.begin(),rooms.end());
     candidate.cavernEdges.assign(edges.begin(),edges.end());
     const auto isNativeCavesStrate=[&candidate](std::int32_t strateIndex)
@@ -2034,6 +2181,537 @@ bool ResolveResolvedCavernGraphV1(ResolvedLayoutSnapshotV3& layout,std::int32_t
     layout=std::move(candidate);return true;
 }
 
+bool PlanResolvedCavernNetworkV1(const ResolvedLayoutSnapshotV3& layout,
+    std::int32_t seed,std::int32_t strateIndex,
+    const ResolvedCavernPlannerSettingsV1& settings,
+    std::vector<ResolvedCavernRoomV1>& rooms,
+    std::vector<ResolvedCavernEdgeV1>& edges,std::string& error)
+{
+    rooms.clear();edges.clear();error.clear();
+    const auto stratum=std::find_if(layout.strates.begin(),layout.strates.end(),
+        [strateIndex](const auto& value){return value.strateIndex==strateIndex;});
+    if(stratum==layout.strates.end()||!IsNativeCaveFieldStratum(*stratum))
+    {error="cavern planner target is not a resolved native cave stratum";return false;}
+    if(settings.roomSpacingMicrometres<=0
+        ||settings.minimumRoomRadiusMicrometres<500000
+        ||settings.maximumRoomRadiusMicrometres<settings.minimumRoomRadiusMicrometres
+        ||settings.maximumRoomRadiusMicrometres>100000000
+        ||settings.destinationRadiusMicrometres<settings.maximumRoomRadiusMicrometres
+        ||settings.destinationRadiusMicrometres>100000000
+        ||settings.floorSupportDepthMicrometres<250000
+        ||settings.floorSupportDepthMicrometres>2000000
+        ||settings.maximumTunnelLengthMicrometres<=0
+        ||settings.minimumTunnelRadiusMicrometres<0
+        ||settings.maximumTunnelRadiusMicrometres<settings.minimumTunnelRadiusMicrometres
+        ||settings.smoothWidthMicrometres<0
+        ||settings.smoothWidthMicrometres>10000000
+        ||settings.irregularityAmplitudeMicrometres<0
+        ||settings.irregularityAmplitudeMicrometres>10000000
+        ||(settings.irregularityAmplitudeMicrometres>0
+            &&settings.irregularityWavelengthMicrometres<=0))
+    {error="cavern planner settings are outside the bounded room, tunnel or relief ranges";return false;}
+
+    const ResolvedLayoutPassageV1* arrivalPassage=nullptr;
+    const ResolvedLayoutPassageV1* departurePassage=nullptr;
+    for(const auto& passage:layout.passages)
+    {
+        if(passage.lowerStrateIndex==strateIndex&&passage.upperStrateIndex<strateIndex
+            &&(!arrivalPassage||passage.stableId<arrivalPassage->stableId))
+            arrivalPassage=&passage;
+        if(passage.upperStrateIndex==strateIndex&&passage.lowerStrateIndex>strateIndex
+            &&(!departurePassage||passage.stableId<departurePassage->stableId))
+            departurePassage=&passage;
+    }
+    if(!arrivalPassage||!departurePassage)
+    {error="cavern planner needs an incoming lower mouth and an outgoing upper mouth in the target strate";return false;}
+    if(layout.coreVoxelSizeMicrometres==0)
+    {error="cavern planner requires a nonzero Core voxel size";return false;}
+
+    using FPoint=std::array<std::int64_t,3>;
+    constexpr std::int64_t MicrometresPerMetre=1000000;
+    constexpr std::int64_t MinimumBoreRadius=1750000;
+    constexpr std::int64_t MinimumBoreHeight=2200000;
+    constexpr std::int64_t PlannedBoreHeight=4500000;
+    constexpr std::int64_t MouthOverlap=3000000;
+    constexpr std::int64_t MaximumBoreRadius=10000000;
+    constexpr std::int64_t MaximumWalkGradeDenominator=5; // 1:5 = 11.31 degrees.
+    const auto integerSqrt=[](std::uint64_t value)
+    {
+        std::uint64_t result=0;
+        std::uint64_t bit=std::uint64_t{1}<<62;
+        while(bit>value)bit>>=2;
+        while(bit!=0)
+        {
+            if(value>=result+bit)
+            {value-=result+bit;result=(result>>1)+bit;}
+            else result>>=1;
+            bit>>=2;
+        }
+        return result;
+    };
+    const auto distance2d=[&](const FPoint& a,const FPoint& b)
+    {
+        const std::int64_t dx=b[0]-a[0],dy=b[1]-a[1];
+        const auto squared=static_cast<std::uint64_t>(dx*dx+dy*dy);
+        return static_cast<std::int64_t>(integerSqrt(squared));
+    };
+    const auto qToMicrometres=[&](std::int64_t q)
+    {return q*static_cast<std::int64_t>(layout.coreVoxelSizeMicrometres)
+        /WorldAssemblySubvoxelsPerVoxel;};
+    const auto mouthInnerPoint=[&](const ResolvedLayoutPassageV1& passage,
+        bool lowerSide,FPoint& point)
+    {
+        for(std::size_t axis=0;axis<3;++axis)
+            point[axis]=qToMicrometres(lowerSide?passage.lowerMouthSubvoxels[axis]
+                :passage.upperMouthSubvoxels[axis]);
+        if(passage.controlPoints.size()<2)return false;
+        const std::size_t mouthIndex=lowerSide
+            ?std::min(passage.lowerMouthControlIndex,passage.controlPoints.size()-1):0;
+        const std::size_t tangentIndex=lowerSide
+            ?(mouthIndex>0?mouthIndex-1:1):1;
+        const std::int64_t tx=passage.controlPoints[mouthIndex].positionSubvoxels[0]
+            -passage.controlPoints[tangentIndex].positionSubvoxels[0];
+        const std::int64_t ty=passage.controlPoints[mouthIndex].positionSubvoxels[1]
+            -passage.controlPoints[tangentIndex].positionSubvoxels[1];
+        const auto tangentLength=static_cast<std::int64_t>(integerSqrt(
+            static_cast<std::uint64_t>(tx*tx+ty*ty)));
+        if(tangentLength<=0)return false;
+        point[0]+=tx*MouthOverlap/tangentLength;
+        point[1]+=ty*MouthOverlap/tangentLength;
+        return true;
+    };
+    FPoint routeStart{},routeEnd{};
+    if(!mouthInnerPoint(*arrivalPassage,true,routeStart)
+        ||!mouthInnerPoint(*departurePassage,false,routeEnd))
+    {error="cavern passage mouth is missing a usable horizontal route tangent";return false;}
+
+    const std::int64_t dx=routeEnd[0]-routeStart[0];
+    const std::int64_t dy=routeEnd[1]-routeStart[1];
+    const std::int64_t dz=routeEnd[2]-routeStart[2];
+    const std::int64_t straightLength=distance2d(routeStart,routeEnd);
+    const std::int64_t horizontalBias=std::clamp<std::int64_t>(
+        settings.tunnelHorizontalBiasQ16,0,65536);
+    const std::int64_t planGradeDenominator=MaximumWalkGradeDenominator
+        +horizontalBias/32768;
+    const std::int64_t requiredLength=std::abs(dz)*planGradeDenominator+1000;
+    std::int64_t lateral=std::max<std::int64_t>(1000000,
+        std::max(settings.tunnelWarpMicrometres,settings.caveWarpMicrometres)/2);
+    const std::uint64_t routeHash=LayoutStampSeed(seed,strateIndex,
+        "tunnel-network/main-route");
+    if((routeHash&1u)!=0)lateral=-lateral;
+    std::int64_t pathAxisX=dx,pathAxisY=dy;
+    const auto makeRoute=[&](std::int64_t lateralOffset)
+    {
+        std::array<FPoint,4> points{};
+        points[0]=routeStart;points[3]=routeEnd;
+        std::int64_t axisX=dx,axisY=dy;
+        std::int64_t axisLength=straightLength;
+        if(axisLength<=0)
+        {
+            axisX=1;axisY=0;axisLength=1;
+        }
+        pathAxisX=axisX;pathAxisY=axisY;
+        const std::int64_t perpX=-axisY*lateralOffset/axisLength;
+        const std::int64_t perpY= axisX*lateralOffset/axisLength;
+        for(std::size_t axis=0;axis<3;++axis)
+        {
+            const std::int64_t delta=routeEnd[axis]-routeStart[axis];
+            points[1][axis]=routeStart[axis]+delta/3;
+            points[2][axis]=routeStart[axis]+(delta*2)/3;
+        }
+        points[1][0]+=perpX;points[1][1]+=perpY;
+        points[2][0]-=perpX;points[2][1]-=perpY;
+        return points;
+    };
+    const auto routeLength=[&](const std::array<FPoint,4>& points)
+    {
+        std::int64_t length=0;
+        for(std::size_t i=1;i<points.size();++i)length+=distance2d(points[i-1],points[i]);
+        return length;
+    };
+    auto routePoints=makeRoute(lateral);
+    for(int attempt=0;routeLength(routePoints)<requiredLength&&attempt<32;++attempt)
+    {
+        if(lateral>0)lateral=std::min<std::int64_t>(lateral*2,1000000000ll);
+        else lateral=std::max<std::int64_t>(lateral*2,-1000000000ll);
+        routePoints=makeRoute(lateral);
+    }
+    const std::int64_t totalRouteLength=routeLength(routePoints);
+    if(totalRouteLength<=0||totalRouteLength<requiredLength)
+    {error="cavern planner cannot fit the fixed mouth floor difference under the 1:5 grade budget";return false;}
+
+    std::array<std::int64_t,4> cumulative{};
+    for(std::size_t i=1;i<routePoints.size();++i)
+        cumulative[i]=cumulative[i-1]+distance2d(routePoints[i-1],routePoints[i]);
+    if(cumulative.back()!=totalRouteLength)
+    {error="cavern planner route length accounting failed";return false;}
+    const std::string backboneFloorProfileId="tunnel-network/strate-"
+        +std::to_string(strateIndex)+"/backbone-floor";
+    const auto routeFloorAt=[&](std::int64_t distance)
+    {
+        distance=std::clamp<std::int64_t>(distance,0,totalRouteLength);
+        return routeStart[2]+dz*distance/totalRouteLength;
+    };
+    const auto routeFloorGradientAt=[&](std::int64_t distance)
+    {
+        distance=std::clamp<std::int64_t>(distance,0,totalRouteLength);
+        std::size_t segment=1;
+        while(segment<cumulative.size()-1&&distance>cumulative[segment])++segment;
+        const auto unitTangent=[&](std::size_t index)
+        {
+            const double tx=static_cast<double>(routePoints[index][0]
+                -routePoints[index-1][0]);
+            const double ty=static_cast<double>(routePoints[index][1]
+                -routePoints[index-1][1]);
+            const double length=std::hypot(tx,ty);
+            return length>0.0?std::array<double,2>{tx/length,ty/length}
+                :std::array<double,2>{0.0,0.0};
+        };
+        auto tangent=unitTangent(segment);
+        if(segment+1<cumulative.size()&&distance==cumulative[segment])
+        {
+            const auto next=unitTangent(segment+1);
+            tangent[0]+=next[0];tangent[1]+=next[1];
+            const double length=std::hypot(tangent[0],tangent[1]);
+            if(length>1.0e-12){tangent[0]/=length;tangent[1]/=length;}
+        }
+        constexpr double Q16=65536.0;
+        const double profileSlope=static_cast<double>(dz)/totalRouteLength;
+        return std::array<std::int32_t,2>{
+            static_cast<std::int32_t>(std::llround(tangent[0]*profileSlope*Q16)),
+            static_cast<std::int32_t>(std::llround(tangent[1]*profileSlope*Q16))};
+    };
+    const auto pointAt=[&](std::int64_t distance)
+    {
+        distance=std::clamp<std::int64_t>(distance,0,totalRouteLength);
+        std::size_t segment=1;
+        while(segment<cumulative.size()-1&&distance>cumulative[segment])++segment;
+        const std::int64_t startDistance=cumulative[segment-1];
+        const std::int64_t segmentLength=std::max<std::int64_t>(1,
+            cumulative[segment]-startDistance);
+        const std::int64_t local=distance-startDistance;
+        FPoint point{};
+        for(std::size_t axis=0;axis<2;++axis)
+        {
+            const std::int64_t delta=routePoints[segment][axis]
+                -routePoints[segment-1][axis];
+            point[axis]=routePoints[segment-1][axis]+delta*local/segmentLength;
+        }
+        point[2]=routeFloorAt(distance);
+        return point;
+    };
+    std::int64_t spacing=std::min(std::max<std::int64_t>(settings.roomSpacingMicrometres,
+        4000000),settings.maximumTunnelLengthMicrometres);
+    spacing=std::max<std::int64_t>(4000000,spacing);
+    if(spacing>settings.maximumTunnelLengthMicrometres)
+        spacing=settings.maximumTunnelLengthMicrometres;
+    if(spacing<=0)
+    {error="cavern planner has no positive route spacing after applying tunnel length limits";return false;}
+    std::vector<std::int64_t> nodeDistances;
+    for(std::size_t leg=1;leg<routePoints.size();++leg)
+    {
+        const std::int64_t legStart=cumulative[leg-1];
+        const std::int64_t legLength=cumulative[leg]-legStart;
+        const std::int64_t segments=std::max<std::int64_t>(1,
+            (legLength+spacing-1)/spacing);
+        if(segments>256)
+        {error="cavern planner would exceed the 256-node bounded backbone";return false;}
+        for(std::int64_t part=1;part<=segments;++part)
+        {
+            const std::int64_t distance=legStart+legLength*part/segments;
+            if(distance>0&&distance<totalRouteLength)nodeDistances.push_back(distance);
+        }
+    }
+    nodeDistances.push_back(totalRouteLength/2);
+    std::sort(nodeDistances.begin(),nodeDistances.end());
+    nodeDistances.erase(std::unique(nodeDistances.begin(),nodeDistances.end()),nodeDistances.end());
+    if(nodeDistances.empty())nodeDistances.push_back(totalRouteLength/2);
+    if(nodeDistances.size()>256)
+    {error="cavern planner would exceed the 256-node bounded backbone";return false;}
+
+    const auto clampQ16=[](std::int32_t value)
+    {return std::clamp<std::int64_t>(value,0,65536);};
+    const std::int64_t smooth=std::clamp<std::int64_t>(
+        std::max<std::int64_t>(1500000,settings.smoothWidthMicrometres),
+        50000,10000000);
+    const std::int64_t irregularity=std::clamp<std::int64_t>(
+        settings.irregularityAmplitudeMicrometres,0,500000);
+    const std::int64_t routeBoreRadius=std::clamp<std::int64_t>(
+        std::max(MinimumBoreRadius,settings.minimumTunnelRadiusMicrometres),
+        MinimumBoreRadius,MaximumBoreRadius);
+    const std::int64_t maximumRouteBoreRadius=std::clamp<std::int64_t>(
+        std::max(routeBoreRadius,settings.maximumTunnelRadiusMicrometres),
+        routeBoreRadius,MaximumBoreRadius);
+    const auto selectedBoreRadius=[&](const std::string& stableId)
+    {
+        const std::uint64_t span=static_cast<std::uint64_t>(
+            maximumRouteBoreRadius-routeBoreRadius);
+        return routeBoreRadius+(span==0?0:static_cast<std::int64_t>(
+            LayoutStampSeed(seed,strateIndex,stableId)%(span+1)));
+    };
+    const auto passageBoreRadius=[&](const ResolvedLayoutPassageV1& passage,bool lowerSide)
+    {
+        std::int64_t radius=qToMicrometres(passage.mouthRadiusSubvoxels);
+        if(!passage.controlPoints.empty())
+        {
+            const auto control=lowerSide?std::min(passage.lowerMouthControlIndex,
+                passage.controlPoints.size()-1):std::size_t{0};
+            radius=std::max(radius,qToMicrometres(passage.controlPoints[control].radiusSubvoxels));
+        }
+        return std::clamp<std::int64_t>(std::max(routeBoreRadius,radius),
+            routeBoreRadius,MaximumBoreRadius);
+    };
+    const auto fractional=[](std::int64_t value,std::int64_t q16)
+    {return value*std::clamp<std::int64_t>(q16,0,65536)/65536;};
+    struct FRoomNode
+    {
+        std::string id;
+        FPoint point{};
+        std::string floorProfileStableId;
+        std::int64_t floorProfileDistanceMicrometres=-1;
+        std::array<std::int32_t,2> floorGradientQ16{};
+        std::int64_t radiusMicrometres=0;
+        bool destination=false;
+    };
+    std::vector<FRoomNode> nodes;nodes.reserve(nodeDistances.size());
+    const std::int64_t floorCutMinimum=clampQ16(settings.roomFloorCutMinimumQ16);
+    const std::int64_t floorCutMaximum=std::max(floorCutMinimum,
+        clampQ16(settings.roomFloorCutMaximumQ16));
+    const std::int64_t floorRange=floorCutMaximum-floorCutMinimum;
+    const std::int64_t heightRatio=std::clamp<std::int64_t>(
+        settings.roomHeightRatioQ16,6554,65536);
+    const std::int64_t baseRoomSpan=settings.maximumRoomRadiusMicrometres
+        -settings.minimumRoomRadiusMicrometres;
+    const std::int64_t mainRouteMidpoint=totalRouteLength/2;
+    for(std::size_t nodeIndex=0;nodeIndex<nodeDistances.size();++nodeIndex)
+    {
+        const std::int64_t distance=nodeDistances[nodeIndex];
+        FRoomNode node;node.point=pointAt(distance);
+        node.floorProfileStableId=backboneFloorProfileId;
+        node.floorProfileDistanceMicrometres=distance;
+        node.floorGradientQ16=routeFloorGradientAt(distance);
+        node.destination=distance==mainRouteMidpoint;
+        node.id="tunnel-network/strate-"+std::to_string(strateIndex)+"/room-"
+            +std::to_string(nodeIndex);
+        const std::uint64_t hash=LayoutStampSeed(seed,strateIndex,node.id);
+        node.radiusMicrometres=node.destination?settings.destinationRadiusMicrometres
+            :settings.minimumRoomRadiusMicrometres
+                +static_cast<std::int64_t>(hash%static_cast<std::uint64_t>(baseRoomSpan+1));
+        nodes.push_back(std::move(node));
+    }
+    const auto makeCompound=[&](const FRoomNode& node,bool destination,
+        std::vector<ResolvedCavernRoomV1>& outRooms)
+    {
+        const std::uint64_t hash=LayoutStampSeed(seed,strateIndex,node.id+"/compound");
+        const std::int64_t variety=clampQ16(settings.roomShapeVarietyQ16);
+        const std::size_t additionalPrimitiveCount=static_cast<std::size_t>(
+            variety*2/65536);
+        const std::size_t primitiveCount=2+static_cast<std::size_t>(
+            hash%(additionalPrimitiveCount+1));
+        std::uint64_t state=hash;
+        const std::int64_t radius=node.radiusMicrometres;
+        const std::int64_t floorCutBase=clampQ16(settings.roomFloorCutMinimumQ16);
+        const std::int64_t componentCount=static_cast<std::int64_t>(primitiveCount);
+        for(std::int64_t component=0;component<componentCount;++component)
+        {
+            const std::string stableId=node.id+"/primitive-"+std::to_string(component);
+            const std::uint64_t componentHash=LayoutStampNext(state);
+            const std::int64_t componentScaleQ16=component==0?65536
+                :42000+static_cast<std::int64_t>(componentHash%22001);
+            const std::int64_t componentRadius=std::max<std::int64_t>(500000,
+                fractional(radius,componentScaleQ16));
+            const std::int64_t axisScaleQ16=component==0?65536
+                :47000+static_cast<std::int64_t>((componentHash>>8)%19001);
+            const bool swapAxes=((componentHash>>24)&1u)!=0;
+            const std::int64_t halfX=swapAxes?fractional(componentRadius,axisScaleQ16)
+                :componentRadius;
+            const std::int64_t halfY=swapAxes?componentRadius
+                :fractional(componentRadius,axisScaleQ16);
+            const std::int64_t halfZ=std::max<std::int64_t>(500000,
+                fractional(componentRadius,heightRatio));
+            const std::int64_t floorCut=floorCutBase
+                +(floorRange==0?0:static_cast<std::int64_t>(componentHash%(
+                    static_cast<std::uint64_t>(floorRange+1))));
+            const std::int64_t offsetMagnitude=component==0?0
+                :fractional(radius,12000+static_cast<std::int64_t>(componentHash%11001));
+            const std::int64_t routeLengthXY=std::max<std::int64_t>(1,
+                static_cast<std::int64_t>(integerSqrt(static_cast<std::uint64_t>(
+                    pathAxisX*pathAxisX+pathAxisY*pathAxisY))));
+            const std::int64_t centerOffset=(component%2==0?1:-1)*offsetMagnitude;
+            ResolvedCavernRoomV1 room;room.stableId=stableId;
+            room.strateIndex=strateIndex;
+            room.floorProfileStableId=node.floorProfileStableId;
+            room.floorProfileDistanceMicrometres=node.floorProfileDistanceMicrometres;
+            room.floorGradientQ16=node.floorGradientQ16;
+            room.halfExtentsMicrometres={halfX,halfY,halfZ};
+            room.centerMicrometres={node.point[0]+(-pathAxisY)*centerOffset/routeLengthXY,
+                node.point[1]+pathAxisX*centerOffset/routeLengthXY,
+                node.point[2]+fractional(halfZ,floorCut)};
+            if(component>0)
+            {
+                const std::int64_t headroom=std::max<std::int64_t>(0,
+                    halfZ-fractional(halfZ,floorCut));
+                const std::int64_t jitterRange=std::min(headroom,halfZ/5);
+                const std::int64_t ceilingJitter=jitterRange>0
+                    ?static_cast<std::int64_t>((componentHash>>32)
+                        %static_cast<std::uint64_t>(jitterRange+1)):0;
+                room.centerMicrometres[2]+=ceilingJitter;
+            }
+            room.floorMicrometres=node.point[2]
+                -(static_cast<std::int64_t>(room.floorGradientQ16[0])
+                    *(room.centerMicrometres[0]-node.point[0])
+                    +static_cast<std::int64_t>(room.floorGradientQ16[1])
+                    *(room.centerMicrometres[1]-node.point[1]))/65536;
+            room.smoothWidthMicrometres=smooth;
+            room.irregularityAmplitudeMicrometres=irregularity;
+            room.irregularityWavelengthMicrometres=settings.irregularityWavelengthMicrometres;
+            outRooms.push_back(std::move(room));
+        }
+        (void)destination;
+    };
+    for(const auto& node:nodes)makeCompound(node,node.destination,rooms);
+
+    const auto componentRoomId=[](const FRoomNode& node)
+    {return node.id+"/primitive-0";};
+    const auto makeEdge=[&](std::string id,std::string startAnchor,
+        const FPoint& start,const FRoomNode& endNode,std::int64_t boreRadius,
+        std::int64_t profileStartDistance,std::int64_t profileEndDistance)
+    {
+        ResolvedCavernEdgeV1 edge;edge.stableId=std::move(id);
+        edge.startAnchorStableId=std::move(startAnchor);
+        edge.endRoomStableId=componentRoomId(endNode);edge.strateIndex=strateIndex;
+        edge.boreRadiusMicrometres=boreRadius;
+        edge.boreHeightMicrometres=std::max(PlannedBoreHeight,MinimumBoreHeight);
+        edge.floorSupportDepthMicrometres=settings.floorSupportDepthMicrometres;
+        if(profileStartDistance>=0&&profileEndDistance>=0)
+        {
+            edge.floorProfileStableId=backboneFloorProfileId;
+            edge.floorProfileStartDistanceMicrometres=profileStartDistance;
+            edge.floorProfileEndDistanceMicrometres=profileEndDistance;
+        }
+        edge.smoothWidthMicrometres=smooth;
+        edge.irregularityAmplitudeMicrometres=irregularity;
+        edge.irregularityWavelengthMicrometres=settings.irregularityWavelengthMicrometres;
+        const FPoint end={endNode.point[0],endNode.point[1],endNode.point[2]};
+        for(std::size_t axis=0;axis<3;++axis)
+        {
+            const std::int64_t delta=end[axis]-start[axis];
+            edge.controlPointsMicrometres[0][axis]=start[axis];
+            edge.controlPointsMicrometres[1][axis]=start[axis]+delta/3;
+            edge.controlPointsMicrometres[2][axis]=start[axis]+(delta*2)/3;
+            edge.controlPointsMicrometres[3][axis]=end[axis];
+        }
+        if(profileStartDistance>=0&&profileEndDistance>=0)
+        {
+            const std::int64_t floorStart=routeFloorAt(profileStartDistance);
+            const std::int64_t floorEnd=routeFloorAt(profileEndDistance);
+            const std::int64_t floorDelta=floorEnd-floorStart;
+            edge.controlPointsMicrometres[0][2]=floorStart;
+            edge.controlPointsMicrometres[1][2]=floorStart+floorDelta/3;
+            edge.controlPointsMicrometres[2][2]=floorStart+(floorDelta*2)/3;
+            edge.controlPointsMicrometres[3][2]=floorEnd;
+        }
+        return edge;
+    };
+    const auto arrivalPoint=routeStart;
+    FPoint previousPoint=arrivalPoint;
+    std::int64_t previousProfileDistance=0;
+    std::string previousAnchor=arrivalPassage->stableId+"/lower-mouth";
+    for(std::size_t nodeIndex=0;nodeIndex<nodes.size();++nodeIndex)
+    {
+        const auto& node=nodes[nodeIndex];
+        const bool portal=nodeIndex==0;
+        const std::string edgeId="tunnel-network/strate-"+std::to_string(strateIndex)
+            +"/route-edge-"+std::to_string(nodeIndex);
+        edges.push_back(makeEdge(edgeId,previousAnchor,previousPoint,node,
+            portal?passageBoreRadius(*arrivalPassage,true):selectedBoreRadius(edgeId),
+            previousProfileDistance,node.floorProfileDistanceMicrometres));
+        previousPoint={node.point[0],node.point[1],node.point[2]};
+        previousProfileDistance=node.floorProfileDistanceMicrometres;
+        previousAnchor=componentRoomId(node);
+    }
+    if(!nodes.empty())
+    {
+        const std::string departureEdgeId="tunnel-network/strate-"
+            +std::to_string(strateIndex)+"/departure-edge";
+        edges.push_back(makeEdge(departureEdgeId,departurePassage->stableId+"/upper-mouth",
+            routeEnd,nodes.back(),passageBoreRadius(*departurePassage,false),
+            totalRouteLength,nodes.back().floorProfileDistanceMicrometres));
+    }
+
+    const std::int64_t roomDensity=clampQ16(settings.roomDensityQ16);
+    const std::int64_t tunnelDensity=clampQ16(settings.tunnelDensityQ16);
+    const std::int64_t branchThreshold=roomDensity*tunnelDensity/65536;
+    const std::size_t branchLimit=settings.originRoomMaximumConnections<=0
+        ?32u:static_cast<std::size_t>(std::clamp(settings.originRoomMaximumConnections,0,32));
+    std::size_t branchCount=0;
+    for(std::size_t nodeIndex=1;nodeIndex+1<nodes.size()
+        &&branchCount<branchLimit&&rooms.size()<2048;++nodeIndex)
+    {
+        const auto& parent=nodes[nodeIndex];
+        const std::uint64_t branchHash=LayoutStampSeed(seed,strateIndex,
+            parent.id+"/side-branch");
+        if(static_cast<std::int64_t>(branchHash&0xffffu)>=branchThreshold)continue;
+        const std::int64_t branchLength=std::max<std::int64_t>(
+            routeBoreRadius*3,std::min(spacing/2,settings.maximumTunnelLengthMicrometres/2));
+        const std::int64_t direction=(branchHash&0x10000u)?1:-1;
+        FRoomNode branch;branch.id=parent.id+"/side-"+std::to_string(branchCount);
+        branch.radiusMicrometres=settings.minimumRoomRadiusMicrometres
+            +static_cast<std::int64_t>((branchHash>>17)
+                %static_cast<std::uint64_t>(baseRoomSpan+1));
+        const std::int64_t pathAxisLength=std::max<std::int64_t>(1,
+            static_cast<std::int64_t>(integerSqrt(static_cast<std::uint64_t>(
+                pathAxisX*pathAxisX+pathAxisY*pathAxisY))));
+        branch.point={parent.point[0]+(-pathAxisY)*branchLength/pathAxisLength*direction,
+            parent.point[1]+pathAxisX*branchLength/pathAxisLength*direction,
+            parent.point[2]};
+        branch.floorProfileStableId=parent.floorProfileStableId;
+        branch.floorProfileDistanceMicrometres=parent.floorProfileDistanceMicrometres;
+        branch.floorGradientQ16=parent.floorGradientQ16;
+        makeCompound(branch,false,rooms);
+        const std::string branchEdgeId="tunnel-network/strate-"
+            +std::to_string(strateIndex)+"/side-edge-"+std::to_string(branchCount);
+        ResolvedCavernEdgeV1 branchEdge;
+        if(settings.tunnelsFlowTowardOrigin)
+            branchEdge=makeEdge(branchEdgeId,componentRoomId(branch),
+                branch.point,parent,selectedBoreRadius(branchEdgeId),-1,-1);
+        else
+            branchEdge=makeEdge(branchEdgeId,componentRoomId(parent),
+                parent.point,branch,selectedBoreRadius(branchEdgeId),-1,-1);
+        // Side branches retain their cave floor SDF, but they do not cap the
+        // descending backbone when their tubes overlap it near a junction.
+        branchEdge.floorSupportDepthMicrometres=0;
+        edges.push_back(std::move(branchEdge));
+        ++branchCount;
+    }
+    if(nodes.empty()||rooms.empty()||edges.empty())
+    {error="cavern planner emitted an empty arrival-to-departure graph";return false;}
+
+    double maximumGradeDegrees=0.0;
+    for(const auto& edge:edges)
+    {
+        const auto& a=edge.controlPointsMicrometres.front();
+        const auto& b=edge.controlPointsMicrometres.back();
+        const std::int64_t run=static_cast<std::int64_t>(integerSqrt(
+            static_cast<std::uint64_t>((b[0]-a[0])*(b[0]-a[0])
+                +(b[1]-a[1])*(b[1]-a[1]))));
+        const std::int64_t rise=std::abs(b[2]-a[2]);
+        if(run<=0&&rise>0)
+        {error="cavern planner produced a vertical route edge with no walking run";return false;}
+        if(run>0&&rise*MaximumWalkGradeDenominator>run)
+        {error="cavern planner emitted an edge beyond its 1:5 route grade budget";return false;}
+        if(run>0)maximumGradeDegrees=std::max(maximumGradeDegrees,
+            std::atan2(static_cast<double>(rise),static_cast<double>(run))
+                *180.0/3.14159265358979323846);
+    }
+    std::printf("[CavernPlannerV1] strate=%d seed=%d mouths=%s->%s route_nodes=%zu room_primitives=%zu edges=%zu branches=%zu route_horizontal_m=%.3f floor_delta_m=%.3f max_grade_deg=%.3f grade_limit_deg=11.310 clearance_radius_m=%.3f clearance_height_m=%.3f destination_radius_m=%.3f status=PASS\n",
+        strateIndex,seed,arrivalPassage->stableId.c_str(),departurePassage->stableId.c_str(),
+        nodes.size(),rooms.size(),edges.size(),branchCount,totalRouteLength/1.0e6,
+        std::abs(dz)/1.0e6,maximumGradeDegrees,routeBoreRadius/1.0e6,PlannedBoreHeight/1.0e6,
+        settings.destinationRadiusMicrometres/1.0e6);
+    return true;
+}
+
 bool ResolveResolvedLayoutOreVeinsV1(ResolvedLayoutSnapshotV3& layout,std::int32_t seed,
     std::span<const ResolvedLayoutOreVeinRequestV1> requests,std::string& error)
 {

Standalone/Core/src/VoxelForgeCore.cpp

@@ -1284,103 +1284,104 @@ namespace VoxelForgeCore
             constexpr double Q=static_cast<double>(WorldAssemblySubvoxelsPerVoxel);
             const double voxelsPerMetre=coreVoxelSizeMicrons>0
                 ?1000000.0/coreVoxelSizeMicrons:4.0;
-            const double bridgeHalfWidthVoxels=3.5*voxelsPerMetre;
-            const double bridgeFootDepthVoxels=(5.0-0.13)*voxelsPerMetre;
+            constexpr double WallReliefMetres=0.21;
+            constexpr double ShellThicknessMetres=0.75;
+            const double tubeExtentMetres=(WallReliefMetres+ShellThicknessMetres)
+                *voxelsPerMetre;
+            const double qPerMetre=Q*voxelsPerMetre;
             const double tileMinimum[3]={static_cast<double>(minimum.X),
                 static_cast<double>(minimum.Y),static_cast<double>(minimum.Z)};
             const double tileMaximum[3]={static_cast<double>(maximum.X),
                 static_cast<double>(maximum.Y),static_cast<double>(maximum.Z)};
             for(const auto& passage:layout.passages)
             {
+                if(passage.controlPoints.size()<2)continue;
+                const std::size_t routeEnd=std::clamp<std::size_t>(
+                    passage.lowerMouthControlIndex,1,passage.controlPoints.size()-1);
+                std::vector<double> routeDistanceMetres(routeEnd+1,0.0);
+                for(std::size_t pointIndex=1;pointIndex<=routeEnd;++pointIndex)
+                {
+                    const auto& a=passage.controlPoints[pointIndex-1].positionSubvoxels;
+                    const auto& b=passage.controlPoints[pointIndex].positionSubvoxels;
+                    const double dx=static_cast<double>(b[0]-a[0])/qPerMetre;
+                    const double dy=static_cast<double>(b[1]-a[1])/qPerMetre;
+                    const double dz=static_cast<double>(b[2]-a[2])/qPerMetre;
+                    routeDistanceMetres[pointIndex]=routeDistanceMetres[pointIndex-1]
+                        +std::sqrt(dx*dx+dy*dy+dz*dz);
+                }
+                const double totalRouteMetres=routeDistanceMetres[routeEnd];
+                double minimumRadiusMetres=std::numeric_limits<double>::infinity();
+                double minimumBodyRadiusMetres=minimumRadiusMetres;
+                for(std::size_t pointIndex=0;pointIndex<=routeEnd;++pointIndex)
+                {
+                    const double radius=static_cast<double>(
+                        passage.controlPoints[pointIndex].radiusSubvoxels)/qPerMetre;
+                    minimumRadiusMetres=std::min(minimumRadiusMetres,radius);
+                    if(routeDistanceMetres[pointIndex]>=10.0
+                        &&totalRouteMetres-routeDistanceMetres[pointIndex]>=10.0)
+                        minimumBodyRadiusMetres=std::min(minimumBodyRadiusMetres,radius);
+                }
+                if(!std::isfinite(minimumBodyRadiusMetres))
+                    minimumBodyRadiusMetres=minimumRadiusMetres;
+                const double bodyFloorDepthMetres=std::max(0.5,
+                    minimumBodyRadiusMetres-0.23);
+                const double upperMouthDepthMetres=std::max(0.0,
+                    static_cast<double>(passage.controlPoints.front().positionSubvoxels[2]
+                        -passage.upperMouthSubvoxels[2])/qPerMetre+0.90);
+                const double lowerMouthDepthMetres=std::max(0.0,
+                    static_cast<double>(passage.controlPoints[routeEnd].positionSubvoxels[2]
+                        -passage.lowerMouthSubvoxels[2])/qPerMetre+0.90);
+                const auto routeFloorDepth=[&](double alongMetres,double radiusMetres)
+                {
+                    const double localRadiusDepth=std::max(0.0,radiusMetres-0.23);
+                    const double mouthDistance=std::max(0.0,std::min(alongMetres,
+                        totalRouteMetres-alongMetres));
+                    const double linear=std::clamp(mouthDistance/5.0,0.0,1.0);
+                    const double ramp=linear*linear*(3.0-2.0*linear);
+                    const double target=alongMetres<=totalRouteMetres*0.5
+                        ?upperMouthDepthMetres:lowerMouthDepthMetres;
+                    return std::lerp(std::clamp(target,0.0,localRadiusDepth),
+                        localRadiusDepth,ramp);
+                };
+                std::vector<double> effectiveCenterZVoxels(
+                    passage.controlPoints.size(),0.0);
+                for(std::size_t pointIndex=0;pointIndex<passage.controlPoints.size();
+                    ++pointIndex)
+                {
+                    const std::size_t radiusIndex=std::min(pointIndex,routeEnd);
+                    const double radius=static_cast<double>(
+                        passage.controlPoints[radiusIndex].radiusSubvoxels)/qPerMetre;
+                    const double localTubeDepth=std::min(bodyFloorDepthMetres,
+                        std::max(0.0,radius-0.23));
+                    const double routeDepth=pointIndex<=routeEnd
+                        ?routeFloorDepth(routeDistanceMetres[pointIndex],radius)
+                        :std::max(0.0,radius-0.23);
+                    effectiveCenterZVoxels[pointIndex]=
+                        static_cast<double>(passage.controlPoints[pointIndex]
+                            .positionSubvoxels[2])/Q
+                        +(-routeDepth+localTubeDepth)*voxelsPerMetre;
+                }
                 for(std::size_t segment=1;segment<passage.controlPoints.size();++segment)
                 {
                     const auto& a=passage.controlPoints[segment-1];
                     const auto& b=passage.controlPoints[segment];
-                    const double radius=static_cast<double>(std::max(
-                        a.radiusSubvoxels,b.radiusSubvoxels))/Q;
-                    const double horizontalExtent=std::max(radius,bridgeHalfWidthVoxels);
-                    const double low[3]={
-                        std::min(a.positionSubvoxels[0],b.positionSubvoxels[0])/Q-horizontalExtent,
-                        std::min(a.positionSubvoxels[1],b.positionSubvoxels[1])/Q-horizontalExtent,
-                        std::min(a.positionSubvoxels[2],b.positionSubvoxels[2])/Q-radius
-                            -bridgeFootDepthVoxels};
-                    const double high[3]={
-                        std::max(a.positionSubvoxels[0],b.positionSubvoxels[0])/Q+horizontalExtent,
-                        std::max(a.positionSubvoxels[1],b.positionSubvoxels[1])/Q+horizontalExtent,
-                        std::max(a.positionSubvoxels[2],b.positionSubvoxels[2])/Q+radius};
+                    const double extent=std::max(static_cast<double>(a.radiusSubvoxels),
+                        static_cast<double>(b.radiusSubvoxels))/Q+tubeExtentMetres;
+                    double low[3]{},high[3]{};
+                    for(int axis=0;axis<3;++axis)
+                    {
+                        const double aPosition=axis==2?effectiveCenterZVoxels[segment-1]
+                            :static_cast<double>(a.positionSubvoxels[axis])/Q;
+                        const double bPosition=axis==2?effectiveCenterZVoxels[segment]
+                            :static_cast<double>(b.positionSubvoxels[axis])/Q;
+                        low[axis]=std::min(aPosition,bPosition)-extent;
+                        high[axis]=std::max(aPosition,bPosition)+extent;
+                    }
                     bool overlap=true;
                     for(int axis=0;axis<3;++axis)
                         overlap&=high[axis]>=tileMinimum[axis]&&low[axis]<=tileMaximum[axis];
                     if(overlap)return true;
                 }
-
-            // L27 can add a long support column around an unsupported passage span. Its
-            // radius can exceed the route and bridge bounds above by the column wall and
-            // flare allowance. The precompile shape-free proof runs before that crossing
-            // plan exists, so conservatively bound every possible column from the resolved
-            // route itself. The center is either the passage anchor or an average of route
-            // samples; in the latter case it lies inside the route's XY bounds.
-            if(passage.controlPoints.size()>=2)
-            {
-                double routeMinimum[3]={std::numeric_limits<double>::infinity(),
-                    std::numeric_limits<double>::infinity(),
-                    std::numeric_limits<double>::infinity()};
-                double routeMaximum[3]={-std::numeric_limits<double>::infinity(),
-                    -std::numeric_limits<double>::infinity(),
-                    -std::numeric_limits<double>::infinity()};
-                double maximumBoreRadius=0.0;
-                for(const auto& point:passage.controlPoints)
-                {
-                    for(int axis=0;axis<3;++axis)
-                    {
-                        const double value=static_cast<double>(point.positionSubvoxels[axis])/Q;
-                        routeMinimum[axis]=std::min(routeMinimum[axis],value);
-                        routeMaximum[axis]=std::max(routeMaximum[axis],value);
-                    }
-                    maximumBoreRadius=std::max(maximumBoreRadius,
-                        static_cast<double>(point.radiusSubvoxels)/Q);
-                }
-
-                double centerMinimumX=routeMinimum[0],centerMaximumX=routeMaximum[0];
-                double centerMinimumY=routeMinimum[1],centerMaximumY=routeMaximum[1];
-                double maximumRouteOffset=std::hypot(
-                    routeMaximum[0]-routeMinimum[0],routeMaximum[1]-routeMinimum[1]);
-                const auto upper=std::find_if(layout.strates.begin(),layout.strates.end(),
-                    [&passage](const auto& stratum)
-                    {return stratum.strateIndex==passage.upperStrateIndex;});
-                if(upper!=layout.strates.end()
-                    &&upper->passageCapabilities.usePassageAnchorForColumns)
-                {
-                    const auto anchor=std::find_if(upper->anchors.begin(),upper->anchors.end(),
-                        [](const auto& value)
-                        {return value.kind=="passage_anchor"&&value.values.size()>=2;});
-                    if(anchor!=upper->anchors.end())
-                    {
-                        const double centerX=static_cast<double>(anchor->values[0])/Q;
-                        const double centerY=static_cast<double>(anchor->values[1])/Q;
-                        centerMinimumX=centerMaximumX=centerX;
-                        centerMinimumY=centerMaximumY=centerY;
-                        maximumRouteOffset=0.0;
-                        for(const auto& point:passage.controlPoints)
-                            maximumRouteOffset=std::max(maximumRouteOffset,std::hypot(
-                                static_cast<double>(point.positionSubvoxels[0])/Q-centerX,
-                                static_cast<double>(point.positionSubvoxels[1])/Q-centerY));
-                    }
-                }
-                // The generated maximum column radius is the route offset, bore radius,
-                // and 13.25 m of wall/irregularity/flare allowance.
-                const double columnRadius=maximumRouteOffset+maximumBoreRadius
-                    +13.25*voxelsPerMetre;
-                const double columnMinimum[3]={centerMinimumX-columnRadius,
-                    centerMinimumY-columnRadius,routeMinimum[2]-6.0*voxelsPerMetre};
-                const double columnMaximum[3]={centerMaximumX+columnRadius,
-                    centerMaximumY+columnRadius,routeMaximum[2]+5.0*voxelsPerMetre};
-                bool overlap=true;
-                for(int axis=0;axis<3;++axis)
-                    overlap=overlap&&columnMaximum[axis]>=tileMinimum[axis]
-                        &&columnMinimum[axis]<=tileMaximum[axis];
-                if(overlap)return true;
-            }
             }
             return false;
         }
@@ -1866,44 +1867,6 @@ namespace VoxelForgeCore
                 static_cast<double>(weighted)/65535.0*sampleVolume});
         return result;
     }
-    std::vector<PassageVoidCrossingAuditV1> World::passageVoidCrossings() const
-    {
-        std::vector<PassageVoidCrossingAuditV1> result;
-        if(!impl_||!impl_->shapeReady.load(std::memory_order_acquire)||!impl_->shape)return result;
-        const auto& crossings=impl_->shape->GetPassageVoidCrossings();
-        result.reserve(crossings.size());
-        for(const auto& crossing:crossings)
-        {
-            PassageVoidCrossingAuditV1 value;
-            value.passageIndex=crossing.passageIndex;value.firstSegment=crossing.firstSegment;
-            value.lastSegment=crossing.lastSegment;value.unsupportedSamples=crossing.unsupportedSamples;
-            value.unsupportedLengthMetres=crossing.unsupportedLengthMetres;
-            value.startPointSubvoxels=crossing.startPoint;value.endPointSubvoxels=crossing.endPoint;
-            value.needsColumn=crossing.needsColumn;
-            value.connectedAtStart=crossing.connectedAtStart;
-            value.connectedAtEnd=crossing.connectedAtEnd;
-            value.columnAxisSubvoxels={crossing.centerXSubvoxels,crossing.centerYSubvoxels};
-            value.columnTopZSubvoxels=crossing.topZSubvoxels;
-            value.columnBottomZSubvoxels=crossing.bottomZSubvoxels;
-            value.columnMinimumRadiusSubvoxels=crossing.minimumRadiusSubvoxels;
-            value.columnNominalRadiusSubvoxels=crossing.nominalRadiusSubvoxels;
-            value.columnMaximumRadiusSubvoxels=crossing.maximumRadiusSubvoxels;
-            value.maximumRouteOffsetMetres=crossing.maximumRouteOffsetMetres;
-            value.maximumBoreRadiusMetres=crossing.maximumBoreRadiusMetres;
-            result.push_back(value);
-        }
-        return result;
-    }
-    bool World::isPassageColumnSample(std::size_t passageIndex,std::size_t segmentIndex,
-        Vec3d position) const
-    {return impl_&&impl_->shapeReady.load(std::memory_order_acquire)&&impl_->shape
-        &&impl_->shape->IsResolvedPassageColumnSample(passageIndex,segmentIndex,
-            {position.x,position.y,position.z});}
-    bool World::isPassageShortSpanSample(std::size_t passageIndex,std::size_t segmentIndex,
-        Vec3d position) const
-    {return impl_&&impl_->shapeReady.load(std::memory_order_acquire)&&impl_->shape
-        &&impl_->shape->IsResolvedPassageShortSpanSample(passageIndex,segmentIndex,
-            {position.x,position.y,position.z});}
     std::string World::densityDispatchPathAt(Vec3d position) const
     { return impl_&&impl_->shapeReady.load(std::memory_order_acquire)&&impl_->shape
         ?impl_->shape->DensityDispatchPathAt(static_cast<float>(position.x),static_cast<float>(position.y),static_cast<float>(position.z))
@@ -2672,30 +2635,6 @@ namespace VoxelForgeCore
                         RelevantMinimum=std::min(RelevantMinimum,FeatureMinimum);
                         RelevantMaximum=std::max(RelevantMaximum,FeatureMaximum);
                     }
-                    // A resolved passage void column can add solid support above or
-                    // below the native heightfield's local Z range. Its finite radial
-                    // and vertical bounds must participate before the mesh band is clipped.
-                    for(const auto& Crossing:tileShape.GetPassageVoidCrossings())
-                    {
-                        if(!Crossing.needsColumn||Crossing.maximumRadiusSubvoxels<=0
-                            ||Crossing.topZSubvoxels<=Crossing.bottomZSubvoxels)continue;
-                        const double Radius=static_cast<double>(Crossing.maximumRadiusSubvoxels)
-                            /static_cast<double>(SubvoxelScale);
-                        const double CenterX=static_cast<double>(Crossing.centerXSubvoxels)
-                            /static_cast<double>(SubvoxelScale);
-                        const double CenterY=static_cast<double>(Crossing.centerYSubvoxels)
-                            /static_cast<double>(SubvoxelScale);
-                        if(CenterX+Radius<TileMinimum[0]/static_cast<double>(SubvoxelScale)
-                            ||CenterX-Radius>TileMaximum[0]/static_cast<double>(SubvoxelScale)
-                            ||CenterY+Radius<TileMinimum[1]/static_cast<double>(SubvoxelScale)
-                            ||CenterY-Radius>TileMaximum[1]/static_cast<double>(SubvoxelScale))continue;
-                        RelevantMinimum=std::min(RelevantMinimum,
-                            static_cast<double>(Crossing.bottomZSubvoxels)
-                                /static_cast<double>(SubvoxelScale));
-                        RelevantMaximum=std::max(RelevantMaximum,
-                            static_cast<double>(Crossing.topZSubvoxels)
-                                /static_cast<double>(SubvoxelScale));
-                    }
                     // The native heightfield band only covers the owner terrain and its
                     // manifest features. Resolved SDF stamps can carve below that band or
                     // add a floor there, so include every overlapping stamp's bounded Z
@@ -2804,7 +2743,6 @@ namespace VoxelForgeCore
         FStandaloneShapeWorld::FScopedPassageTileQueries passageFeatureScope(
             tileShape,queryMinimum,queryMaximum,
             options.profilePassageComposer?&out.diagnostics:nullptr,key.lod,
-            options.diagnosticPassageSourceFloorVoxels,
             options.diagnosticPassageCarveReachMetres);
         passageFeatureScope.prepareSweptTubeBase(origin,MeshStep,MeshCells,
             static_cast<std::uint32_t>(world.Seed));
@@ -2904,10 +2842,6 @@ namespace VoxelForgeCore
             {
                 out.diagnostics.passageTileRouteSegmentCandidates=
                     passageFeatureScope.routeSegmentCandidateCount();
-                for(std::size_t passageIndex=0;
-                    passageIndex<impl_->resolvedLayout->passages.size();++passageIndex)
-                    out.diagnostics.passageTileMouthSegmentCandidates+=
-                        passageFeatureScope.mouthSupportSegments(passageIndex).size();
             }
         }
         FGridFillWorkerContext GridFillWorkerContext;

Standalone/Research/WorldAssemblerM3/Astra2/PROMPT.md

@@ -0,0 +1,60 @@
+# VoxelForge — read-only design review: stamp tunnels along the curve instead of searching the curve per sample
+
+You are reviewing a design idea against real code. **Do not edit, create or delete any file, and do not build or run
+anything.** Read code and reason. Be thorough; mark each claim as [CODE] (with file:line), [MEASURED] (from a report),
+or [INFERENCE].
+
+## Context
+
+VoxelForge Core (`Standalone/Core`) generates a density field: negative = solid, positive = air, 25 cm voxels, 32³ tiles,
+LOD0–8 clipmap, marching cubes. Same seed must give the same world on every machine and worker count; byte-identity with
+older versions is NOT required.
+
+Inter-strate tunnels ("passages") on branch `wip/l111-halfpipe-spline` (HEAD `707bc61`) are a smooth spline swept tube:
+- a D-shaped walkable floor with a continuous floor height h(s), max grade 15°;
+- a rounded crest;
+- an open-air half-pipe where there is no rock;
+- a thin solid shell ("plain shell rule": interior always air, a thin shell always solid);
+- smooth-min where tunnels meet;
+- mouths that ramp into the ground.
+
+Key code: `Standalone/Core/src/ShapeWorldInternal.h` (`FScopedPassageTileQueries`, `prepareSweptTubeBase`, around
+lines 374 and 11436), `Standalone/Core/src/VoxelForgeCore.cpp:~2747`. Reports: `Standalone/Research/WorldAssemblerM3/Linux109`
+… `Linux119` (REPORT.md each).
+
+Cost today [MEASURED, L117–L119]: the smooth tunnel is 13–22% slower than the coarse L109 tunnel at LOD0–2 readiness.
+About 95% of the extra cost is in passage-feature tiles. The causes: sweep pairs (sample × curve segment), exact
+fallbacks that re-scan segments for off-grid queries, segment visits, per-tile prepare, and smooth-min.
+
+## The idea to review (owner + orchestrator)
+
+Today each sample "pulls": it searches for the nearest curve segment and computes a distance. Proposed: "push"/stamp.
+For each tile, walk the curve in rings (frames: position, forward, up, as with an Unreal spline mesh) and write the
+tube's signed distance into only the lattice samples near each ring. Only a narrow band of about one voxel around the
+wall needs an exact value; deep rock only needs "rock" and the tunnel interior only needs "air". Off-grid queries
+(normals/gradients at seams, gameplay point queries) would read the stamped lattice, interpolating, instead of
+re-searching the curve. Also: accept curve accuracy at one voxel (25 cm) rather than 2.5–10 cm, and build the curve
+frames once per route rather than per tile.
+
+## Questions
+
+1. Is the cost model right? From the code, what does a passage-feature tile do per sample today, and what would stamping
+   replace? Estimate the work ratio (samples touched × ops) for a typical LOD0 tile and for LOD1–2.
+2. What breaks or gets harder, concretely in this code?
+   - box verdicts/tile skipping (`ClassifyTile`-style proofs that a tile is all solid/air);
+   - seams between tiles: both neighbours must stamp identical values at shared samples, deterministically;
+   - LOD levels (coarser lattices);
+   - off-grid `DensityAtUnclamped` queries and gradients;
+   - min/smooth-min with the rest of the field and with other tunnels;
+   - the floor h(s), half-pipe, shell, mouths;
+   - walk audits;
+   - thread safety.
+3. Is "write into the lattice" compatible with how density is evaluated (lattice fill vs point evaluation)? If the
+   pipeline is pull-only, what is the smallest change that gets the push benefit? For example, a per-tile sparse
+   band list built by walking rings, then lookups during fill.
+4. Is there a simpler or cheaper alternative we're missing? For example, an analytic local frame per sample from the
+   spline parameter, or a coarse per-tile segment window that is already almost free.
+5. Recommendation: the design you would build, in steps, with the evidence bar (what measurement would prove or kill
+   it). Keep it short.
+
+Write your answer as plain Markdown, at most ~1,200 words.

Standalone/Research/WorldAssemblerM3/Astra2/REVIEW.md

@@ -0,0 +1,53 @@
+[INFERENCE] **Build a tighter passage cache first.** Ring stamping is promising, but the code already pushes segment contributions into a tile lattice. The main opportunity is reducing coverage, overlap, and off-grid fallback work—not reversing a purely pull-based architecture.
+
+References below: `H` = `Standalone/Core/src/ShapeWorldInternal.h`; `C` = `Standalone/Core/src/VoxelForgeCore.cpp`; `M` = `Standalone/Source/VoxelForge/Private/VoxelMarchingCubesMesher.cpp`. Report references mean `Standalone/Research/WorldAssemblerM3/LinuxNNN/REPORT.md`.
+
+**1. Cost model**
+
+[CODE] `prepareSweptTubeBase` allocates per-passage `(cells+3)³` sample records, including a one-step halo. It walks candidate bounding boxes, lattice samples, then fine segments; individual segment bounds reject some pairs. Accepted pairs compute projection, deterministic square root, radius, floor profile, and potentially shell distance. A final pass flattens the selected tube and caches relief (`H:11436`, `H:11489`, `H:11521`, `H:11659`).
+
+[CODE] During density evaluation, aligned samples retrieve that cache, then still perform relief weighting, floor shaping, mouth fades, shell composition, and host-field composition. Misses invoke two hierarchy searches: nearest body/floor and nearby shells—not necessarily a linear whole-route scan (`H:1247`, `H:1346`, `H:1427`, `H:1435`).
+
+[MEASURED] L119 reports 29.40M preparation pairs and 24.46M segment visits across its four LOD0–2 profiles. Preparation fell 59.8% versus L118, while fallbacks increased 55–58%. L117 attributes about 95% of the tile-worker regression to passage-feature tiles. Thus the diagnosis is sound, but those counters do not establish equivalent costs per operation.
+
+[INFERENCE] For an illustrative straight crossing, assume exposed surface perimeter \(P=16\) m, tile crossing length \(L=32\Delta\), and exact band ±one lattice step. Unique band samples are approximately \(2PL/\Delta^2\):
+
+| [INFERENCE] Normal tile | LOD0 | LOD1 | LOD2 |
+|---|---:|---:|---:|
+| Spacing Δ | 0.25 m | 0.5 m | 1 m |
+| Band samples | 4,096 | 2,048 | 1,024 |
+| Fraction of 35³ prepared grid | 9.6% | 4.8% | 2.4% |
+
+[INFERENCE] These are geometric estimates, not measured “typical tiles”; clipping, floor/outer-shell surfaces, bends, and crossings change them. If today fraction \(f\) of the grid receives \(k\) segment evaluations, and stamping averages \(m\) contributions per band sample, the expensive-work ratio is approximately `(table fraction × m)/(f × k)`, multiplied by relative operation cost. For `f=0.5, k=2, m=2`, that is 19%, 9.6%, 4.8%. Allocation, classification, host evaluation, and composition remain. Dense rings can make \(m\) large and erase this advantage.
+
+**2. Concrete difficulties**
+
+- [CODE] **Skipping:** passage-bound overlap currently vetoes a uniform proof and returns `Mixed` (`H:5684`). [INFERENCE] Retain conservative bounds, expanded for approximation and reconstruction support. “No band samples written” proves neither solid nor air: a thin shell can lie between samples.
+
+- [CODE] **Seams:** cache coordinates derive from integer world coordinates; nearest ties use segment index (`H:11448`, `H:11564`, `H:11640`). [INFERENCE] Generate rings from route-global parameters, never tile-local stepping. Include every contributor affecting shared samples and halos; use stable ordering and arithmetic. Independently clipping ring generation at tile boundaries risks disagreement.
+
+- [INFERENCE] **LOD:** distinguish 25 cm curve error from lattice spacing. A fixed 25 cm band is insufficient around 1 m cells; cover interpolation corners and gradient support. A 0.75 m shell can disappear at coarse resolution. Independently reconstructed LOD fields also disagree off-grid; define one canonical point-query field.
+
+- [CODE] **Off-grid:** `sweptSample` rejects non-lattice coordinates; public point queries call `DensityAtUnclamped` without requiring a tile cache (`H:11721`; `C:1875`). Gradients normally use density-grid values, with unclamped fallback (`M:1972`). [INFERENCE] Interpolation needs all eight valid corners, including outside loaded tiles. It changes gradients and zero crossings; interpolating only passage channels also differs from interpolating the fully composed field.
+
+- [CODE] **Composition:** passage bodies and shells use separate hard minima; final composition includes host/native-floor exceptions (`H:1496–1519`). The floor corner uses smooth-max, implemented through smooth-min (`H:3493`, `H:6940`). [INFERENCE] Do not assume the measured smooth-min cost is tunnel-to-tunnel joining. Preserve separate void/shell channels and composition order. Arbitrary “rock/air” magnitudes can move interpolated crossings and smooth blends.
+
+- [CODE] **Shape:** floor selection is independent of minimum round distance; the code explicitly warns that minimizing flattened segment distances makes downhill floors jump (`H:1328`). Floor evaluation uses horizontal projection, authored `h(s)`, and XY mouth blending (`H:3448`). Shells have rounded crests and five-metre mouth fades (`H:1461`, `H:3503`). [INFERENCE] A conventional spline-normal D-section changes this geometry. Stamped isolated rings can scallop; rasterize bounded slabs between frames while preserving gravity-relative floors, shell unions, and interior priority.
+
+- [MEASURED] **Walkability:** L119’s maximum sampled floor-root offset was 7.884 mm; L113 documented voxel-aligned floor ribs. [INFERENCE] A 25 cm curve tolerance does not authorize 25 cm floor error or guarantee ≤15° grade. Audit reconstructed density and meshes, including mouths, overlaps, support, and headroom—not only planner `h(s)`.
+
+- [CODE] **Threads:** copied worker scopes share prepared data through a shared pointer; active scope is thread-local (`H:353`, `H:11411`; `C:2854`). [INFERENCE] Finish stamping before publication. Parallel ring writes require deterministic reductions; mutable shared lazy caches introduce synchronization and lifetime problems.
+
+**3. Smallest useful change**
+
+[CODE] Preparation already precedes `GenerateMesh`, whose fill calls lattice evaluators (`C:2747`, `C:2892`; `M:1488`). Route segments are already prepared globally with 10 cm chord tolerance and an 8 m spacing cap (`H:4601`).
+
+[INFERENCE] Replace preparation’s broad pair enumeration with conservative slab-to-grid spans or small spatial bins. Keep indexed lookups during pull-based fill. Store body/floor ownership and shell contributions separately; use explicit outside/interior/band states. A dense index plus compact records may beat a sparse hash map. Additional frames can be route-global, but this does not eliminate existing per-tile rasterization.
+
+**4. Alternatives and recommendation**
+
+[MEASURED] L115 already reduced coarse candidates 74.1% without reducing fine pairs. L118’s restriction of fallbacks to prepared spans changed output; L116’s 20.8% pair reduction did not recover readiness.
+
+[INFERENCE] An analytic local frame still needs a parameter lookup, especially near folds. Prefer globally indexed conservative segment bins, shared by preparation and exact point queries.
+
+[INFERENCE] Build in three steps: instrument unique samples, overlap, misses by caller, memory, and per-LOD time; implement tighter bins/slabs while preserving geometry; then separately trial 25 cm curve tolerance and canonical interpolation. Accept only reproducible near-readiness gains against L119/L109 plus unchanged walk/seam requirements and cross-compiler/worker determinism. Kill the approach if saved distance work becomes rasterization, memory, or reconstruction cost.

Standalone/Research/WorldAssemblerM3/Astra3/PROMPT.md

@@ -0,0 +1,53 @@
+# VoxelForge — read-only design review: per-block bit flags (masks) to skip work
+
+You are reviewing an idea against real code. **Do not edit, create or delete any file, and do not build or run
+anything.** Read code and reason. Be thorough; mark each claim as [CODE] (with file:line), [MEASURED] (from a report),
+or [INFERENCE].
+
+## Context
+
+VoxelForge Core (`Standalone/Core`) generates a density field: negative = solid, positive = air, 25 cm voxels, 32³ tiles,
+LOD0–8 clipmap, marching cubes. The production mesher for the standalone path is in `Standalone/Source/...`, and Core is
+also compiled into Unreal. Same seed must give the same world on every machine and worker count; byte-identity with
+older versions is NOT required.
+
+Branch `assembler/world` (HEAD `7022d02`'s parent line, currently `4041d4b`) has the smooth inter-strate tunnels.
+Readiness target: LOD0–2 near terrain in ~1 s on arrival. Measured today: SW ~0.7–0.8 s, VS ~0.9 s, FI/TN ~1.0 s
+headless. In Linux UE, start→LOD0–2 is ~2.2–2.5 s. At 30 m/s in UE, near coverage is ~40%.
+
+Reports to consult: `Standalone/Research/WorldAssemblerM3/Linux1*/REPORT.md`, especially:
+- L77/L87/L117, cost census;
+- L124, shape gate and per-strate timings;
+- L125/L126/L128, tunnel prepare: per-segment work, 67–73% pair overlap, support up to 7.9 m;
+- L128, where an exact one-owner slab scheme halved pairs but got slower because of per-sample ownership tests.
+
+Tile skipping already exists at tile level: box verdicts / classify, where all-solid or all-air tiles are skipped.
+
+## The idea (from the owner)
+
+Keep bit flags per block (1 bit per voxel, or per small brick of e.g. 4³ or 8³ voxels): for example "certainly deep
+solid", "certainly deep air", "near a surface", "touched by a tunnel", "already evaluated". Use them to skip work in
+generation and meshing, and possibly elsewhere (collision, streaming, edits, AI/nav, lighting).
+
+## Questions
+
+1. Where does time actually go per tile today, and which of that work a mask could skip? Cover density fill,
+   classification, mesher (marching cubes, normals/gradients, skirts, seam refinement), passage prepare, and composition.
+   Use measured census numbers where they exist.
+2. Which flags are sound? A flag must be a proof, never a guess. A mask "certainly solid" must be derived from a
+   conservative bound (the bound must be the supremum of what the evaluator produces), or the mesher will delete
+   surfaces. Which bounds already exist in the code (box verdicts, interval bounds) that a brick-level mask could reuse?
+   Name the existing hazards. For example, `Identity` means `Sdf >= T`, and `ClassifyTile` proving below-TerrainZ solid
+   would delete 3D caves.
+3. Granularity: per-voxel bits vs per-brick (4³/8³) vs per-column. What does each cost to build, versus what it
+   saves? Consider SIMD and cache behaviour. Marching cubes needs only cells whose corner signs differ, and gradients
+   need a halo.
+4. "Already evaluated" or "checked" flags: where would they be correct (memoisation of identical queries), and where
+   would they be wrong (nearest-of-many queries like the tunnel segments)? Is there repeated identical evaluation today,
+   for example halo samples shared between neighbouring tiles, or skirt/seam re-queries?
+5. Other uses: player edits (diff layer), collision-only server tiles, streaming priority, nav. Which would a mask
+   genuinely help, given the multiplayer rule (replicate seed + layout + diff, never geometry)?
+6. Recommendation: the one or two mask designs most likely to pay off, with the expected gain (a range, marked
+   inference), the measurement that would prove or kill each, and the risks. Keep it short.
+
+Write plain Markdown, at most ~1,200 words.

Standalone/Research/WorldAssemblerM3/Astra3/REVIEW.md

@@ -0,0 +1,83 @@
+[INFERENCE] **Improve the existing brick-proof system; don’t add a general “checked voxel” mask.** The promising savings are avoided field/composition work and passage preparation, rather than marching-cubes classification.
+
+References below abbreviate these files: [M = production mesher](/home/jahni/VoxelM/Plugins/VoxelForge/Standalone/Source/VoxelForge/Private/VoxelMarchingCubesMesher.cpp), [S = ShapeWorldInternal.h](/home/jahni/VoxelM/Plugins/VoxelForge/Standalone/Core/src/ShapeWorldInternal.h), [C = VoxelForgeCore.cpp](/home/jahni/VoxelM/Plugins/VoxelForge/Standalone/Core/src/VoxelForgeCore.cpp), [O = VoxelDensityOpStack.cpp](/home/jahni/VoxelM/Plugins/VoxelForge/Standalone/Source/VoxelForge/Private/VoxelDensityOpStack.cpp). L-numbers identify `Standalone/Research/WorldAssemblerM3/LinuxN/REPORT.md`.
+
+[INFERENCE] Evidence limitation: Linux128 is absent from this checkout. Its supplied 7.9 m support and slower one-owner result remain unverified here. The reports below span different revisions; their percentages must not be presented as today’s census.
+
+**1. Where time goes**
+
+[MEASURED] L77’s exclusive FI/TN LOD0–2 worker account:
+
+| Work | FI | TN |
+|---|---:|---:|
+| Total worker CPU/view | 23.02 s | 21.50 s |
+| Passage composition | 65.9% | 65.4% |
+| Support | 14.2% | 13.5% |
+| Base field | 3.2% | 3.3% |
+| Bands / dispatch | 2.3% / 3.7% | 2.6% / 3.6% |
+| MC / skirts / refinement, excluding field work | 1.6% / 0.1% / 7.5% | 1.2% / 0.2% / 8.6% |
+
+[MEASURED] Grid fill occupied 19.058/17.377 worker-seconds, **overlapping** those phase categories. L117’s later census found support zero; its smooth-tunnel regression concentrated about 95% of added tile CPU in passage-feature tiles. Cell classification stayed flat; gradients became cheaper. Thus L77 identifies historical leverage, not current percentages.
+
+[MEASURED] L125 preparation averaged 514/130/4.1 ms worker work at LOD0/1/2. LOD0 evaluated 3.11–8.49 million segment/sample pairs with 67–72% overlap, although the exact band occupied only 5–7% of grid samples. L126 reduced floor-profile calls 20–23% without useful readiness improvement.
+
+[CODE] Brick skipping already exists: configurable block verdicts, required-sample masks, and active-block gradient halos (M:646,1053). MC already rejects cells without sign crossings before interpolation/gradients (M:2186). Gradient memoisation exists (M:1956). Skirts issue separate density queries (M:2491); refinement has its own stage (C:2975).
+
+[INFERENCE] A pre-evaluation proof can save density, composition, and subsequent MC work. A sign mask constructed after density evaluation saves only downstream scanning. Classification itself costs work; no isolated current classification percentage is established by these reports.
+
+[MEASURED] L124’s shape gate alone took 179–226 ms, with no near-tile admissions before completion. L87 additionally measured a 107–153 ms far-parent gate and no headless collision wait. Tile masks cannot directly remove these gates or explain the supplied UE/headless gap.
+
+**2. Sound flags and hazards**
+
+[INFERENCE] For final density \(D\), solid requires an upper bound covering the evaluator’s supremum, \(U<0\); air requires a valid lower bound \(L\ge0\). “Deep” requires an explicit margin. Unknown must stay unknown. A crossing proves an MC-active cell; uniform sampled signs do not prove an entire continuous volume surface-free.
+
+[CODE] Reusable proofs include:
+
+- S:5428: halo-expanded box classification.
+- S:5974: exact surface-column extrema over the MC lattice.
+- S:6019: floating-island density intervals.
+- S:5197,6032: native cave intervals.
+- S:5949: stamp interval composition with outward numerical padding.
+- O:6998: operation-specific inactivity from SDF lower bounds.
+
+[CODE] These proofs have guards: passage overlap returns unknown (S:5687); unsupported stamp composition returns unknown (S:5655); stratum crossings are rejected (S:5900). The mesher distinguishes uniform candidates from `bSkipSafe` certificates (M:715,756), and otherwise validates candidates against actual samples (M:1838).
+
+[CODE] For gated modifiers, `Identity` can mean `Sdf >= T`, not `abs(Sdf) >= T` (O:1496,6998). Noise bounds cannot assume nominal `[-1,1]`: FBM and unclamped cellular bounds are explicitly discussed at O:848,301.
+
+[INFERENCE] Additional hazards: confusing internal SDF with final density; certifying base terrain before caves, tunnels, shells, stamps or edits; using lattice proofs for off-grid queries; ignoring rounding or gradient halos. A below-`TerrainZ` solid rule applied to 3D caves deletes caves. “Touched by tunnel” should mean conservative candidate overlap; only **proven untouched** authorizes omission.
+
+**3. Granularity**
+
+[INFERENCE] Storage per flag for 32³ cells:
+
+| Granularity | Storage | Construction/tradeoff |
+|---|---:|---|
+| Per cell | 4 KiB | 32,768 decisions; expensive unless derived during necessary work |
+| 4³ bricks | 64 B | 512 proofs; tighter boundaries, more proof/halo overhead |
+| 8³ bricks | 8 B | 64 proofs; cheaper, more mixed bricks |
+| XY columns | 128 B | 1,024 decisions; useful mainly for heightfield-specific proofs |
+
+[INFERENCE] These exclude corner/halo samples, metadata and additional flags. A column bit cannot represent multiple cave intervals. Brick runs preserve contiguous X sampling and SIMD better than scattered per-sample branches; bit packing alone does not establish a speedup.
+
+[CODE] The existing implementation uses byte arrays and preserves exact samples around active bricks (M:1053–1083). Replacing densities with arbitrary same-sign constants near active cells would corrupt interpolated positions or normals.
+
+**4. “Already evaluated”**
+
+[CODE] Passage preparation precedes meshing (C:2747). Its valid sentinel is initialization, not completion: subsequent segments update floor selection, shell minima and round-distance winners, with deterministic tie-breaking (S:11520,11615,11688).
+
+[INFERENCE] Therefore first-visitor wins is invalid. Memoisation is valid only for identical queries against identical seed/layout/evaluator/edit revision and query semantics; cache the value, not just a bit.
+
+[MEASURED] L77 counted 8.6–8.9% halo points, without measuring duplicate coordinates. L125 verified 4,900 shared samples across four neighbour pairs and still recorded substantial off-grid skirt/midpoint fallbacks.
+
+[CODE] Shared-grid reuse and gradient validity caches already exist (M:1031,1956). [INFERENCE] Skirt/seam queries are reuse candidates only when coordinates and semantics actually match; “same voxel” is insufficient.
+
+**5. Other uses**
+
+[INFERENCE] Edit-dirty bricks can localize regeneration, with invalidation expanded through normals, seams and collision dependencies. Collision-only servers can skip certified empty regions, but still need boundary geometry. Streaming can prioritize potentially occupied bricks; nav and lighting need clearance/connectivity or visibility data beyond occupancy. Keep masks derived locally from replicated seed + layout + diff, versioned with that state.
+
+**6. Recommendation**
+
+- [INFERENCE] **Tune existing 4³/8³ certificates**, caching proofs and emitting contiguous required-sample runs. Plausible incremental near-ready saving: **0–5%**, with regressions possible. Measure proof time, samples avoided, halo expansion, total worker CPU and UE readiness.
+- [INFERENCE] **Pass required-region information into passage preparation**, rejecting whole segment/brick regions conservatively. Plausible near-ready saving: **0–10%**, highly uncertain because passage overlap currently vetoes proofs. Kill it if pair reduction fails to reduce preparation and end-to-end time.
+
+[INFERENCE] Both require seam/off-grid validation, adversarial bound checks, edit invalidation tests, and cross-machine/worker determinism. Preserve canonical tie-breaking; old-version byte identity need not constrain a sound redesign.

Standalone/Research/WorldAssemblerM3/Astra4/PROMPT.md

@@ -0,0 +1,37 @@
+# VoxelForge — open read-only review: what should be done and focused on now?
+
+You are an outside reviewer. **Do not edit, create or delete any file, and do not build or run anything.** Read the
+repository and its reports and reason. Be thorough; mark each claim as [CODE] (with file:line), [MEASURED] (from a
+report), or [INFERENCE].
+
+## The project, in the owner's words and goals
+
+- VoxelForge is a UE5 density-field voxel plugin for a **multiplayer** game. The world is made of stacked "strates"
+  (layers) linked by tunnels. The standalone C++ generator is "Core" (`Standalone/Core`), and it is also compiled into
+  Unreal. Read `CLAUDE.md`, `CODEMAP.md`, `ARCHITECTURE.md`, `GDD.md`, `COMPOSER-NOTES.md` and `fable-idea.md` at the
+  plugin root for the game and design intent. The research reports are in
+  `Standalone/Research/WorldAssemblerM3/Linux*/REPORT.md` and the `Astra*/` reviews.
+- Owner goals stated over time:
+  - the near terrain (LOD0–2) ready in ~1 s on arrival, then far terrain keeps filling in at low priority;
+  - same seed gives the same world on every machine (byte-identity with older versions is NOT required);
+  - smooth for VR and high-speed travel;
+  - all generators (surface, floating islands, caves, dreamcore/liminal, etc.) working in Core;
+  - the system should eventually *invent* strates and worlds from tagged authored content plus a seed;
+  - keep things simple and avoid over-complication.
+- The current main line is branch `assembler/world`.
+
+## Your task
+
+Look at the whole: code, reports, design docs, and the direction of the last few dozen rounds. Tell the owner what
+should be done and focused on now, whatever it is. That may be performance, features, generation quality, gameplay
+foundations, architecture, technical debt, tooling, process, or stopping something. Don't assume the recent focus is
+the right one; question it if the evidence says so.
+
+Answer:
+1. **Where the project actually is**: what works, what's fragile, what's missing for the game to exist.
+2. **The top 3–5 things to focus on now**, ranked, each with why, a rough size, what "done" looks like, and how to
+   tell it worked.
+3. **What to stop or deprioritise**, and why.
+4. **Risks or debts** that will bite later if ignored.
+
+Write plain Markdown, at most ~1,500 words, readable by a non-specialist owner (explain jargon briefly).

Standalone/Research/WorldAssemblerM3/Astra4/REVIEW.md

@@ -0,0 +1,72 @@
+[INFERENCE] **Focus now on a small, persistent multiplayer descent, supported by end-to-end Unreal performance work.** VoxelForge has substantial terrain technology. The larger uncertainty is whether it delivers a reliable, interesting shared expedition. Another sequence of tunnel optimizations will not resolve that.
+
+Read-only review of `assembler/world` at `044cd6d`; no files changed and no builds, tests, or project execution. [CODE] identifies repository evidence; design-document references establish intent, not implementation. [MEASURED] means reported results, not independently reproduced measurements.
+
+**1. Where the project actually is**
+
+[CODE] **Core is real and integrated, but generator implementation remains mixed.** Unreal compiles the canonical standalone sources through validated wrappers. Core dispatches to surface, floating-island and native graph evaluators, with a legacy-generator fallback when appropriate. That is useful coverage, but does not mean every family has a complete native implementation. See [VoxelForgeCore.Build.cs:15](/home/jahni/VoxelM/Plugins/VoxelForge/Source/VoxelForgeCore/VoxelForgeCore.Build.cs:15) and [ShapeWorldInternal.h:600](/home/jahni/VoxelM/Plugins/VoxelForge/Standalone/Core/src/ShapeWorldInternal.h:600).
+
+[MEASURED] **Passage correctness and speed have improved substantially.** L109 simplified tunnels and removed support-restoration machinery, reducing floating-island and tunnel-network near-generation times by approximately 31–38%. L125 reports 21,425/21,425 open route samples, a maximum sampled floor offset of 7.884 mm, and matching cross-compiler/worker hashes. These are meaningful engineering achievements. [L109](Standalone/Research/WorldAssemblerM3/Linux109/REPORT.md), [L125](Standalone/Research/WorldAssemblerM3/Linux125/REPORT.md).
+
+[MEASURED] **The one-second player experience is not achieved.** L125’s standalone near-ready medians were approximately 0.78–1.03 seconds. L127’s Unreal startup took 2.22–2.51 seconds; at 30 m/s, median near coverage fell from 43.36% to 40.22% after the smooth-tunnel merge. Both versions recovered after stopping. L130 found no measurable timing improvement and skipped Unreal movement. These are different workloads; subtracting their times would not isolate a particular overhead. [L125](Standalone/Research/WorldAssemblerM3/Linux125/REPORT.md), [L127](Standalone/Research/WorldAssemblerM3/Linux127/REPORT.md), [L130](Standalone/Research/WorldAssemblerM3/Linux130/REPORT.md).
+
+[CODE] **Multiplayer foundations exist; the multiplayer lifecycle remains unfinished here.** There are season hashes, local terrain-edit entry points, and streaming-anchor infrastructure. The architecture explicitly leaves join replication, authoritative edit messages, late-join snapshots and networked season changes open. The edit entry point directly dispatches to local modification paths. [VoxelWorld.h:574](/home/jahni/VoxelM/Plugins/VoxelForge/Source/VoxelForge/Public/VoxelWorld.h:574), [VoxelWorld.cpp:15771](/home/jahni/VoxelM/Plugins/VoxelForge/Source/VoxelForge/Private/VoxelWorld.cpp:15771), [ARCHITECTURE.md:2408](/home/jahni/VoxelM/Plugins/VoxelForge/ARCHITECTURE.md:2408).
+
+[CODE] **The composer is more than a proposal, but less than the full content vision.** It rolls candidates, measures them and rejects failures. The intended system additionally combines tagged geometry, materials, decorations, creatures and audio; theme coherence remains an open design item. [VoxelSeasonManifest.cpp:2523](/home/jahni/VoxelM/Plugins/VoxelForge/Source/VoxelForge/Private/VoxelSeasonManifest.cpp:2523), [COMPOSER-NOTES.md:1512](/home/jahni/VoxelM/Plugins/VoxelForge/COMPOSER-NOTES.md:1512).
+
+[INFERENCE] **This repository demonstrates a capable terrain platform, not yet a validated game loop.** I cannot establish what gameplay exists outside the plugin. The GDD itself identifies “Is the descent fun?” as its first unresolved question. [GDD.md:441](/home/jahni/VoxelM/Plugins/VoxelForge/GDD.md:441).
+
+**2. Ranked priorities**
+
+All sizes below are [INFERENCE]: rough focused engineering estimates, excluding substantial art production.
+
+**1 — Deliver one multiplayer expedition that survives reconnecting. Size: 3–6 weeks.**
+
+[INFERENCE] Use two contrasting strates and their connecting tunnel. Two players descend, find one destination, uncover or collect something, return, and retain their changes after restart. Include a third player joining after terrain has changed. Avoid expanding immediately into the city economy or community backend.
+
+[INFERENCE] The necessary foundation is a server-approved world identity, ordered authoritative edits, a saved snapshot plus subsequent edits, and collision around separated players. Give edits stable identifiers so retransmission cannot apply them twice. Reuse the existing local edit and anchor systems.
+
+[INFERENCE] **Done:** host/client edits agree; a late joiner reconstructs the same terrain; restart preserves it; incompatible world versions fail clearly; remote players have ground when away from the host.
+
+[INFERENCE] **Success:** a complete shared expedition without terrain disagreement, lost progress or collision failures—and players can explain what made them want to descend again. This tests the project’s central assumptions much earlier than a complete RPG would.
+
+**2 — Close the Unreal arrival and movement gap. Size: 1–3 weeks initially.**
+
+[MEASURED] L122 found idle generation capacity while admission was blocked; L123 measured approximately 19 ms from region request to game-thread observation despite sub-millisecond median compilation. Simple lookahead and admission relaxations failed their combined coverage/frame-time gates. L121’s modest extra far-apply lane did improve far coverage. [L121](Standalone/Research/WorldAssemblerM3/Linux121/REPORT.md), [L122](Standalone/Research/WorldAssemblerM3/Linux122/REPORT.md), [L123](Standalone/Research/WorldAssemblerM3/Linux123/REPORT.md).
+
+[INFERENCE] First establish one current-build, rendered Unreal baseline on intended hardware. Measure arrival through **visible terrain and usable collision**, then walking, 30 m/s travel, turns, teleporting and edits. Record near coverage, far gaps, frame-time spikes and memory separately. NullRHI measurements, which omit rendering, cannot certify VR comfort.
+
+[INFERENCE] Investigate region completion, admission and publication before another broad density rewrite. A game-thread continuation is a candidate experiment, not a proven solution; keep world mutation on its owning thread.
+
+[INFERENCE] **Done:** approximately one-second LOD0–2 readiness on the agreed hardware, safe collision, steadily filling far terrain, and an explicit movement-coverage target without worsening frame spikes. **Success:** repeated rendered runs meet those limits, including with remote-player collision work. Do not redefine “ready” as only the nearest handful of tiles.
+
+**3 — Turn generator breadth into playable variety. Size: 2–4 weeks for the first content slice.**
+
+[INFERENCE] Establish a compact capability matrix for each family: Core dispatch, entrances/exits, interior connectivity, edits, materials/content, far terrain and performance. Include dreamcore/liminal explicitly; motif generation alone does not establish a convincing anomalous place.
+
+[MEASURED] L58 illustrates the distinction: its lattice route passed reachability checks, but the review described a corkscrew tunnel rather than the intended open vertical shaft. Technical validity and identity are separate acceptance criteria. [L58](Standalone/Research/WorldAssemblerM3/Linux58/REPORT.md).
+
+[INFERENCE] Extend the existing offline composer with a small tagged library: a grounded cave, an island environment and one liminal identity, each with coherent materials, sound and a destination. Preserve owner review before publishing a season.
+
+[INFERENCE] **Done:** several seeds produce recognizably different, coherent places with findable progression and usable encounter spaces. **Success:** players navigate them without developer guidance and distinguish their identities without reading labels. Expand families after this works.
+
+**4 — Consolidate the evidence and remove obsolete complexity incrementally. Size: 2–4 days, then ongoing.**
+
+[INFERENCE] Maintain one short current-state page: shipped behavior, open defects, active baseline, hardware and next acceptance test. Keep historical reports as history. Remove rejected experimental paths when they no longer serve a specific comparison; extract stable responsibilities when touching oversized files.
+
+[INFERENCE] **Done:** a contributor can identify the current evaluator, runtime path, multiplayer gaps and performance limits without reconstructing 130 rounds. **Success:** new work no longer repeats rejected experiments or treats old measurements as current.
+
+**3. Stop or deprioritise**
+
+- [INFERENCE] **Pause open-ended tunnel micro-optimization.** L125/L126 missed the older speed bar; L130’s reach correction produced no measurable speedup. Keep correctness fixes, but require a current cost account and a bounded experiment before further optimization. [MEASURED: L126](Standalone/Research/WorldAssemblerM3/Linux126/REPORT.md), [L130](Standalone/Research/WorldAssemblerM3/Linux130/REPORT.md).
+- [INFERENCE] **Do not implement Astra recommendations mechanically.** Astra1’s support-heavy model predates L109’s simplification. Astra2/3 offer hypotheses, not measured wins for today’s code. Re-profile before adopting them.
+- [INFERENCE] **Defer GPU generation, a mesher replacement, elaborate discovery systems and large-scale backend work.** None currently resolves the missing shared expedition.
+- [INFERENCE] **Stop requiring old-version byte identity for deliberate generation changes.** Preserve peer agreement within a version and explicit versioned saves; permit reviewed geometry improvements.
+
+**4. Debts that will bite later**
+
+[INFERENCE] **Determinism needs a deployment contract.** Linux Clang/GCC and worker-count agreement is strong evidence, not proof across every intended machine. Include Windows/compiler/CPU combinations actually supported. World identity must cover evaluator, content, layout, scale and edit state—not merely seed. [MEASURED: L130](Standalone/Research/WorldAssemblerM3/Linux130/REPORT.md).
+
+[CODE] **Scale and fallback complexity remain active integration risks.** Core is opt-in, and runtime queries bridge different coordinate conventions. [VoxelWorld.h:300](/home/jahni/VoxelM/Plugins/VoxelForge/Source/VoxelForge/Public/VoxelWorld.h:300), [VoxelWorld.cpp:15612](/home/jahni/VoxelM/Plugins/VoxelForge/Source/VoxelForge/Private/VoxelWorld.cpp:15612). [INFERENCE] Test edits, content placement and collision across those boundaries before broadening authoring.
+
+[INFERENCE] **A passing route audit does not prove a whole strate is enjoyable, connected or suitable for six-player combat.** Retain geometric checks, but add actual traversal, destination discovery and encounter-space evaluation. Those are now more valuable than another polished tunnel benchmark.

Standalone/Research/WorldAssemblerM3/Linux109/MeasureL109FloorRamp.py

@@ -0,0 +1,111 @@
+#!/usr/bin/env python3
+"""Measure the fixed-depth floor and derived centreline against the authored grade."""
+from __future__ import annotations
+
+import json
+import math
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[4]
+LAYOUT = ROOT / "Saved/Linux/l109/audit-layout-final-stamp.json"
+OUT = ROOT / "Saved/Linux/l109/floor-ramp-grade-final.json"
+MOUTH_RAMP_METRES = 5.0
+
+
+def route_m(layout: dict, passage: dict):
+    scale = layout["core_voxel_size_um"] / 256.0 / 1_000_000.0
+    controls = [(tuple(float(v) * scale for v in row["position_q"]),
+                 float(row["radius_q"]) * scale) for row in passage["control_points"]]
+    lower_q = passage["lower_mouth_q"]
+    player_center_above_floor_q = round(
+        1.13 * 256.0 * 1_000_000.0 / layout["core_voxel_size_um"])
+    lower_index = next((index for index, row in enumerate(passage["control_points"])
+                        if row["position_q"][0] == lower_q[0]
+                        and row["position_q"][1] == lower_q[1]
+                        and row["position_q"][2] - row["radius_q"]
+                            + player_center_above_floor_q == lower_q[2]), None)
+    if lower_index is None:
+        raise RuntimeError(f"{passage['stable_id']}: lower mouth control is unbound")
+    return controls[:lower_index + 1]
+
+
+def maximum_grade(profile):
+    grades = []
+    for (p0, z0), (p1, z1) in zip(profile, profile[1:]):
+        horizontal = math.hypot(p1[0] - p0[0], p1[1] - p0[1])
+        if horizontal > 1.0e-8:
+            grades.append(math.degrees(math.atan2(abs(z1 - z0), horizontal)))
+    return max(grades, default=0.0)
+
+
+def main() -> int:
+    layout = json.loads(LAYOUT.read_text(encoding="utf-8"))
+    rows = []
+    for passage in layout["passages"]:
+        route = route_m(layout, passage)
+        segment_lengths = [math.dist(a[0], b[0]) for a, b in zip(route, route[1:])]
+        prefix = [0.0]
+        for length in segment_lengths:
+            prefix.append(prefix[-1] + length)
+        interior_radii = [radius for distance, (_, radius) in zip(prefix, route)
+                          if distance >= 10.0 and prefix[-1] - distance >= 10.0]
+        minimum_body_radius = min(interior_radii or [radius for _, radius in route])
+        body_depth = max(0.5, minimum_body_radius - 0.23)
+        scale = layout["core_voxel_size_um"] / 256.0 / 1_000_000.0
+        upper_target = max(0.0, route[0][0][2]
+                           - passage["upper_mouth_q"][2] * scale + 1.13)
+        lower_target = max(0.0, route[-1][0][2]
+                           - passage["lower_mouth_q"][2] * scale + 1.13)
+        floor_profile = []
+        raw_fixed_depth_profile = []
+        maximum_centerline_shift = 0.0
+        for segment_index, (a, b, length) in enumerate(
+                zip(route, route[1:], segment_lengths)):
+            steps = max(1, math.ceil(length / 0.05))
+            for step in range(steps + 1):
+                if floor_profile and step == 0:
+                    continue
+                t = step / steps
+                position = tuple(a[0][axis] + (b[0][axis] - a[0][axis]) * t
+                                 for axis in range(3))
+                radius = a[1] + (b[1] - a[1]) * t
+                along = prefix[segment_index] + length * t
+                mouth_distance = max(0.0, min(along, prefix[-1] - along))
+                u = max(0.0, min(1.0, mouth_distance / MOUTH_RAMP_METRES))
+                ramp = u * u * (3.0 - 2.0 * u)
+                local_radius_depth = max(0.0, radius - 0.23)
+                mouth_target = upper_target if along <= prefix[-1] * 0.5 else lower_target
+                mouth_depth = max(0.0, min(radius, mouth_target))
+                route_depth = mouth_depth + (local_radius_depth - mouth_depth) * ramp
+                tube_depth = min(body_depth, local_radius_depth)
+                effective_center = position[2] - route_depth + tube_depth
+                floor_profile.append((position, effective_center - tube_depth))
+                raw_fixed_depth_profile.append((position, position[2] - tube_depth))
+                maximum_centerline_shift = max(maximum_centerline_shift,
+                    abs(effective_center - position[2]))
+        rows.append({
+            "passage": passage["stable_id"],
+            "path_m": prefix[-1],
+            "body_floor_depth_m": body_depth,
+            "mouth_blend_m": MOUTH_RAMP_METRES,
+            "preserved_route_floor_max_grade_deg": maximum_grade(floor_profile),
+            "unshifted_fixed_depth_floor_max_grade_deg": maximum_grade(
+                raw_fixed_depth_profile),
+            "maximum_derived_centerline_shift_m": maximum_centerline_shift,
+        })
+    OUT.write_text(json.dumps({
+        "method": "0.05 m route samples; fixed body depth from the minimum interior radius; "
+                  "cubic smoothstep mouth blend",
+        "passages": rows,
+    }, indent=2) + "\n", encoding="utf-8")
+    for row in rows:
+        print(f"[L109FloorRamp] {row['passage']} "
+              f"depth_m={row['body_floor_depth_m']:.3f} "
+              f"preserved_grade_deg={row['preserved_route_floor_max_grade_deg']:.3f} "
+              f"unshifted_fixed_depth_grade_deg={row['unshifted_fixed_depth_floor_max_grade_deg']:.3f} "
+              f"max_center_shift_m={row['maximum_derived_centerline_shift_m']:.3f}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

Standalone/Research/WorldAssemblerM3/Linux109/REPORT.md

@@ -0,0 +1,70 @@
+# Linux 109 — tunnel floor and shell
+
+Baseline: `assembler/world` HEAD `95cf8de10723f2dffe7b3dfc116ecf350c7114ae`. Candidate source stamp: `9457b9bed7d57a4a387828961f9dd53632ec602eb5ea8f247b087be6ca5cc312` (evaluator v217).
+
+## Shape
+
+The per-tile swept tube now uses a horizontal flat floor, a 0.50 m smooth floor-to-wall corner, and a 0.75 m shell. Body floor depth is fixed per passage at the minimum interior radius minus 0.23 m, capped at the local radius. Five-metre cubic blends ramp both mouths into their ground; the lower mouth follows the matching authored native-cave floor where one exists. The route centreline shifts through radius tapers to keep that floor walkable. A small wall relief remains for naturalness.
+
+The passage interiors are unioned first and always win over every passage shell, including at coiled crossings. The shell writes solid only where the host field is air; rock remains unchanged. There is no rock classification. Open-air shell and interior samples pass in the owner audit.
+
+## Removed and retained
+
+Removed route support restoration, deferred support after later bores, support columns, mouth-support restoration, and their floor-specific capsule checks and profiling records. Replaced the old open-air/column audit gates with the shell audit and removed the no-long-bridge and rock-column gates, which measured the deleted support mechanism. Walkability, capsule, headroom, grade, mouth, naturalness, and ripple gates remain. Floor-ripple probes fully masked by another passage's interior air are skipped only when the overlap is explicit; capsule walkability still has to pass.
+
+Kept the native Lattice field's explicitly authored below-floor substrate because its spine needs it for walkability; this is part of that native field, not route support restoration. The legacy landmark demo retains its compatibility floor-cap path, which is separate from the resolved route sweep. Native Caves and Lattice fast checks pass.
+
+## Performance
+
+Five interleaved L80 pairs per cell, after one warmup for each binary; seconds are median `LOD0–2 near-ready / whole-view-ready`. Baseline binary is from `assembler/world` HEAD. Tile counts matched within each cell.
+
+| Strate | W8 near | W8 whole | W12 near | W12 whole |
+|---|---:|---:|---:|---:|
+| SW | 0.680 → 0.737 (+8.3%) | 1.686 → 1.674 (−0.7%) | 0.700 → 0.761 (+8.8%) | 1.643 → 1.644 (+0.1%) |
+| FI | 1.455 → 0.911 (−37.4%) | 2.419 → 1.772 (−26.8%) | 1.394 → 0.863 (−38.0%) | 2.288 → 1.664 (−27.3%) |
+| TN | 1.423 → 0.956 (−32.8%) | 2.310 → 1.778 (−23.0%) | 1.294 → 0.895 (−30.9%) | 2.132 → 1.661 (−22.1%) |
+| VS | 1.008 → 0.789 (−21.8%) | 1.917 → 1.639 (−14.5%) | 0.854 → 0.738 (−13.5%) | 1.716 → 1.541 (−10.2%) |
+
+There is no broad regression. SW near-ready is about 0.06 s slower; its whole-view time is flat. **Measured:** SW LOD0–2 skirt triangles rise from 133,894 to 196,602, and summed skirt work rises from 455 to 743 ms at W8 and 650 to 1,044 ms at W12. **Inference:** the added open-air shell surface and skirts explain the small SW near-ready cost. FI, TN, and VS improve at both readiness marks.
+
+## L77 phase census
+
+Two samples per case. Entries are `share / estimated worker-s`; estimates use weighted L77 samples times `WorldAssemblerLODWorker` CPU, so they are not direct phase timers.
+
+| Case | Base passage | Base support | Candidate passage | Candidate support |
+|---|---:|---:|---:|---:|
+| FI near | 44.62% / 4.108 | 14.84% / 1.366 | 22.90% / 1.133 | 0% / 0 |
+| FI whole | 29.31% / 4.874 | 8.08% / 1.344 | 10.59% / 1.206 | 0% / 0 |
+| TN near | 38.57% / 3.318 | 14.74% / 1.267 | 19.43% / 1.019 | 0% / 0 |
+| TN whole | 24.48% / 3.824 | 7.59% / 1.186 | 9.25% / 1.074 | 0% / 0 |
+
+The sampled support share disappears. Baseline estimates were 1.19–1.37 worker-s across these cases, below the 1.6–1.9 s expectation.
+
+## Audits and determinism
+
+The full-view all-LOD audit passes: 21,425/21,425 canonical passage centreline samples are open. All three interfaces pass capsule and support checks with zero blocked or unsupported samples, zero grade failures, and both mouths bound (2/2). Capsule samples: 266,175 (0–1), 15,120 (1–2), 64,764 (2–3). The shell audit sampled 89,365 floor points, 8,423 wall points, and 10,527 interior points: zero floor, wall, or interior violations; 22,755 open-air shell samples. Naturalness, mouth profile, and ripple summaries pass 3/3, with negative controls detected. Derived floor grade peaks at 14.49°; fast native Caves headroom is 0.30 m and native Lattice headroom is 2.20 m.
+
+Fast tier: PASS, 259 PASS checks plus 5 diagnostics across 80 runs. The Clang/GCC hash matrix matches at W1/W8/W12 for the owner four-strata layout and both two-strata canonical recipes. The v217 canonical pins were written for Surface World style 1 and Floating Islands style 2.
+
+Because Core sources are Unreal-compiled, the Linux VFHostEditor build passed with `-NoUBA -MaxParallelActions=4` (27 actions, 67.4 s). Three FI and three SW `after` arrival traces also passed, each settling its full requested LOD0–2 set. Unreal processes had no `prlimit`.
+
+## Change metrics
+
+LOD0–2 camera mesh windows; “moved” means a shared global lattice crossing displaced by more than 0.01 cm. Coverage is the paired edge-classifier coverage.
+
+| Strate | Added | Removed | Moved | Coverage |
+|---|---:|---:|---:|---:|
+| SW | 78,201 | 14,530 | 2,758 | 99.75% |
+| FI | 37,810 | 42,398 | 78,215 | 98.31% |
+| TN | 44,694 | 48,925 | 48,816 | 98.29% |
+| VS | 7,055 | 7,096 | 8,790 | 97.38% |
+
+## Captures
+
+84 PNGs: before and after, each with both mouths and 25/50/75% stations for all three passages. Each of the nine open-air sections has three views in each set. The [capture index](captures/INDEX.md) lists every image and flags every **OPEN AIR** view. All 84 source meshes and rendered views contain geometry; minimum visible triangles are 55,420 before and 55,484 after. Six before and seven after images met the pixel floor but were flagged low contrast for review.
+
+Open-air sections (route distance in metres): `0–1`: 7.226–10.217, 13.456–17.443, 23.174–24.918, 1036.913–1489.321; `1–2`: 101.820–104.499; `2–3`: 7.677–34.747, 368.909–376.660, 393.828–396.510, 444.932–447.654.
+
+## Failed runs
+
+Early fast runs found a 0.79 m lower-mouth mismatch at native Caves and loss of the native Lattice route substrate; the mouth now matches the authored cave edge floor, and the explicitly authored Lattice floor field is retained. A sentinel check also initially treated the “no native floor” value as support; this was fixed. The first Unreal build stopped before compilation because UnrealBuildTool could not rotate its log under read-only `~/.config`; redirecting `XDG_CONFIG_HOME` to `/tmp` fixed it. Final audits, timings, hash runs, crossings, captures, build, and six traces passed. No run exceeded 30 minutes.

Standalone/Research/WorldAssemblerM3/Linux109/RunL109Captures.py

@@ -0,0 +1,511 @@
+#!/usr/bin/env python3
+"""Capture before/after passage mouths, route stations, and exposed shell sections."""
+from __future__ import annotations
+
+import json
+import math
+import re
+from array import array
+import struct
+import subprocess
+import time
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[4]
+RESEARCH = Path(__file__).resolve().parent
+CAPTURES = RESEARCH / "captures"
+SAVED = ROOT / "Saved/Linux/l109/captures"
+DATA = ROOT / "Standalone/data/owner_world.json"
+BINARIES = {
+    "before": ROOT / "Saved/Linux/l109/bin-baseline/VoxelForgeCoreHeadless-l109-baseline",
+    "after": ROOT / "Saved/Linux/l109/bin-final/VoxelForgeCoreHeadless-l109-clang",
+}
+LAYOUTS = {
+    "before": ROOT / "Saved/Linux/l109/audit-layout-baseline.json",
+    "after": ROOT / "Saved/Linux/l109/audit-layout-final-stamp.json",
+}
+AUDIT_LOG = ROOT / "Saved/Linux/l109/audit-owner4-final-stamp.log"
+RENDERERS = {
+    "before": ROOT / "Saved/Linux/l109/bin-baseline/VoxelForgeOffscreenCapture-l109-baseline",
+    "after": ROOT / "Saved/Linux/l109/bin-final/VoxelForgeOffscreenCapture-l109-clang",
+}
+SECTION = re.compile(
+    r"^\[PassageOpenAirSection\] id=(\S+) section=(\d+) start_m=([\d.]+) "
+    r"end_m=([\d.]+) samples=(\d+) midpoint_xyz_m=\(([^)]*)\) "
+    r"midpoint_radius_m=([\d.]+)", re.M)
+DRAWN = re.compile(r"drawn_triangles=(\d+)")
+QUALITY_METRICS = re.compile(r"stddev_luma=([\d.]+).*coverage_luma_gt12=([\d.]+)")
+
+
+def vector(text: str) -> tuple[float, float, float]:
+    return tuple(float(value) for value in text.split(","))
+
+
+def add(a, b):
+    return tuple(x + y for x, y in zip(a, b))
+
+
+def sub(a, b):
+    return tuple(x - y for x, y in zip(a, b))
+
+
+def mul(a, scale):
+    return tuple(x * scale for x in a)
+
+
+def norm(a):
+    length = math.sqrt(sum(x * x for x in a))
+    if length <= 1.0e-9:
+        return (0.0, 1.0, 0.0)
+    return tuple(value / length for value in a)
+
+
+def angles(camera, target):
+    delta = sub(target, camera)
+    horizontal = math.hypot(delta[0], delta[1])
+    return (math.degrees(math.atan2(delta[0], delta[1])),
+            math.degrees(math.atan2(delta[2], horizontal)))
+
+
+def fmt(values) -> str:
+    return ",".join(f"{value:.6f}" for value in values)
+
+
+def mesh_triangle_count(mesh: Path) -> int:
+    """Read only the VFMSH headers and count indexed triangles."""
+    with mesh.open("rb") as stream:
+        magic = stream.read(8)
+        materials = magic == b"VFMSH002"
+        if magic not in (b"VFMSH001", b"VFMSH002"):
+            raise RuntimeError(f"unsupported mesh scene header: {mesh}")
+        version, _voxel_um, tile_count = struct.unpack("<3I", stream.read(12))
+        if version != (2 if materials else 1):
+            raise RuntimeError(f"mesh scene version mismatch: {mesh}")
+        stream.seek(40, 1)
+        triangles = 0
+        for _ in range(tile_count):
+            stream.seek(16, 1)
+            surface_count, _skirts, vertex_count, index_count, mask_count = struct.unpack(
+                "<5I", stream.read(20))
+            triangles += index_count // 3
+            stream.seek(vertex_count * 24 + (10 * vertex_count if materials else 0)
+                        + index_count * 4 + mask_count, 1)
+        if stream.read(1):
+            raise RuntimeError(f"trailing data in mesh scene: {mesh}")
+    return triangles
+
+
+def dot(a, b):
+    return sum(x * y for x, y in zip(a, b))
+
+
+def nearest_surface_anchor(mesh: Path, point, margin_m=200.0) -> dict:
+    """Return a nearby VFMSH surface vertex and its normal."""
+    best_distance_sq = float("inf")
+    best_position = best_normal = None
+    with mesh.open("rb") as stream:
+        magic = stream.read(8)
+        materials = magic == b"VFMSH002"
+        if magic not in (b"VFMSH001", b"VFMSH002"):
+            raise RuntimeError(f"unsupported mesh scene header: {mesh}")
+        version, voxel_um, tile_count = struct.unpack("<3I", stream.read(12))
+        if version != (2 if materials else 1):
+            raise RuntimeError(f"mesh scene version mismatch: {mesh}")
+        stream.seek(40, 1)
+        for _ in range(tile_count):
+            key = struct.unpack("<4i", stream.read(16))
+            surface_count, _skirts, vertex_count, index_count, mask_count = struct.unpack(
+                "<5I", stream.read(20))
+            extent = (32 << key[3]) * voxel_um / 1_000_000.0
+            origin = (key[0] * extent, key[1] * extent, key[2] * extent)
+            if any(point[axis] < origin[axis] - margin_m or
+                   point[axis] > origin[axis] + extent + margin_m for axis in range(3)):
+                stream.seek(vertex_count * 24 + (10 * vertex_count if materials else 0)
+                            + index_count * 4 + mask_count, 1)
+                continue
+            positions = array("f")
+            positions.frombytes(stream.read(vertex_count * 12))
+            normals = array("f")
+            normals.frombytes(stream.read(vertex_count * 12))
+            if materials:
+                stream.seek(vertex_count * 10, 1)
+            stream.seek(index_count * 4 + mask_count, 1)
+            if struct.pack("=I", 1) != struct.pack("<I", 1):
+                positions.byteswap()
+                normals.byteswap()
+            for index in range(min(surface_count, vertex_count)):
+                offset = index * 3
+                position = tuple(float(positions[offset + axis]) for axis in range(3))
+                distance_sq = dot(sub(position, point), sub(position, point))
+                if distance_sq >= best_distance_sq:
+                    continue
+                raw_normal = tuple(float(normals[offset + axis]) for axis in range(3))
+                if dot(raw_normal, raw_normal) < 1.0e-10:
+                    continue
+                best_distance_sq = distance_sq
+                best_position = position
+                best_normal = norm(raw_normal)
+    if best_position is None:
+        raise RuntimeError(f"no nearby surface vertices in mesh scene: {mesh}")
+    return {"position_m": best_position, "normal": best_normal,
+            "distance_m": math.sqrt(best_distance_sq)}
+
+
+def sanitize(value: str) -> str:
+    return re.sub(r"[^A-Za-z0-9_-]+", "-", value).strip("-").replace("/", "-")
+
+
+def route_for(layout: dict, passage: dict) -> list[tuple[tuple[float, float, float], float]]:
+    scale = layout["core_voxel_size_um"] / 256.0 / 1_000_000.0
+    controls = [(tuple(value * scale for value in row["position_q"]),
+                 row["radius_q"] * scale) for row in passage["control_points"]]
+    lower_q = passage["lower_mouth_q"]
+    player_center_above_floor_q = round(
+        1.13 * 256.0 * 1_000_000.0 / layout["core_voxel_size_um"])
+    lower_index = next((index for index, row in enumerate(passage["control_points"])
+                        if row["position_q"][0] == lower_q[0]
+                        and row["position_q"][1] == lower_q[1]
+                        and row["position_q"][2] - row["radius_q"]
+                            + player_center_above_floor_q == lower_q[2]), None)
+    if lower_index is None:
+        raise RuntimeError(f"{passage['stable_id']}: lower mouth control is unbound")
+    result = controls[:lower_index + 1]
+    if len(result) < 2:
+        raise RuntimeError(f"{passage['stable_id']}: incomplete route controls")
+    return result
+
+
+def route_lengths(route):
+    lengths = [math.dist(a[0], b[0]) for a, b in zip(route, route[1:])]
+    prefix = [0.0]
+    for length in lengths:
+        prefix.append(prefix[-1] + length)
+    return lengths, prefix
+
+
+def at_distance(route, lengths, prefix, distance):
+    distance = max(0.0, min(distance, prefix[-1]))
+    for index, length in enumerate(lengths):
+        if distance <= prefix[index + 1] or index + 1 == len(lengths):
+            t = 0.0 if length <= 1.0e-9 else (distance - prefix[index]) / length
+            a, b = route[index], route[index + 1]
+            return (tuple(a[0][axis] + (b[0][axis] - a[0][axis]) * t
+                          for axis in range(3)),
+                    a[1] + (b[1] - a[1]) * t)
+    return route[-1]
+
+
+def tangent_at(route, lengths, prefix, distance):
+    before = at_distance(route, lengths, prefix, max(0.0, distance - 0.25))[0]
+    after = at_distance(route, lengths, prefix, min(prefix[-1], distance + 0.25))[0]
+    return norm(sub(after, before))
+
+
+def derived_route(route, passage, layout):
+    """Apply L109's fixed-depth floor offset to raw layout controls."""
+    lengths, prefix = route_lengths(route)
+    total = prefix[-1]
+    body_radii = [radius for index, (_, radius) in enumerate(route)
+                  if prefix[index] >= 10.0 and total - prefix[index] >= 10.0]
+    body_depth = max(0.5, min(body_radii or [radius for _, radius in route]) - 0.23)
+    scale = layout["core_voxel_size_um"] / 256.0 / 1_000_000.0
+    upper_z = passage["upper_mouth_q"][2] * scale
+    lower_z = passage["lower_mouth_q"][2] * scale
+    upper_target = max(0.0, route[0][0][2] - upper_z + 1.13)
+    lower_target = max(0.0, route[-1][0][2] - lower_z + 1.13)
+    transformed = []
+    for index, (point, radius) in enumerate(route):
+        local_radius_depth = max(0.0, radius - 0.23)
+        mouth_distance = max(0.0, min(prefix[index], total - prefix[index]))
+        u = max(0.0, min(1.0, mouth_distance / 5.0))
+        ramp = u * u * (3.0 - 2.0 * u)
+        target = upper_target if prefix[index] <= total * 0.5 else lower_target
+        mouth_depth = max(0.0, min(radius, target))
+        route_depth = mouth_depth + (local_radius_depth - mouth_depth) * ramp
+        tube_depth = min(body_depth, local_radius_depth)
+        transformed.append(((point[0], point[1], point[2] - route_depth + tube_depth),
+                            radius))
+    return transformed, body_depth
+
+
+def make_views(layout: dict, audit_sections: list[dict], *, use_derived_shape: bool,
+               include_open_air: bool = True) -> list[dict]:
+    views = []
+    passages = sorted(layout["passages"], key=lambda item: item["stable_id"])
+    for passage in passages:
+        raw_route = route_for(layout, passage)
+        if use_derived_shape:
+            route, body_depth = derived_route(raw_route, passage, layout)
+        else:
+            route = raw_route
+            body_depth = max(point[1] for point in route)
+        def floor_depth(radius: float) -> float:
+            if not use_derived_shape:
+                return radius
+            return min(body_depth, max(0.0, radius - 0.23))
+        lengths, prefix = route_lengths(route)
+        total = prefix[-1]
+        passage_id = passage["stable_id"]
+        safe_id = sanitize(passage_id)
+
+        for end_name, distance, outward_sign in (
+                ("upper-mouth", 0.0, -1.0), ("lower-mouth", total, 1.0)):
+            center, radius = at_distance(route, lengths, prefix, distance)
+            tangent = tangent_at(route, lengths, prefix, distance)
+            outward = mul(tangent, outward_sign)
+            floor_z = center[2] - floor_depth(radius)
+            eye = (center[0] + outward[0] * 4.5,
+                   center[1] + outward[1] * 4.5,
+                   floor_z + 1.55)
+            if end_name == "lower-mouth":
+                eye = add(eye, add(mul(outward, 5.0), (0.0, 0.0, 2.0)))
+            target = (center[0] - outward[0] * 2.0,
+                      center[1] - outward[1] * 2.0,
+                      floor_z + 1.45)
+            yaw, pitch = angles(eye, target)
+            views.append({"name": f"{safe_id}-{end_name}", "passage": passage_id,
+                          "kind": "mouth", "station": end_name,
+                          "camera": eye, "target": target, "yaw": yaw, "pitch": pitch,
+                          "focus": center, "center_z": center[2]})
+
+        for fraction in (0.25, 0.50, 0.75):
+            distance = total * fraction
+            center, radius = at_distance(route, lengths, prefix, distance)
+            tangent = tangent_at(route, lengths, prefix, distance)
+            depth = floor_depth(radius)
+            floor_z = center[2] - depth
+            horizontal = norm((-tangent[1], tangent[0], 0.0))
+            eye = (center[0] + horizontal[0] * 0.75,
+                   center[1] + horizontal[1] * 0.75, floor_z + 1.60)
+            target_center, target_radius = at_distance(route, lengths, prefix,
+                min(total, distance + 14.0))
+            target_depth = floor_depth(target_radius)
+            target = (target_center[0] + horizontal[0] * 2.35,
+                      target_center[1] + horizontal[1] * 2.35,
+                      target_center[2] - target_depth + 1.80)
+            yaw, pitch = angles(eye, target)
+            views.append({"name": f"{safe_id}-station-{int(fraction * 100)}",
+                          "passage": passage_id, "kind": "station",
+                          "station": f"{int(fraction * 100)}%", "camera": eye,
+                          "target": target, "yaw": yaw, "pitch": pitch,
+                          "focus": center, "center_z": center[2],
+                          "route_distance_m": distance,
+                          "route_length_m": total,
+                          "requested_route_distance_m": distance})
+
+        if not include_open_air:
+            continue
+        for section in (item for item in audit_sections if item["passage"] == passage_id):
+            distance = 0.5 * (section["start_m"] + section["end_m"])
+            center, radius = (section["midpoint"], section["radius_m"])
+            tangent = tangent_at(route, lengths, prefix, distance)
+            side = norm((-tangent[1], tangent[0], 0.0))
+            depth = floor_depth(radius)
+            section_id = section["section"]
+            interior_eye = (center[0], center[1], center[2] - depth + 1.25)
+            interior_target = add(interior_eye, mul(tangent, 7.0))
+            side_eye = add(center, add(mul(side, radius + 5.0), (0.0, 0.0, 1.2)))
+            oblique_eye = add(center, add(mul(side, -(radius + 3.5)), (0.0, 0.0, 4.0)))
+            targets = (add(center, (0.0, 0.0, 0.1)),
+                       add(center, mul(tangent, 2.0)), interior_target)
+            for view_index, (eye, target) in enumerate(zip(
+                    (side_eye, oblique_eye, interior_eye), targets), start=1):
+                yaw, pitch = angles(eye, target)
+                views.append({"name": f"{safe_id}-openair-{section_id}-view-{view_index}",
+                              "passage": passage_id, "kind": "open-air",
+                              "station": f"section-{section_id}/view-{view_index}",
+                              "camera": eye, "target": target, "yaw": yaw,
+                              "pitch": pitch, "focus": center,
+                              "center_z": center[2], "open_air_section": section_id,
+                              "route_distance_m": distance,
+                              "requested_route_distance_m": distance,
+                              "route_length_m": total})
+    return views
+
+
+def run_process(name: str, command: list[str], log: Path, *, allow_failure=False) -> str:
+    wrapped = ["taskset", "-c", "0-11", "prlimit", "--as=6442450944", "--",
+               "timeout", "1740s", *map(str, command)]
+    started = time.monotonic()
+    with log.open("w", encoding="utf-8") as stream:
+        result = subprocess.run(wrapped, cwd=ROOT, stdout=stream,
+                                stderr=subprocess.STDOUT, check=False, timeout=1740)
+    if result.returncode and not allow_failure:
+        raise RuntimeError(f"{name}: exit={result.returncode} log={log}")
+    print(f"[L109CaptureProcess] name={name} elapsed_s={time.monotonic()-started:.2f} "
+          f"exit={result.returncode} log={log.relative_to(ROOT)}", flush=True)
+    return log.read_text(encoding="utf-8", errors="replace")
+
+
+def capture(variant: str, view: dict, layout: dict) -> dict:
+    variant_saved = SAVED / variant
+    variant_saved.mkdir(parents=True, exist_ok=True)
+    variant_capture = CAPTURES / variant
+    variant_capture.mkdir(parents=True, exist_ok=True)
+    scene = variant_saved / f"{view['name']}.vfmesh"
+    image = variant_capture / f"{view['name']}.png"
+    candidates = [(view["camera"], view["target"], view["focus"],
+                   view.get("route_distance_m"), 0.0, "requested")]
+    if variant == "before" and view["kind"] != "mouth":
+        passage = next(item for item in layout["passages"]
+                       if item["stable_id"] == view["passage"])
+        route = route_for(layout, passage)
+        lengths, prefix = route_lengths(route)
+        total = prefix[-1]
+        eye_offset = sub(view["camera"], view["focus"])
+        target_offset = sub(view["target"], view["focus"])
+        requested = view["route_distance_m"]
+        for delta in (5, -5, 10, -10, 20, -20, 35, -35, 50, -50, 75, -75,
+                      100, -100, 150, -150, 200, -200):
+            distance = max(0.0, min(total, requested + delta))
+            if abs(distance - requested) < 0.01:
+                continue
+            center, _radius = at_distance(route, lengths, prefix, distance)
+            candidates.append((add(center, eye_offset), add(center, target_offset), center,
+                               distance, distance - requested, "route-offset"))
+
+    attempt_errors = []
+    for attempt_index, (probe_camera, probe_target, probe_focus, route_distance,
+                        route_adjustment, attempt_kind) in enumerate(candidates):
+        suffix = "" if attempt_index == 0 else f"-attempt-{attempt_index:02d}"
+        headless_log = variant_saved / f"{view['name']}{suffix}-mesh.log"
+        render_log = variant_saved / f"{view['name']}{suffix}-render.log"
+        quality_log = variant_saved / f"{view['name']}{suffix}-quality.log"
+        yaw, pitch = angles(probe_camera, probe_target)
+        camera = (*probe_camera, yaw, pitch)
+        export_command = [
+            str(BINARIES[variant]), str(DATA), "--world-assembler-prototype",
+            "--stream-benchmark", "--stream-pipeline", "--workers=12",
+            "--clip-radius=3", "--max-clip-level=0", "--max-level=0", "--force-generate",
+            "--recipe=surface-world", f"--camera={fmt(camera)}",
+            "--render-distance-chunks=1", f"--mesh-export={scene}",
+        ]
+        export_output = run_process(f"{variant}-{view['name']}{suffix}-mesh", export_command,
+                                    headless_log)
+        if "status=PASS" not in export_output or not scene.is_file():
+            raise RuntimeError(f"{variant}-{view['name']}: missing successful scene export")
+        source_triangles = mesh_triangle_count(scene)
+        if source_triangles <= 0:
+            attempt_errors.append(f"{attempt_kind}@{route_distance}: empty mesh")
+            if variant == "after" or view["kind"] == "mouth":
+                break
+            continue
+
+        actual_camera, actual_target = probe_camera, probe_target
+        actual_focus = probe_focus
+        surface_distance = None
+        if attempt_kind == "route-offset":
+            anchor = nearest_surface_anchor(scene, probe_focus)
+            surface_distance = anchor["distance_m"]
+            actual_camera = add(anchor["position_m"], mul(anchor["normal"], 12.0))
+            actual_target = anchor["position_m"]
+            actual_focus = anchor["position_m"]
+        actual_yaw, actual_pitch = angles(actual_camera, actual_target)
+        render_command = [
+            str(RENDERERS[variant]), str(scene), str(image), "--size=1920x1080",
+            f"--camera-m={fmt(actual_camera)}", f"--yaw-deg={actual_yaw:.6f}",
+            f"--pitch-deg={actual_pitch:.6f}", "--shade=readable", "--headlamp=on",
+            "--height-tint=on", f"--height-center-m={actual_focus[2]:.6f}",
+            "--height-range-m=32",
+        ]
+        render_output = run_process(f"{variant}-{view['name']}{suffix}-render", render_command,
+                                    render_log)
+        drawn = DRAWN.search(render_output)
+        if drawn is None or int(drawn.group(1)) <= 0 or not image.is_file():
+            attempt_errors.append(f"{attempt_kind}@{route_distance}: no drawn geometry")
+            if variant == "after" or view["kind"] == "mouth":
+                break
+            continue
+        quality = run_process(f"{variant}-{view['name']}{suffix}-quality", [
+            "python3", "-B", str(ROOT / "Standalone/Tools/verify/capture_quality.py"),
+            str(image)], quality_log, allow_failure=True)
+        quality_status = "pass"
+        if "status=PASS" not in quality:
+            metrics = QUALITY_METRICS.search(quality)
+            coverage = float(metrics.group(2)) if metrics else 0.0
+            if coverage >= 0.001:
+                quality_status = "low_contrast_geometry_visible"
+            else:
+                attempt_errors.append(f"{attempt_kind}@{route_distance}: no visible pixel coverage")
+                if variant == "after" or view["kind"] == "mouth":
+                    break
+                continue
+
+        row = {**view, "variant": variant, "capture": str(image.relative_to(RESEARCH)),
+               "scene": str(scene.relative_to(ROOT)), "drawn_triangles": int(drawn.group(1)),
+               "source_triangles": source_triangles, "verified": True,
+               "pixel_quality_status": quality_status,
+               "route_distance_actual_m": route_distance,
+               "route_station_adjustment_m": route_adjustment,
+               "fallback_kind": attempt_kind if attempt_kind != "requested" else None,
+               "fallback_surface_distance_m": surface_distance,
+               "camera_actual_m": actual_camera, "target_actual_m": actual_target,
+               "logs": [str(headless_log.relative_to(ROOT)),
+                        str(render_log.relative_to(ROOT)),
+                        str(quality_log.relative_to(ROOT))]}
+        print("[L109Capture] " + json.dumps(row, sort_keys=True), flush=True)
+        return row
+    raise RuntimeError(f"{variant}-{view['name']}: no verified image; "
+                       + "; ".join(attempt_errors))
+
+
+def main() -> int:
+    if any(not path.is_file() for path in (*BINARIES.values(), *RENDERERS.values(),
+                                            *LAYOUTS.values(), AUDIT_LOG)):
+        raise SystemExit("missing L109 capture binary, layout, or audit log")
+    sections = []
+    for passage, section, start, end, samples, midpoint, radius in SECTION.findall(
+            AUDIT_LOG.read_text(encoding="utf-8", errors="replace")):
+        sections.append({"passage": passage, "section": int(section),
+                         "start_m": float(start), "end_m": float(end),
+                         "samples": int(samples), "midpoint": vector(midpoint),
+                         "radius_m": float(radius)})
+    if not sections:
+        raise SystemExit("no open-air sections found in the final passage audit")
+    all_rows = []
+    for variant in ("before", "after"):
+        layout = json.loads(LAYOUTS[variant].read_text(encoding="utf-8"))
+        include_open_air = True
+        views = make_views(layout, sections, use_derived_shape=variant == "after",
+                           include_open_air=include_open_air)
+        expected = 5 * len(layout["passages"])
+        if include_open_air:
+            expected += 3 * len(sections)
+        if len(views) != expected:
+            raise RuntimeError(f"unexpected camera count: {len(views)} != {expected}")
+        print(f"[L109CapturePlan] variant={variant} passage_views={5*len(layout['passages'])} "
+              f"open_air_sections={len(sections) if include_open_air else 0} "
+              f"open_air_views={3*len(sections) if include_open_air else 0} "
+              f"total={len(views)}", flush=True)
+        for view in views:
+            all_rows.append(capture(variant, view, layout))
+    CAPTURES.mkdir(parents=True, exist_ok=True)
+    (CAPTURES / "capture-index.json").write_text(
+        json.dumps({"open_air_sections": sections, "captures": all_rows}, indent=2) + "\n",
+        encoding="utf-8")
+    lines = ["# L109 passage captures", "",
+             "Each PNG was rendered from its own exported mesh. The renderer reported visible "
+             "triangles, and each decoded image met the pixel coverage floor. "
+             "`low_contrast_geometry_visible` marks images that met this floor below the "
+             "preferred contrast target.", ""]
+    for variant in ("before", "after"):
+        lines.extend([f"## {variant.title()}", ""])
+        for row in all_rows:
+            if row["variant"] != variant:
+                continue
+            flag = " **OPEN AIR**" if row["kind"] == "open-air" else ""
+            quality_note = (f"; {row['pixel_quality_status']}"
+                            if row["pixel_quality_status"] != "pass" else "")
+            lines.append(f"- [{row['station']}]({variant}/{Path(row['capture']).name}) — "
+                         f"{row['passage']}{flag}; {row['drawn_triangles']} visible triangles"
+                         f"{quality_note}")
+        lines.append("")
+    (CAPTURES / "INDEX.md").write_text("\n".join(lines), encoding="utf-8")
+    failures = [row for row in all_rows if not row["verified"]]
+    print(f"[L109CaptureSummary] captures={len(all_rows)} verified={len(all_rows)-len(failures)} "
+          f"failed={len(failures)} open_air_sections={len(sections)}", flush=True)
+    return 1 if failures else 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

Standalone/Research/WorldAssemblerM3/Linux109/RunL109CrossingDelta.py

@@ -0,0 +1,140 @@
+#!/usr/bin/env python3
+"""Compare LOD0-2 owner mesh edge crossings before and after L109."""
+from __future__ import annotations
+
+import json
+import math
+import statistics
+import struct
+import subprocess
+import time
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[4]
+OUT = ROOT / "Saved/Linux/l109/crossings"
+DATA = ROOT / "Standalone/data/owner_world.json"
+BINARIES = {
+    "before": ROOT / "Saved/Linux/l109/bin-baseline/VoxelForgeCoreHeadless-l109-baseline",
+    "after": ROOT / "Saved/Linux/l109/bin-final/VoxelForgeCoreHeadless-l109-clang",
+}
+CAMERAS = {
+    "SW": (8.0, 16.0, -372.2626953125),
+    "FI": (25.8369140625, 3.19921875, -737.0),
+    "TN": (19.166015625, -14.2001953125, -753.3740234375),
+    "VS": (-20.0625, -5.068359375, -900.353515625),
+}
+
+
+def read_exact(stream, size):
+    data = stream.read(size)
+    if len(data) != size:
+        raise RuntimeError("truncated VFMSH export")
+    return data
+
+
+def crossing_map(path: Path) -> tuple[dict, dict]:
+    crossings = {}
+    meta = {"tiles": 0, "surface_vertices": 0, "classified": 0}
+    with path.open("rb") as stream:
+        magic = read_exact(stream, 8)
+        materials = magic == b"VFMSH002"
+        if magic not in (b"VFMSH001", b"VFMSH002"):
+            raise RuntimeError(f"invalid mesh export {path}")
+        version, voxel_um, tile_count = struct.unpack("<3I", read_exact(stream, 12))
+        if version != (2 if materials else 1):
+            raise RuntimeError("VFMSH version mismatch")
+        read_exact(stream, 40)
+        voxel_m = voxel_um / 1_000_000.0
+        for _ in range(tile_count):
+            key = struct.unpack("<4i", read_exact(stream, 16))
+            surface_count, _skirts, vertex_count, index_count, mask_count = struct.unpack(
+                "<5I", read_exact(stream, 20))
+            positions = struct.iter_unpack("<3f", read_exact(stream, vertex_count * 12))
+            values = list(positions)
+            read_exact(stream, vertex_count * 12)  # normals
+            if materials:
+                read_exact(stream, vertex_count * 10)
+            read_exact(stream, index_count * 4 + mask_count)
+            meta["tiles"] += 1
+            meta["surface_vertices"] += min(surface_count, vertex_count)
+            unit = voxel_m * (1 << key[3])
+            for position in values[:min(surface_count, vertex_count)]:
+                grid = tuple(value / unit for value in position)
+                fractions = tuple(abs(value - round(value)) for value in grid)
+                axis = max(range(3), key=fractions.__getitem__)
+                if fractions[axis] < 1.0e-4:
+                    continue
+                fixed = [index for index in range(3) if index != axis]
+                if any(fractions[index] > 0.01 for index in fixed):
+                    continue
+                edge = [0, 0, 0]
+                for index in fixed:
+                    edge[index] = int(round(grid[index]))
+                edge[axis] = math.floor(grid[axis])
+                identity = (key[3], axis, *edge)
+                crossings.setdefault(identity, tuple(float(v) for v in position))
+                meta["classified"] += 1
+        if stream.read(1):
+            raise RuntimeError("trailing data in mesh export")
+    meta["classified_pct"] = (100.0 * meta["classified"] / meta["surface_vertices"]
+                              if meta["surface_vertices"] else 0.0)
+    return crossings, meta
+
+
+def export(strate: str, variant: str) -> Path:
+    OUT.mkdir(parents=True, exist_ok=True)
+    mesh = OUT / f"{strate.lower()}-{variant}.vfmesh"
+    log = OUT / f"{strate.lower()}-{variant}.log"
+    camera = ",".join(f"{v:.12f}" for v in CAMERAS[strate]) + ",0,0"
+    command = [
+        "taskset", "-c", "0-11", "prlimit", "--as=6442450944", "--",
+        "timeout", "1740s", str(BINARIES[variant]), str(DATA),
+        "--world-assembler-prototype", "--stream-benchmark", "--stream-pipeline",
+        "--workers=8", "--clip-radius=3", "--max-clip-level=2", "--max-level=2",
+        "--force-generate", "--recipe=surface-world", f"--camera={camera}",
+        "--render-distance-chunks=0",
+        f"--mesh-export={mesh}",
+    ]
+    started = time.monotonic()
+    with log.open("w", encoding="utf-8") as stream:
+        result = subprocess.run(command, cwd=ROOT, stdout=stream,
+                                stderr=subprocess.STDOUT, timeout=1740, check=False)
+    output = log.read_text(encoding="utf-8", errors="replace")
+    if result.returncode or not mesh.is_file() or "status=PASS" not in output:
+        raise RuntimeError(f"mesh export failed: {strate}/{variant} exit={result.returncode} {log}")
+    return mesh
+
+
+def main() -> int:
+    summaries = []
+    for strate in CAMERAS:
+        paths = {variant: export(strate, variant) for variant in ("before", "after")}
+        maps = {variant: crossing_map(path) for variant, path in paths.items()}
+        old, old_meta = maps["before"]
+        new, new_meta = maps["after"]
+        old_keys, new_keys = set(old), set(new)
+        shared = old_keys & new_keys
+        moved_cm = [100.0 * math.dist(old[key], new[key]) for key in shared
+                    if math.dist(old[key], new[key]) > 0.0001]
+        row = {
+            "strate": strate, "scope": "camera LOD0-2 mesh window; global lattice edge IDs",
+            "added_crossings": len(new_keys - old_keys),
+            "removed_crossings": len(old_keys - new_keys),
+            "shared_crossings": len(shared), "moved_crossings": len(moved_cm),
+            "mean_moved_cm": statistics.mean(moved_cm) if moved_cm else 0.0,
+            "max_moved_cm": max(moved_cm, default=0.0),
+            "before": old_meta, "after": new_meta,
+            "classified_pair_coverage_pct": min(old_meta["classified_pct"],
+                                                  new_meta["classified_pct"]),
+            "exports": {key: str(value.relative_to(ROOT)) for key, value in paths.items()},
+        }
+        summaries.append(row)
+        print("[L109CrossingDelta] " + json.dumps(row, sort_keys=True), flush=True)
+    (OUT / "summary.json").write_text(
+        json.dumps({"measurement": "per-vertex marching-cubes edge classifier; compare shared global lattice edges; positions >0.01cm count as moved",
+                    "strata": summaries}, indent=2) + "\n", encoding="utf-8")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

Standalone/Research/WorldAssemblerM3/Linux109/RunL109HashMatrix.py

@@ -0,0 +1,103 @@
+#!/usr/bin/env python3
+"""Check Clang/GCC hash identity at W1/W8/W12 and repin v217 owner hashes."""
+from __future__ import annotations
+
+import json
+import re
+import subprocess
+import time
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[4]
+OUT = ROOT / "Saved/Linux/l109/hashes"
+DATASETS = {
+    "owner4": ROOT / "Standalone/data/owner_world.json",
+    "owner2": ROOT / "Standalone/data/regression/owner_world_2strate_l61.json",
+}
+BINARIES = {
+    "clang": ROOT / "Saved/Linux/l109/bin-final/VoxelForgeCoreHeadless-l109-clang",
+    "gcc": ROOT / "Saved/Linux/l109/bin-final/VoxelForgeCoreHeadless-l109-gcc",
+}
+HASH = re.compile(r"^\[Pipeline(?:FullDetailHash|WholeViewHash|LODHash|ExactQueryHash)\].*$", re.M)
+STAMP = re.compile(r"^\[VoxelForgeSourceStamp\] sha256=(\S+)$", re.M)
+
+
+def run(dataset: str, recipe: str, workers: int, compiler: str) -> dict:
+    OUT.mkdir(parents=True, exist_ok=True)
+    label = f"{dataset}-{recipe}-w{workers}-{compiler}"
+    log = OUT / f"{label}.log"
+    layout = OUT / f"{label}.layout.json"
+    command = [
+        "taskset", "-c", "0-11", "prlimit", "--as=6442450944", "--",
+        "timeout", "1740s", str(BINARIES[compiler]), str(DATASETS[dataset]),
+        f"--workers={workers}", "--world-assembler-prototype", "--stream-benchmark",
+        "--stream-pipeline", "--force-generate", "--clip-radius=3",
+        "--max-clip-level=8", "--max-level=8", f"--recipe={recipe}",
+        "--stream-hash", f"--resolved-layout-output={layout}",
+    ]
+    started = time.monotonic()
+    with log.open("w", encoding="utf-8") as stream:
+        result = subprocess.run(command, cwd=ROOT, stdout=stream,
+                                stderr=subprocess.STDOUT, check=False, timeout=1740)
+    output = log.read_text(encoding="utf-8", errors="replace")
+    hashes = HASH.findall(output)
+    if result.returncode or len(hashes) != 12:
+        raise SystemExit(f"hash run failed or incomplete: {label}; exit={result.returncode} "
+                         f"hashes={len(hashes)} log={log}")
+    row = {"dataset": dataset, "recipe": recipe, "workers": workers,
+           "compiler": compiler, "exit_code": result.returncode,
+           "hashes": hashes, "source_stamp": (STAMP.search(output).group(1)
+                                                if STAMP.search(output) else None),
+           "elapsed_s": round(time.monotonic() - started, 3),
+           "log": str(log.relative_to(ROOT))}
+    print("[L109HashRun] " + json.dumps(row, sort_keys=True), flush=True)
+    return row
+
+
+def main() -> int:
+    if any(not path.is_file() for path in (*DATASETS.values(), *BINARIES.values())):
+        raise SystemExit("missing L109 hash input")
+    rows = []
+    recipes = {"owner4": ("surface-world",),
+               "owner2": ("surface-world", "floating-islands")}
+    for dataset, dataset_recipes in recipes.items():
+        for recipe in dataset_recipes:
+            for workers in (1, 8, 12):
+                for compiler in ("clang", "gcc"):
+                    rows.append(run(dataset, recipe, workers, compiler))
+                    (OUT / "runs-partial.json").write_text(
+                        json.dumps(rows, indent=2) + "\n", encoding="utf-8")
+    groups = {}
+    for dataset, dataset_recipes in recipes.items():
+        for recipe in dataset_recipes:
+            key = f"{dataset}/{recipe}"
+            group = [row for row in rows if row["dataset"] == dataset
+                     and row["recipe"] == recipe]
+            reference = group[0]["hashes"]
+            groups[key] = {"all_worker_compiler_hashes_match": all(
+                row["hashes"] == reference for row in group),
+                "source_stamps": sorted({row["source_stamp"] for row in group}),
+                "runs": [{"workers": row["workers"], "compiler": row["compiler"],
+                          "log": row["log"]} for row in group]}
+            if not groups[key]["all_worker_compiler_hashes_match"]:
+                raise SystemExit(f"determinism mismatch: {key}")
+            if dataset == "owner2":
+                canonical = next(row for row in group if row["workers"] == 12
+                                 and row["compiler"] == "clang")
+                recipe_name = "floating-islands" if recipe == "floating-islands" else "surface-world"
+                style = 1 if recipe_name == "surface-world" else 2
+                baseline = (ROOT / "Standalone/Tools/verify/baselines/owner-hash/canonical"
+                            / f"owner-hash-{recipe_name}-style{style}-v217.baseline")
+                baseline.write_text("\n".join(canonical["hashes"]) + "\n",
+                                    encoding="utf-8")
+    summary = {"runs": rows, "groups": groups,
+               "owner4_all_four_strata": "owner4/surface-world",
+               "owner2_canonical_pins": ["owner2/surface-world",
+                                          "owner2/floating-islands"]}
+    (OUT / "summary.json").write_text(json.dumps(summary, indent=2) + "\n",
+                                      encoding="utf-8")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

Standalone/Research/WorldAssemblerM3/Linux109/RunL109PhaseCensus.py

@@ -0,0 +1,112 @@
+#!/usr/bin/env python3
+"""Estimate L77 passage/support worker shares before and after L109."""
+from __future__ import annotations
+
+import json
+import re
+import statistics
+import subprocess
+import time
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[4]
+OUT = ROOT / "Saved/Linux/l109/l77-phase-census"
+DATA = ROOT / "Standalone/data/owner_world.json"
+BINARIES = {
+    "base": ROOT / "Saved/Linux/l109/bin-baseline-profile/VoxelForgeCoreHeadless-l109-base-profile",
+    "candidate": ROOT / "Saved/Linux/l109/bin-final-profile/VoxelForgeCoreHeadless-l109-candidate-profile",
+}
+CAMERAS = {
+    "FI": (25.8369140625, 3.19921875, -737.0),
+    "TN": (19.166015625, -14.2001953125, -753.3740234375),
+}
+WORKER = re.compile(r"^\[WorldAssemblerLODWorker\] (.*)$", re.M)
+SAMPLE = re.compile(r"^\[L77Sample\] (.*)$", re.M)
+SAMPLER = re.compile(r"^\[L77Sampler\] action=stop (.*)$", re.M)
+FIELDS = re.compile(r"([a-zA-Z0-9_]+)=([^ ]+)")
+
+
+def run(stratum: str, variant: str, replicate: int) -> dict:
+    OUT.mkdir(parents=True, exist_ok=True)
+    label = f"{stratum.lower()}-{variant}-r{replicate}"
+    log = OUT / f"{label}.log"
+    camera = ",".join(f"{value:.12f}" for value in CAMERAS[stratum]) + ",0,0"
+    command = [
+        "taskset", "-c", "0-11", "prlimit", "--as=6442450944", "--",
+        "timeout", "1740s", str(BINARIES[variant]), str(DATA),
+        "--world-assembler-prototype", "--stream-benchmark", "--stream-pipeline",
+        "--workers=8", "--clip-radius=3", "--max-clip-level=8", "--max-level=8",
+        f"--camera={camera}", "--profile-lod-work", "--stream-scheduler-policy=3",
+        "--linux77-profile",
+    ]
+    started = time.monotonic()
+    with log.open("w", encoding="utf-8") as stream:
+        result = subprocess.run(command, cwd=ROOT, stdout=stream,
+                                stderr=subprocess.STDOUT, check=False, timeout=1740)
+    content = log.read_text(encoding="utf-8", errors="replace")
+    sampler_match = SAMPLER.search(content)
+    workers = [dict(FIELDS.findall(match.group(1))) for match in WORKER.finditer(content)]
+    samples = [dict(FIELDS.findall(match.group(1))) for match in SAMPLE.finditer(content)]
+    if result.returncode or sampler_match is None or len(workers) < 9 or not samples:
+        raise SystemExit(f"phase census failed or markers missing: {log}; exit={result.returncode}")
+    sampler = dict(FIELDS.findall(sampler_match.group(1)))
+    if sampler.get("status") != "PASS":
+        raise SystemExit(f"L77 sampler failed: {log}: {sampler}")
+    lod_seconds = {int(row["lod"]): float(row["worker_sum_s"]) for row in workers}
+
+    def estimate(scope: str) -> dict:
+        lods = set(range(3)) if scope == "lod0-2" else set(range(9))
+        eligible = [row for row in samples if int(row["lod"]) in lods]
+        weights = {phase: sum(int(row.get("weight", "1")) for row in eligible
+                              if row.get("phase") == phase)
+                   for phase in ("passages", "support")}
+        total_weight = sum(int(row.get("weight", "1")) for row in eligible)
+        cpu_seconds = sum(value for lod, value in lod_seconds.items() if lod in lods)
+        result_row = {"worker_cpu_s": round(cpu_seconds, 4),
+                      "weighted_samples": total_weight}
+        for phase, weight in weights.items():
+            share = weight / total_weight if total_weight else 0.0
+            result_row[f"{phase}_share_pct"] = round(share * 100.0, 3)
+            result_row[f"{phase}_worker_s_estimate"] = round(cpu_seconds * share, 4)
+        return result_row
+
+    row = {
+        "stratum": stratum, "variant": variant, "replicate": replicate,
+        "exit_code": result.returncode,
+        "sampler": sampler,
+        "lod0_2": estimate("lod0-2"), "whole_view": estimate("whole"),
+        "log": str(log.relative_to(ROOT)),
+        "runner_wall_s": round(time.monotonic() - started, 3),
+    }
+    print("[L109PhaseCensusRun] " + json.dumps(row, sort_keys=True), flush=True)
+    return row
+
+
+def main() -> int:
+    if any(not path.is_file() for path in BINARIES.values()):
+        raise SystemExit("missing L77 profile binary")
+    rows = [run(stratum, variant, replicate)
+            for stratum in ("FI", "TN")
+            for replicate in (1, 2)
+            for variant in ("base", "candidate")]
+    summary = []
+    for stratum in ("FI", "TN"):
+        for variant in ("base", "candidate"):
+            entry = {"stratum": stratum, "variant": variant, "replicates": 2}
+            group = [row for row in rows if row["stratum"] == stratum
+                     and row["variant"] == variant]
+            for scope in ("lod0_2", "whole_view"):
+                for key in ("worker_cpu_s", "passages_share_pct", "passages_worker_s_estimate",
+                            "support_share_pct", "support_worker_s_estimate"):
+                    values = [row[scope][key] for row in group]
+                    entry[f"{scope}_{key}_runs"] = values
+                    entry[f"{scope}_{key}_median"] = statistics.median(values)
+            summary.append(entry)
+    (OUT / "summary.json").write_text(
+        json.dumps({"runs": rows, "summary": summary}, indent=2) + "\n",
+        encoding="utf-8")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

Standalone/Research/WorldAssemblerM3/Linux109/RunL109Timing.py

@@ -0,0 +1,105 @@
+#!/usr/bin/env python3
+"""Five interleaved L80 readiness pairs for L109, at W8 and W12."""
+from __future__ import annotations
+
+import json
+import re
+import statistics
+import subprocess
+import time
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[4]
+OUT = ROOT / "Saved/Linux/l109/timing"
+DATA = ROOT / "Standalone/data/owner_world.json"
+BINARIES = {
+    "base": ROOT / "Saved/Linux/l109/bin-baseline/VoxelForgeCoreHeadless-l109-baseline",
+    "candidate": ROOT / "Saved/Linux/l109/bin-final/VoxelForgeCoreHeadless-l109-clang",
+}
+CAMERAS = {
+    0: (8.0, 16.0, -372.2626953125),
+    1: (25.8369140625, 3.19921875, -737.0),
+    2: (19.166015625, -14.2001953125, -753.3740234375),
+    3: (-20.0625, -5.068359375, -900.353515625),
+}
+PIPELINE = re.compile(r"^\[StreamPipeline\] (.*)$", re.M)
+FIELDS = re.compile(r"([a-zA-Z0-9_]+)=([^ ]+)")
+NEAR = re.compile(
+    r"^\[WorldAssemblerNearLODReady\] lods=0-2 tiles=(\d+)/(\d+) cold_ready_s=([0-9.]+)$",
+    re.M,
+)
+
+
+def run(strate: int, workers: int, variant: str, phase: str, replicate: int) -> dict:
+    OUT.mkdir(parents=True, exist_ok=True)
+    label = f"s{strate}-w{workers}-{variant}-{phase}{replicate}"
+    log = OUT / f"{label}.log"
+    camera = ",".join(f"{v:.12f}" for v in CAMERAS[strate]) + ",0,0"
+    command = [
+        "taskset", "-c", "0-11", "prlimit", "--as=6442450944", "--",
+        "timeout", "1740s", str(BINARIES[variant]), str(DATA),
+        "--world-assembler-prototype", "--stream-benchmark", "--stream-pipeline",
+        f"--workers={workers}", "--clip-radius=3", "--max-clip-level=8", "--max-level=8",
+        f"--camera={camera}", "--profile-lod-work", "--stream-scheduler-policy=3",
+    ]
+    start = time.monotonic()
+    with log.open("w", encoding="utf-8") as output:
+        result = subprocess.run(command, cwd=ROOT, stdout=output,
+                                stderr=subprocess.STDOUT, check=False, timeout=1740)
+    content = log.read_text(encoding="utf-8", errors="replace")
+    pipeline_match = PIPELINE.search(content)
+    near_match = NEAR.search(content)
+    if result.returncode or pipeline_match is None or near_match is None:
+        raise SystemExit(f"readiness run failed or trace missing: {log}; exit={result.returncode}")
+    pipeline = dict(FIELDS.findall(pipeline_match.group(1)))
+    row = {
+        "strate": strate, "workers": workers, "variant": variant,
+        "phase": phase, "replicate": replicate, "exit_code": result.returncode,
+        "wall_s": round(time.monotonic() - start, 3),
+        "lod02_ready_s": float(near_match.group(3)),
+        "whole_view_ready_s": float(pipeline["whole_view_ready"].removesuffix("s")),
+        "whole_tiles": int(pipeline["whole_tiles"].split("/")[0]),
+        "log": str(log.relative_to(ROOT)),
+    }
+    print("[L109TimingRun] " + json.dumps(row, sort_keys=True), flush=True)
+    return row
+
+
+def main() -> int:
+    if not DATA.is_file() or any(not binary.is_file() for binary in BINARIES.values()):
+        raise SystemExit("missing data or timing binary")
+    rows = []
+    for strate in range(4):
+        for workers in (8, 12):
+            run(strate, workers, "base", "warmup", 0)
+            run(strate, workers, "candidate", "warmup", 0)
+            for replicate in range(1, 6):
+                order = ("base", "candidate") if replicate % 2 else ("candidate", "base")
+                for variant in order:
+                    rows.append(run(strate, workers, variant, "measured", replicate))
+                    (OUT / "runs-partial.json").write_text(
+                        json.dumps(rows, indent=2) + "\n", encoding="utf-8")
+    summary = []
+    for strate in range(4):
+        for workers in (8, 12):
+            item = {"strate": strate, "workers": workers, "pairs": 5}
+            for variant in ("base", "candidate"):
+                group = [row for row in rows if row["strate"] == strate
+                         and row["workers"] == workers and row["variant"] == variant]
+                for key in ("lod02_ready_s", "whole_view_ready_s"):
+                    values = [row[key] for row in group]
+                    item[f"{variant}_{key}"] = values
+                    item[f"{variant}_{key}_median"] = statistics.median(values)
+            for key in ("lod02_ready_s", "whole_view_ready_s"):
+                base = item[f"base_{key}_median"]
+                candidate = item[f"candidate_{key}_median"]
+                item[f"{key}_delta_pct"] = round(100 * (candidate / base - 1), 2)
+            summary.append(item)
+    (OUT / "summary.json").write_text(
+        json.dumps({"runs": rows, "summary": summary}, indent=2) + "\n",
+        encoding="utf-8")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

Standalone/Research/WorldAssemblerM3/Linux109/RunL109UnrealArrival.py

@@ -0,0 +1,484 @@
+#!/usr/bin/env python3
+"""Run one pinned null-RHI SW/FI arrival for Linux 107."""
+from __future__ import annotations
+
+import argparse
+from collections import Counter
+from datetime import datetime
+import json
+import re
+import subprocess
+import time
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[4]
+HOST = Path("/home/jahni/VFHost")
+HOST_PLUGIN_SAVED = HOST / "Plugins/VoxelForge/Saved/Linux/l109"
+HOST_SAVED = HOST / "Saved/Linux/l109"
+EDITOR = Path("/home/jahni/UnrealEngine/Engine/Binaries/Linux/UnrealEditor-Cmd")
+OUT = ROOT / "Saved/Linux/l109/unreal-arrival"
+SPAWNS_M = {
+    "FI": (25.8369140625, 3.19921875, -737.0),
+    "SW": (25.8369140625, 3.19921875, -200.0),
+}
+MOVE_START_S = 20.0
+MOVE_DURATION_S = 12.0
+EXIT_S = MOVE_START_S + MOVE_DURATION_S + 5.0
+RUN_TIMEOUT_S = 1740
+
+
+def read_events(path: Path) -> list[dict]:
+    events = []
+    if path.is_file():
+        for line in path.read_text(encoding="utf-8", errors="replace").splitlines():
+            if line.startswith("{"):
+                try:
+                    events.append(json.loads(line))
+                except json.JSONDecodeError:
+                    pass
+    return events
+
+
+def percentile(values: list[float], fraction: float) -> float | None:
+    if not values:
+        return None
+    ordered = sorted(values)
+    return ordered[min(len(ordered) - 1, int(round((len(ordered) - 1) * fraction)))]
+
+
+def sample_unreal_peak_rss() -> tuple[int, int] | None:
+    executable = str(EDITOR).encode()
+    project = str(HOST / "VFHost.uproject").encode()
+    peak: tuple[int, int] | None = None
+    for process in Path("/proc").glob("[0-9]*"):
+        try:
+            command_line = (process / "cmdline").read_bytes()
+            if executable not in command_line or project not in command_line:
+                continue
+            status = (process / "status").read_text(encoding="ascii", errors="ignore")
+            match = re.search(r"^VmHWM:\s+(\d+)\s+kB$", status, re.MULTILINE)
+            if match:
+                rss_kib = int(match.group(1))
+                if peak is None or rss_kib > peak[0]:
+                    peak = (rss_kib, int(process.name))
+        except (FileNotFoundError, PermissionError, ProcessLookupError, OSError):
+            continue
+    return peak
+
+
+def endphysics_gap(stdout: Path) -> tuple[float | None, int | None]:
+    release_times: dict[int, dict[int, datetime]] = {}
+    pattern = re.compile(
+        r"^\[(\d{4}\.\d{2}\.\d{2}-\d{2}\.\d{2}\.\d{2}:\d{3})\].*"
+        r"LogTick: tick\s+(\d+)\s+-+\s+Release tick group ([34])")
+    with stdout.open(encoding="utf-8", errors="replace") as stream:
+        for line in stream:
+            match = pattern.match(line)
+            if not match:
+                continue
+            timestamp = datetime.strptime(match.group(1), "%Y.%m.%d-%H.%M.%S:%f")
+            frame = int(match.group(2))
+            group = int(match.group(3))
+            release_times.setdefault(frame, {})[group] = timestamp
+    gaps = [(round((times[4] - times[3]).total_seconds() * 1000.0, 3), frame)
+            for frame, times in release_times.items() if 3 in times and 4 in times]
+    return max(gaps, default=(None, None))
+
+
+def run(stratum: str, repeat: int, phase: str, ready_depth: int,
+        background_cap: int, worker_reserve: int, helper_background: int,
+        tile_background_low: int, tile_background_normal: int, startup_boost: int,
+        tick_logs: bool, insights: bool) -> dict:
+    label = (f"{phase}-{stratum.lower()}-rd{ready_depth}-bg{background_cap}"
+             f"-reserve{worker_reserve}-helperbg{helper_background}"
+             f"-tilelow{tile_background_low}-tilenormal{tile_background_normal}"
+             f"-ticklog{int(tick_logs)}"
+             f"-boost{startup_boost}-{repeat}")
+    trace = OUT / f"{label}.trace.jsonl"
+    utrace = OUT / f"{label}.utrace"
+    abslog = OUT / f"{label}.log"
+    stdout = OUT / f"{label}.stdout.log"
+    output = HOST_PLUGIN_SAVED / "output" / label
+    userdir = HOST_SAVED / "user" / label
+    zendata = HOST_SAVED / "zen" / label
+    ddc = HOST_SAVED / "DDC"
+    for directory in (OUT, output, userdir, zendata, ddc):
+        directory.mkdir(parents=True, exist_ok=True)
+    trace.unlink(missing_ok=True)
+    utrace.unlink(missing_ok=True)
+    spawn_cm = ",".join(f"{value * 100.0:.3f}" for value in SPAWNS_M[stratum])
+    command = [
+        str(EDITOR), str(HOST / "VFHost.uproject"), "-nullrhi", "-unattended",
+        "-nop4", "-nosplash", "-NoSound", "-corelimit=12", "-NoZenAutoLaunch",
+        "-notraceserver", "-nocrashreports",
+        "-DisablePlugins=AndroidFileServer",
+        "-ddc=NoZenLocalFallback", f"-userdir={userdir}", f"-ZenDataPath={zendata}",
+        f"-LocalDataCachePath={ddc}", "-run=VoxelForgeExplore", "-seed=0",
+        "-archetype=TunnelNetwork", f"-slot={1 if stratum == 'FI' else 0}",
+        "-modes=export", "-gameconfig=1", "-coreworldtest=1",
+        "-ownerfourstrateplan=1", "-corevoxelsizecm=25", "-coreislandtest=0",
+        "-corefallbacktest=0", "-coreofftest=0", "-coreplantest=0",
+        "-settingsref=/Game/VoxelForge/DA_Settings.DA_Settings", "-corehybridmesher=0",
+        f"-out={output}", "-voxel.StreamingCoarseFirst=1",
+        "-voxel.StreamingDynamicNearReserve=1",
+        "-voxel.StreamingLegacyPendingResultCap=0",
+        "-voxel.CoreRegionConcurrentAdmission=1",
+        "-voxel.StreamingResultBacklogCap=256",
+        f"-voxel.StreamingReadyQueueDepth={ready_depth}",
+        f"-voxel.StreamingBackgroundWorkerCap={background_cap}",
+        f"-voxel.StreamingBackgroundWorkerReserve={worker_reserve}",
+        f"-voxel.CoreParallelForBackgroundPriority={helper_background}",
+        f"-voxel.StreamingTileBackgroundLow={tile_background_low}",
+        f"-voxel.StreamingTileBackgroundNormal={tile_background_normal}",
+        f"-voxel.TestLogTickFunctions={int(tick_logs)}",
+        "-voxel.CoreColumnCacheDirectMapped=0", "-voxel.CoreMCGradientMemoization=1",
+        "-voxel.StreamingCollisionLookahead=1",
+        f"-voxel.StreamingStartupApplyBoost={startup_boost}",
+        "-voxel.TestMove=1", "-voxel.TestMoveSpeedCmPerSecond=140.0",
+        f"-voxel.TestMoveStartSeconds={MOVE_START_S:.1f}",
+        f"-voxel.TestMoveDurationSeconds={MOVE_DURATION_S:.1f}",
+        f"-voxel.TestExitSeconds={EXIT_S:.1f}", "-voxel.TestMoveDirection=1,0,0",
+        "-voxel.StreamingSchedulerPolicy=3", "-voxel.TestExitOnSteady=0",
+        "-coreworldstreamtrace=1", f"-voxel.StartupTraceSpawnCm={spawn_cm}",
+        "-voxel.StartupTraceUntilEndPlay=1", "-voxel.StartupTraceThroughCrossing=1",
+        "-voxel.StartupTraceMaxTiles=16384", f"-voxel.StartupTraceFile={trace}",
+        f"-abslog={abslog}",
+    ]
+    if insights:
+        command += ["-trace=cpu,task,frame", "-statnamedevents",
+                    f"-tracefile={utrace}", "-tracefiletrunc"]
+    wrapped = ["taskset", "-c", "0-11", "timeout", f"{RUN_TIMEOUT_S}s", *command]
+    started = time.monotonic()
+    with stdout.open("w", encoding="utf-8") as stream:
+        process = subprocess.Popen(wrapped, cwd=ROOT, stdout=stream,
+                                   stderr=subprocess.STDOUT)
+        peak_rss = None
+        while process.poll() is None:
+            observed = sample_unreal_peak_rss()
+            if observed and (peak_rss is None or observed[0] > peak_rss[0]):
+                peak_rss = observed
+            time.sleep(0.25)
+        exit_code = process.wait()
+        observed = sample_unreal_peak_rss()
+        if observed and (peak_rss is None or observed[0] > peak_rss[0]):
+            peak_rss = observed
+
+    events = read_events(trace)
+    worker_backend_config = next(
+        (e for e in events if e.get("type") == "worker_backend_config"), {})
+    worker_cap_readback = next(
+        (e for e in events if e.get("type") == "streaming_worker_cap_readback"), {})
+    samples = [e for e in events if e.get("type") == "streaming_sample"]
+    arrival = [e for e in samples if float(e.get("t_s", 0.0)) < MOVE_START_S]
+    full = next((e for e in arrival
+                 if sum(e.get("desired_by_lod", [0] * 9)[:3]) > 0
+                 and sum(e.get("settled_by_lod", [0] * 9)[:3])
+                 >= sum(e.get("desired_by_lod", [0] * 9)[:3])), None)
+    first_admission = next((e for e in events if e.get("type") == "scheduler_admission"
+                            and e.get("desired_rebuilt") == 1), None)
+    request_events = [e for e in events if e.get("type") == "request"]
+    initial_request_events = [e for e in request_events if e.get("epoch") == 0]
+    tile_by_ordinal = {e.get("request_ordinal"): e for e in events
+                       if e.get("type") == "tile"}
+    collision_count = sum(1 for e in events if e.get("type") == "collision_ready"
+                          and e.get("level", 9) <= 2
+                          and full and e.get("t_s", 1e30) <= full["t_s"])
+    initial_tiles = [tile_by_ordinal[r.get("request_ordinal")] for r in initial_request_events
+                     if full and r.get("t_s", 1e30) <= full["t_s"]
+                     and r.get("request_ordinal") in tile_by_ordinal
+                     and tile_by_ordinal[r["request_ordinal"]].get("level", 9) <= 2]
+    burst_start_s = float(first_admission["t_s"]) if first_admission else -1.0
+    burst_end_s = float(full["t_s"]) if full else -1.0
+    work_by_frame = {int(e.get("frame", -1)): e for e in events
+                     if e.get("type") == "game_thread_work_frame"}
+    work_values = [e for e in work_by_frame.values()
+                   if burst_start_s <= float(e.get("t_s", -1.0)) <= burst_end_s]
+    arrival_frame_ids = {int(e.get("frame", -1)) for e in work_values}
+    scheduler_by_frame = {int(e.get("frame", -1)): e for e in events
+                          if e.get("type") == "scheduler_admission"}
+    scheduler_values = [e for e in events if e.get("type") == "scheduler_admission"
+                        and burst_start_s <= float(e.get("t_s", -1.0)) <= burst_end_s]
+    apply_by_frame = {int(e.get("frame", -1)): e for e in events
+                      if e.get("type") == "apply_frame"}
+    commandlet_tick_by_frame = {int(e.get("frame", -1)): e for e in events
+                                if e.get("type") == "commandlet_world_tick_frame"}
+    garbage_collect_by_frame = {int(e.get("frame", -1)): e for e in events
+                                if e.get("type") == "garbage_collect"}
+    # apply_frame N+1 records the interval containing AVoxelWorld::Tick N. Its apply_ms
+    # belongs to Tick N+1, so use apply_by_frame[N] for the same-tick apply breakdown.
+    near_intervals = [apply_by_frame[frame + 1] for frame in arrival_frame_ids
+                      if frame + 1 in apply_by_frame]
+    rmc_by_frame: dict[int, dict[str, float]] = {}
+    for event in events:
+        frame = int(event.get("frame", -1))
+        if event.get("type") != "rmc_submit" or frame not in arrival_frame_ids:
+            continue
+        values = rmc_by_frame.setdefault(frame, {
+            "component_acquire_ms": 0.0,
+            "create_section_group_ms": 0.0,
+            "create_sections_ms": 0.0,
+            "collision_handoff_ms": 0.0,
+        })
+        for key in values:
+            values[key] += float(event.get(key, 0.0))
+    collision_completion_by_frame: dict[int, float] = {}
+    for event in events:
+        frame = int(event.get("frame", -1))
+        if event.get("type") == "collision_completion" and frame in arrival_frame_ids:
+            collision_completion_by_frame[frame] = collision_completion_by_frame.get(frame, 0.0) \
+                + float(event.get("handler_ms", 0.0))
+    near_requests = [r for r in request_events if full and r.get("t_s", 1e30) <= full["t_s"]
+                     and r.get("level", 9) <= 2]
+    initial_lod0_counts = Counter(
+        tuple(r.get("tile", ())) for r in near_requests if r.get("level") == 0)
+    repeated_lod0_keys = [
+        {"tile": list(tile), "requests": count}
+        for tile, count in sorted(initial_lod0_counts.items()) if count > 1]
+    initial_scheduler_frames = [e for e in events
+        if e.get("type") == "scheduler_admission" and full
+        and e.get("t_s", 1e30) <= full["t_s"]]
+    worst_interval = max(near_intervals, key=lambda e: float(e.get("frame_ms", 0.0)), default=None)
+    worst_interval_frame_id = int(worst_interval.get("frame", -1)) if worst_interval else -1
+    worst_work_frame_id = worst_interval_frame_id - 1
+    worst_work = work_by_frame.get(worst_work_frame_id, {})
+    worst_apply = apply_by_frame.get(worst_work_frame_id, {})
+    worst_scheduler = scheduler_by_frame.get(worst_work_frame_id, {})
+    worst_rmc = rmc_by_frame.get(worst_work_frame_id, {})
+    worst_collision_completion_ms = collision_completion_by_frame.get(worst_work_frame_id, 0.0)
+    rmc_values = [rmc_by_frame.get(frame, {}) for frame in arrival_frame_ids]
+    apply_values = [apply_by_frame[frame] for frame in arrival_frame_ids
+                    if frame in apply_by_frame]
+    commandlet_tick_values = [commandlet_tick_by_frame[frame] for frame in arrival_frame_ids
+                              if frame in commandlet_tick_by_frame]
+    garbage_collect_values = [garbage_collect_by_frame[frame] for frame in arrival_frame_ids
+                              if frame in garbage_collect_by_frame]
+
+    def work_stat(name: str, fraction: float | None = None) -> float | None:
+        values = [float(e.get(name, 0.0)) for e in work_values]
+        return percentile(values, fraction) if fraction is not None else max(values, default=0.0)
+
+    def scheduler_stat(name: str, fraction: float | None = None) -> float | None:
+        values = [float(e.get(name, 0.0)) for e in scheduler_values]
+        return percentile(values, fraction) if fraction is not None else max(values, default=0.0)
+
+    def rmc_stat(name: str, fraction: float | None = None) -> float | None:
+        values = [float(e.get(name, 0.0)) for e in rmc_values]
+        return percentile(values, fraction) if fraction is not None else max(values, default=0.0)
+
+    def commandlet_stat(name: str, fraction: float | None = None) -> float | None:
+        values = [float(e.get(name, 0.0)) for e in commandlet_tick_values]
+        return percentile(values, fraction) if fraction is not None else max(values, default=0.0)
+
+    row = {
+        "label": label, "stratum": stratum, "repeat": repeat, "phase": phase,
+        "ready_queue_depth": ready_depth,
+        "requested_background_worker_cap": background_cap,
+        "background_worker_cap": worker_cap_readback.get(
+            "streaming_background_worker_cap", background_cap),
+        "requested_background_worker_reserve": worker_reserve,
+        "background_worker_reserve": worker_backend_config.get(
+            "streaming_background_worker_reserve", worker_cap_readback.get(
+                "background_worker_reserve", worker_reserve)),
+        "requested_core_parallel_for_background_priority": helper_background,
+        "core_parallel_for_background_priority": worker_backend_config.get(
+            "core_parallel_for_background_priority", helper_background),
+        "requested_streaming_tile_background_low": tile_background_low,
+        "streaming_tile_background_low": worker_backend_config.get(
+            "streaming_tile_background_low", tile_background_low),
+        "requested_streaming_tile_background_normal": tile_background_normal,
+        "streaming_tile_background_normal": worker_backend_config.get(
+            "streaming_tile_background_normal", tile_background_normal),
+        "effective_tile_slots": worker_backend_config.get("effective_tile_slots"),
+        "background_worker_threads": worker_backend_config.get("taskgraph_background_threads"),
+        "startup_apply_boost": startup_boost, "insights": insights,
+        "engine_tick_logs": tick_logs,
+        "exit_code": exit_code,
+        "launch_wrapper": "taskset -c 0-11 timeout 1740s (no address-space limit)",
+        "peak_rss_kib": peak_rss[0] if peak_rss else None,
+        "peak_rss_mib": round(peak_rss[0] / 1024.0, 1) if peak_rss else None,
+        "peak_rss_pid": peak_rss[1] if peak_rss else None,
+        "endphysics_to_postphysics_gap_ms_max": None,
+        "endphysics_gap_frame": None,
+        "wall_s": round(time.monotonic() - started, 3),
+        "desired_t_s": next((e.get("t_s") for e in events if e.get("type") == "desired_set"), None),
+        "first_admission_t_s": first_admission.get("t_s") if first_admission else None,
+        "lod02_full_settled_t_s": full.get("t_s") if full else None,
+        "near_settle_from_first_admission_s": round(
+            float(full["t_s"]) - float(first_admission["t_s"]), 4)
+            if full and first_admission else None,
+        "desired_by_lod": full.get("desired_by_lod", [])[:3] if full else None,
+        "settled_by_lod": full.get("settled_by_lod", [])[:3] if full else None,
+        "near_admitted_tiles": len(near_requests),
+        "initial_lod0_admissions": sum(initial_lod0_counts.values()),
+        "repeated_lod0_request_keys_before_settle": repeated_lod0_keys,
+        "peak_result_backlog_before_settle": max(
+            (int(e.get("result_backlog", 0)) for e in initial_scheduler_frames), default=0),
+        "last_admission_t_s": max((e.get("t_s", 0.0) for e in near_requests), default=None),
+        "near_tile_events": len(initial_tiles),
+        "near_worker_seconds": round(sum(e.get("worker_task_s", 0.0) for e in initial_tiles), 4),
+        "near_core_seconds": round(sum(e.get("core_generate_s", 0.0) for e in initial_tiles), 4),
+        "collision_ready_by_settle": collision_count,
+        "burst_frame_count": len(near_intervals),
+        "burst_frame_ms_p95": round(sorted(e.get("frame_ms", 0.0) for e in near_intervals)[
+            min(len(near_intervals) - 1, int(.95 * (len(near_intervals) - 1)))], 3)
+            if near_intervals else None,
+        "burst_frame_ms_max": round(max((e.get("frame_ms", 0.0) for e in near_intervals), default=0.0), 3),
+        "burst_full_world_tick_ms_p99": round(commandlet_stat("world_tick_ms", .99) or 0.0, 3),
+        "burst_full_world_tick_ms_max": round(commandlet_stat("world_tick_ms") or 0.0, 3),
+        "burst_world_pre_actor_ms_p99": round(commandlet_stat("world_pre_actor_ms", .99) or 0.0, 3),
+        "burst_world_pre_actor_ms_max": round(commandlet_stat("world_pre_actor_ms") or 0.0, 3),
+        "burst_world_actor_phase_ms_p99": round(commandlet_stat("world_actor_phase_ms", .99) or 0.0, 3),
+        "burst_world_actor_phase_ms_max": round(commandlet_stat("world_actor_phase_ms") or 0.0, 3),
+        "burst_world_post_actor_ms_p99": round(commandlet_stat("world_post_actor_ms", .99) or 0.0, 3),
+        "burst_world_post_actor_ms_max": round(commandlet_stat("world_post_actor_ms") or 0.0, 3),
+        "burst_commandlet_readiness_check_ms_max": round(
+            commandlet_stat("readiness_check_ms") or 0.0, 3),
+        "burst_commandlet_sleep_ms_p99": round(commandlet_stat("sleep_ms", .99) or 0.0, 3),
+        "burst_commandlet_sleep_ms_max": round(commandlet_stat("sleep_ms") or 0.0, 3),
+        "burst_commandlet_loop_ms_max": round(commandlet_stat("loop_ms") or 0.0, 3),
+        "burst_gc_count": len(garbage_collect_values),
+        "burst_gc_duration_ms_max": round(max(
+            (float(e.get("duration_ms", 0.0)) for e in garbage_collect_values), default=0.0), 3),
+        "burst_max_meshes_per_frame": max((e.get("meshes_applied", 0) for e in apply_values), default=0),
+        "burst_max_apply_ms": round(max((e.get("apply_ms", 0.0) for e in apply_values), default=0.0), 3),
+        "burst_max_admit_per_frame": max((sum(e.get("submitted_by_lod", [0] * 9)[:3])
+                                            for e in scheduler_values), default=0),
+        "burst_frame_timing_count": len(work_values),
+        "burst_active_generation_workers_p95": round(work_stat("active_generation_workers", .95) or 0.0, 3),
+        "burst_active_generation_workers_max": round(work_stat("active_generation_workers") or 0.0, 3),
+        "burst_active_tile_tasks_max": round(work_stat("active_tile_tasks") or 0.0, 3),
+        "burst_world_tick_ms_p99": round(work_stat("tick_ms", .99) or 0.0, 3),
+        "burst_world_tick_ms_max": round(work_stat("tick_ms") or 0.0, 3),
+        "burst_update_chunks_ms_p99": round(work_stat("update_chunks_ms", .99) or 0.0, 3),
+        "burst_update_chunks_ms_max": round(work_stat("update_chunks_ms") or 0.0, 3),
+        "burst_pending_drain_ms_p99": round(work_stat("pending_drain_ms", .99) or 0.0, 3),
+        "burst_pending_drain_ms_max": round(work_stat("pending_drain_ms") or 0.0, 3),
+        "burst_post_apply_callbacks_ms_p99": round(work_stat("post_apply_callbacks_ms", .99) or 0.0, 3),
+        "burst_post_apply_callbacks_ms_max": round(work_stat("post_apply_callbacks_ms") or 0.0, 3),
+        "burst_visibility_reconcile_ms_p99": round(work_stat("visibility_reconcile_ms", .99) or 0.0, 3),
+        "burst_visibility_reconcile_ms_max": round(work_stat("visibility_reconcile_ms") or 0.0, 3),
+        "burst_collision_probe_ms_max": round(work_stat("collision_probe_ms") or 0.0, 3),
+        "burst_visible_apply_probe_ms_max": round(work_stat("visible_apply_probe_ms") or 0.0, 3),
+        "burst_lod0_ready_ms_max": round(work_stat("lod0_ready_ms") or 0.0, 3),
+        "burst_steady_trace_ms_max": round(work_stat("steady_trace_ms") or 0.0, 3),
+        "burst_modification_ms_max": round(work_stat("modification_ms") or 0.0, 3),
+        "burst_lifecycle_audit_ms_p99": round(work_stat("lifecycle_audit_ms", .99) or 0.0, 3),
+        "burst_lifecycle_audit_ms_max": round(work_stat("lifecycle_audit_ms") or 0.0, 3),
+        "burst_collision_gate_ms_p99": round(work_stat("collision_gate_ms", .99) or 0.0, 3),
+        "burst_collision_gate_ms_max": round(work_stat("collision_gate_ms") or 0.0, 3),
+        "burst_unload_ms_p99": round(work_stat("unload_ms", .99) or 0.0, 3),
+        "burst_unload_ms_max": round(work_stat("unload_ms") or 0.0, 3),
+        "burst_trace_json_ms_p99": round(work_stat("trace_json_ms", .99) or 0.0, 3),
+        "burst_trace_json_ms_max": round(work_stat("trace_json_ms") or 0.0, 3),
+        "burst_desired_rebuild_ms_max": round(scheduler_stat("desired_rebuild_ms") or 0.0, 3),
+        "burst_rmc_create_group_ms_p99": round(rmc_stat("create_section_group_ms", .99) or 0.0, 3),
+        "burst_rmc_create_group_ms_max": round(rmc_stat("create_section_group_ms") or 0.0, 3),
+        "burst_rmc_collision_handoff_ms_max": round(rmc_stat("collision_handoff_ms") or 0.0, 3),
+        "burst_collision_completion_ms_max": round(max(
+            collision_completion_by_frame.values(), default=0.0), 3),
+        "worst_frame_breakdown": {
+            "frame": worst_work_frame_id,
+            "world_tick_frame": worst_work_frame_id,
+            "frame_interval_end_frame": worst_interval_frame_id,
+            "frame_interval_ms": round(float(worst_interval.get("frame_ms", 0.0)), 3)
+                if worst_interval else None,
+            "apply_ms": round(float(worst_apply.get("apply_ms", 0.0)), 3)
+                if worst_apply else None,
+            "result_drain_ms": round(float(worst_apply.get("drain_ms", 0.0)), 3)
+                if worst_apply else None,
+            "outside_world_tick_ms": round(max(0.0, float(worst_interval.get("frame_ms", 0.0))
+                - float(worst_work.get("tick_ms", 0.0))), 3)
+                if worst_interval else None,
+            "outside_full_world_tick_ms": round(max(0.0, float(worst_interval.get("frame_ms", 0.0))
+                - float(commandlet_tick_by_frame.get(worst_work_frame_id, {}).get("world_tick_ms", 0.0))), 3)
+                if worst_interval else None,
+            "full_world_tick_ms": round(float(commandlet_tick_by_frame.get(
+                worst_work_frame_id, {}).get("world_tick_ms", 0.0)), 3),
+            "world_pre_actor_ms": round(float(commandlet_tick_by_frame.get(
+                worst_work_frame_id, {}).get("world_pre_actor_ms", 0.0)), 3),
+            "world_actor_phase_ms": round(float(commandlet_tick_by_frame.get(
+                worst_work_frame_id, {}).get("world_actor_phase_ms", 0.0)), 3),
+            "world_post_actor_ms": round(float(commandlet_tick_by_frame.get(
+                worst_work_frame_id, {}).get("world_post_actor_ms", 0.0)), 3),
+            "commandlet_readiness_check_ms": round(float(commandlet_tick_by_frame.get(
+                worst_work_frame_id, {}).get("readiness_check_ms", 0.0)), 3),
+            "commandlet_sleep_ms": round(float(commandlet_tick_by_frame.get(
+                worst_work_frame_id, {}).get("sleep_ms", 0.0)), 3),
+            "commandlet_loop_ms": round(float(commandlet_tick_by_frame.get(
+                worst_work_frame_id, {}).get("loop_ms", 0.0)), 3),
+            "gc_duration_ms": round(float(garbage_collect_by_frame.get(
+                worst_work_frame_id, {}).get("duration_ms", 0.0)), 3),
+            "world_tick_ms": round(float(worst_work.get("tick_ms", 0.0)), 3),
+            "active_generation_workers": int(worst_work.get("active_generation_workers", 0)),
+            "active_tile_tasks": int(worst_work.get("active_tile_tasks", 0)),
+            "collision_gate_ms": round(float(worst_work.get("collision_gate_ms", 0.0)), 3),
+            "update_chunks_ms": round(float(worst_work.get("update_chunks_ms", 0.0)), 3),
+            "desired_rebuild_ms": round(float(worst_scheduler.get("desired_rebuild_ms", 0.0)), 3),
+            "pending_drain_ms": round(float(worst_work.get("pending_drain_ms", 0.0)), 3),
+            "post_apply_callbacks_ms": round(float(worst_work.get("post_apply_callbacks_ms", 0.0)), 3),
+            "collision_probe_ms": round(float(worst_work.get("collision_probe_ms", 0.0)), 3),
+            "visibility_reconcile_ms": round(float(worst_work.get("visibility_reconcile_ms", 0.0)), 3),
+            "visible_apply_probe_ms": round(float(worst_work.get("visible_apply_probe_ms", 0.0)), 3),
+            "lod0_ready_ms": round(float(worst_work.get("lod0_ready_ms", 0.0)), 3),
+            "unload_ms": round(float(worst_work.get("unload_ms", 0.0)), 3),
+            "steady_trace_ms": round(float(worst_work.get("steady_trace_ms", 0.0)), 3),
+            "modification_ms": round(float(worst_work.get("modification_ms", 0.0)), 3),
+            "lifecycle_audit_ms": round(float(worst_work.get("lifecycle_audit_ms", 0.0)), 3),
+            "trace_json_ms": round(float(worst_work.get("trace_json_ms", 0.0)), 3),
+            "trace_archive_serialize_ms": round(float(
+                worst_work.get("trace_archive_serialize_ms", 0.0)), 3),
+            "collision_completion_ms": round(worst_collision_completion_ms, 3),
+            **{key: round(float(worst_rmc.get(key, 0.0)), 3)
+               for key in ("component_acquire_ms", "create_section_group_ms",
+                           "create_sections_ms", "collision_handoff_ms")},
+        },
+        "trace": str(trace.relative_to(ROOT)), "utrace": str(utrace.relative_to(ROOT)),
+        "abslog": str(abslog.relative_to(ROOT)), "stdout": str(stdout.relative_to(ROOT)),
+    }
+    stdout_text = stdout.read_text(encoding="utf-8", errors="replace")
+    gap_ms, gap_frame = endphysics_gap(stdout)
+    row["endphysics_to_postphysics_gap_ms_max"] = gap_ms
+    row["endphysics_gap_frame"] = gap_frame
+    failure = re.search(r"\[VoxelForgeCoreWorldTest\] failed: ([^\r\n]+)", stdout_text)
+    if failure:
+        row["world_test_failure"] = failure.group(1)
+    with (OUT / "runs.jsonl").open("a", encoding="utf-8") as stream:
+        stream.write(json.dumps(row, sort_keys=True) + "\n")
+    (OUT / f"{label}.summary.json").write_text(json.dumps(row, indent=2) + "\n",
+                                                 encoding="utf-8")
+    print("[L109UnrealArrival] " + json.dumps(row, sort_keys=True), flush=True)
+    if exit_code:
+        raise SystemExit(f"Unreal commandlet failed: {stdout}; exit={exit_code}")
+    if not full or len(arrival) < 8:
+        raise SystemExit(f"Unreal arrival trace incomplete: {stdout}; exit={exit_code}")
+    if insights and (not utrace.is_file() or utrace.stat().st_size < 1024):
+        raise SystemExit(f"Unreal Insights trace missing or empty: {utrace}")
+    return row
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--stratum", choices=("FI", "SW"), required=True)
+    parser.add_argument("--repeat", type=int, default=1)
+    parser.add_argument("--phase", choices=("probe", "before", "after", "snapshot"), default="probe")
+    parser.add_argument("--ready-depth", type=int, default=64)
+    parser.add_argument("--background-cap", type=int, choices=(0, 1), default=1)
+    parser.add_argument("--worker-reserve", type=int, choices=(0, 1, 2, 3), default=0)
+    parser.add_argument("--helper-background", type=int, choices=(0, 1), default=1)
+    parser.add_argument("--tile-background-low", type=int, choices=(0, 1), default=1)
+    parser.add_argument("--tile-background-normal", type=int, choices=(0, 1), default=0)
+    parser.add_argument("--startup-boost", type=int, choices=(0, 1), default=1)
+    parser.add_argument("--tick-logs", action="store_true")
+    parser.add_argument("--insights", action="store_true")
+    args = parser.parse_args()
+    if not EDITOR.is_file():
+        raise SystemExit(f"missing Unreal commandlet executable: {EDITOR}")
+    run(args.stratum, args.repeat, args.phase, args.ready_depth,
+        args.background_cap, args.worker_reserve, args.helper_background,
+        args.tile_background_low, args.tile_background_normal, args.startup_boost,
+        args.tick_logs, args.insights)
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

Standalone/Research/WorldAssemblerM3/Linux109/captures/INDEX.md

@@ -0,0 +1,93 @@
+# L109 passage captures
+
+Each PNG was rendered from its own exported mesh. The renderer reported visible triangles, and each decoded image met the pixel coverage floor. `low_contrast_geometry_visible` marks images that met this floor below the preferred contrast target.
+
+## Before
+
+- [upper-mouth](before/passage-0-1-0-upper-mouth.png) — passage/0-1/0; 188440 visible triangles
+- [lower-mouth](before/passage-0-1-0-lower-mouth.png) — passage/0-1/0; 564520 visible triangles
+- [25%](before/passage-0-1-0-station-25.png) — passage/0-1/0; 234197 visible triangles
+- [50%](before/passage-0-1-0-station-50.png) — passage/0-1/0; 315900 visible triangles
+- [75%](before/passage-0-1-0-station-75.png) — passage/0-1/0; 256359 visible triangles
+- [section-1/view-1](before/passage-0-1-0-openair-1-view-1.png) — passage/0-1/0 **OPEN AIR**; 211462 visible triangles
+- [section-1/view-2](before/passage-0-1-0-openair-1-view-2.png) — passage/0-1/0 **OPEN AIR**; 208255 visible triangles
+- [section-1/view-3](before/passage-0-1-0-openair-1-view-3.png) — passage/0-1/0 **OPEN AIR**; 221239 visible triangles
+- [section-2/view-1](before/passage-0-1-0-openair-2-view-1.png) — passage/0-1/0 **OPEN AIR**; 218664 visible triangles
+- [section-2/view-2](before/passage-0-1-0-openair-2-view-2.png) — passage/0-1/0 **OPEN AIR**; 223817 visible triangles
+- [section-2/view-3](before/passage-0-1-0-openair-2-view-3.png) — passage/0-1/0 **OPEN AIR**; 221239 visible triangles
+- [section-3/view-1](before/passage-0-1-0-openair-3-view-1.png) — passage/0-1/0 **OPEN AIR**; 224005 visible triangles
+- [section-3/view-2](before/passage-0-1-0-openair-3-view-2.png) — passage/0-1/0 **OPEN AIR**; 230599 visible triangles; low_contrast_geometry_visible
+- [section-3/view-3](before/passage-0-1-0-openair-3-view-3.png) — passage/0-1/0 **OPEN AIR**; 229291 visible triangles
+- [section-4/view-1](before/passage-0-1-0-openair-4-view-1.png) — passage/0-1/0 **OPEN AIR**; 407353 visible triangles
+- [section-4/view-2](before/passage-0-1-0-openair-4-view-2.png) — passage/0-1/0 **OPEN AIR**; 406732 visible triangles
+- [section-4/view-3](before/passage-0-1-0-openair-4-view-3.png) — passage/0-1/0 **OPEN AIR**; 363878 visible triangles
+- [upper-mouth](before/passage-1-2-0-upper-mouth.png) — passage/1-2/0; 576595 visible triangles
+- [lower-mouth](before/passage-1-2-0-lower-mouth.png) — passage/1-2/0; 555286 visible triangles
+- [25%](before/passage-1-2-0-station-25.png) — passage/1-2/0; 667602 visible triangles
+- [50%](before/passage-1-2-0-station-50.png) — passage/1-2/0; 602756 visible triangles
+- [75%](before/passage-1-2-0-station-75.png) — passage/1-2/0; 620502 visible triangles
+- [section-1/view-1](before/passage-1-2-0-openair-1-view-1.png) — passage/1-2/0 **OPEN AIR**; 636438 visible triangles
+- [section-1/view-2](before/passage-1-2-0-openair-1-view-2.png) — passage/1-2/0 **OPEN AIR**; 593960 visible triangles
+- [section-1/view-3](before/passage-1-2-0-openair-1-view-3.png) — passage/1-2/0 **OPEN AIR**; 622040 visible triangles
+- [upper-mouth](before/passage-2-3-0-upper-mouth.png) — passage/2-3/0; 59568 visible triangles
+- [lower-mouth](before/passage-2-3-0-lower-mouth.png) — passage/2-3/0; 294430 visible triangles; low_contrast_geometry_visible
+- [25%](before/passage-2-3-0-station-25.png) — passage/2-3/0; 55420 visible triangles; low_contrast_geometry_visible
+- [50%](before/passage-2-3-0-station-50.png) — passage/2-3/0; 458134 visible triangles
+- [75%](before/passage-2-3-0-station-75.png) — passage/2-3/0; 475978 visible triangles
+- [section-1/view-1](before/passage-2-3-0-openair-1-view-1.png) — passage/2-3/0 **OPEN AIR**; 104612 visible triangles
+- [section-1/view-2](before/passage-2-3-0-openair-1-view-2.png) — passage/2-3/0 **OPEN AIR**; 81078 visible triangles
+- [section-1/view-3](before/passage-2-3-0-openair-1-view-3.png) — passage/2-3/0 **OPEN AIR**; 94962 visible triangles; low_contrast_geometry_visible
+- [section-2/view-1](before/passage-2-3-0-openair-2-view-1.png) — passage/2-3/0 **OPEN AIR**; 476052 visible triangles
+- [section-2/view-2](before/passage-2-3-0-openair-2-view-2.png) — passage/2-3/0 **OPEN AIR**; 479198 visible triangles
+- [section-2/view-3](before/passage-2-3-0-openair-2-view-3.png) — passage/2-3/0 **OPEN AIR**; 479198 visible triangles
+- [section-3/view-1](before/passage-2-3-0-openair-3-view-1.png) — passage/2-3/0 **OPEN AIR**; 485580 visible triangles
+- [section-3/view-2](before/passage-2-3-0-openair-3-view-2.png) — passage/2-3/0 **OPEN AIR**; 478866 visible triangles
+- [section-3/view-3](before/passage-2-3-0-openair-3-view-3.png) — passage/2-3/0 **OPEN AIR**; 490304 visible triangles; low_contrast_geometry_visible
+- [section-4/view-1](before/passage-2-3-0-openair-4-view-1.png) — passage/2-3/0 **OPEN AIR**; 475662 visible triangles
+- [section-4/view-2](before/passage-2-3-0-openair-4-view-2.png) — passage/2-3/0 **OPEN AIR**; 496128 visible triangles
+- [section-4/view-3](before/passage-2-3-0-openair-4-view-3.png) — passage/2-3/0 **OPEN AIR**; 496508 visible triangles; low_contrast_geometry_visible
+
+## After
+
+- [upper-mouth](after/passage-0-1-0-upper-mouth.png) — passage/0-1/0; 253282 visible triangles
+- [lower-mouth](after/passage-0-1-0-lower-mouth.png) — passage/0-1/0; 559326 visible triangles
+- [25%](after/passage-0-1-0-station-25.png) — passage/0-1/0; 145630 visible triangles
+- [50%](after/passage-0-1-0-station-50.png) — passage/0-1/0; 430386 visible triangles
+- [75%](after/passage-0-1-0-station-75.png) — passage/0-1/0; 294374 visible triangles
+- [section-1/view-1](after/passage-0-1-0-openair-1-view-1.png) — passage/0-1/0 **OPEN AIR**; 289074 visible triangles
+- [section-1/view-2](after/passage-0-1-0-openair-1-view-2.png) — passage/0-1/0 **OPEN AIR**; 298057 visible triangles
+- [section-1/view-3](after/passage-0-1-0-openair-1-view-3.png) — passage/0-1/0 **OPEN AIR**; 322309 visible triangles
+- [section-2/view-1](after/passage-0-1-0-openair-2-view-1.png) — passage/0-1/0 **OPEN AIR**; 283642 visible triangles
+- [section-2/view-2](after/passage-0-1-0-openair-2-view-2.png) — passage/0-1/0 **OPEN AIR**; 316771 visible triangles
+- [section-2/view-3](after/passage-0-1-0-openair-2-view-3.png) — passage/0-1/0 **OPEN AIR**; 322309 visible triangles
+- [section-3/view-1](after/passage-0-1-0-openair-3-view-1.png) — passage/0-1/0 **OPEN AIR**; 262299 visible triangles
+- [section-3/view-2](after/passage-0-1-0-openair-3-view-2.png) — passage/0-1/0 **OPEN AIR**; 310919 visible triangles; low_contrast_geometry_visible
+- [section-3/view-3](after/passage-0-1-0-openair-3-view-3.png) — passage/0-1/0 **OPEN AIR**; 363071 visible triangles
+- [section-4/view-1](after/passage-0-1-0-openair-4-view-1.png) — passage/0-1/0 **OPEN AIR**; 378007 visible triangles
+- [section-4/view-2](after/passage-0-1-0-openair-4-view-2.png) — passage/0-1/0 **OPEN AIR**; 471132 visible triangles
+- [section-4/view-3](after/passage-0-1-0-openair-4-view-3.png) — passage/0-1/0 **OPEN AIR**; 382187 visible triangles
+- [upper-mouth](after/passage-1-2-0-upper-mouth.png) — passage/1-2/0; 570417 visible triangles; low_contrast_geometry_visible
+- [lower-mouth](after/passage-1-2-0-lower-mouth.png) — passage/1-2/0; 511318 visible triangles
+- [25%](after/passage-1-2-0-station-25.png) — passage/1-2/0; 661032 visible triangles
+- [50%](after/passage-1-2-0-station-50.png) — passage/1-2/0; 602996 visible triangles
+- [75%](after/passage-1-2-0-station-75.png) — passage/1-2/0; 615520 visible triangles
+- [section-1/view-1](after/passage-1-2-0-openair-1-view-1.png) — passage/1-2/0 **OPEN AIR**; 629052 visible triangles
+- [section-1/view-2](after/passage-1-2-0-openair-1-view-2.png) — passage/1-2/0 **OPEN AIR**; 588414 visible triangles
+- [section-1/view-3](after/passage-1-2-0-openair-1-view-3.png) — passage/1-2/0 **OPEN AIR**; 616234 visible triangles
+- [upper-mouth](after/passage-2-3-0-upper-mouth.png) — passage/2-3/0; 59562 visible triangles
+- [lower-mouth](after/passage-2-3-0-lower-mouth.png) — passage/2-3/0; 294522 visible triangles; low_contrast_geometry_visible
+- [25%](after/passage-2-3-0-station-25.png) — passage/2-3/0; 55484 visible triangles; low_contrast_geometry_visible
+- [50%](after/passage-2-3-0-station-50.png) — passage/2-3/0; 457754 visible triangles
+- [75%](after/passage-2-3-0-station-75.png) — passage/2-3/0; 475706 visible triangles
+- [section-1/view-1](after/passage-2-3-0-openair-1-view-1.png) — passage/2-3/0 **OPEN AIR**; 104612 visible triangles
+- [section-1/view-2](after/passage-2-3-0-openair-1-view-2.png) — passage/2-3/0 **OPEN AIR**; 81078 visible triangles
+- [section-1/view-3](after/passage-2-3-0-openair-1-view-3.png) — passage/2-3/0 **OPEN AIR**; 94962 visible triangles; low_contrast_geometry_visible
+- [section-2/view-1](after/passage-2-3-0-openair-2-view-1.png) — passage/2-3/0 **OPEN AIR**; 475688 visible triangles
+- [section-2/view-2](after/passage-2-3-0-openair-2-view-2.png) — passage/2-3/0 **OPEN AIR**; 479134 visible triangles
+- [section-2/view-3](after/passage-2-3-0-openair-2-view-3.png) — passage/2-3/0 **OPEN AIR**; 479134 visible triangles
+- [section-3/view-1](after/passage-2-3-0-openair-3-view-1.png) — passage/2-3/0 **OPEN AIR**; 485216 visible triangles
+- [section-3/view-2](after/passage-2-3-0-openair-3-view-2.png) — passage/2-3/0 **OPEN AIR**; 479198 visible triangles
+- [section-3/view-3](after/passage-2-3-0-openair-3-view-3.png) — passage/2-3/0 **OPEN AIR**; 490436 visible triangles; low_contrast_geometry_visible
+- [section-4/view-1](after/passage-2-3-0-openair-4-view-1.png) — passage/2-3/0 **OPEN AIR**; 475298 visible triangles
+- [section-4/view-2](after/passage-2-3-0-openair-4-view-2.png) — passage/2-3/0 **OPEN AIR**; 495856 visible triangles
+- [section-4/view-3](after/passage-2-3-0-openair-4-view-3.png) — passage/2-3/0 **OPEN AIR**; 496236 visible triangles; low_contrast_geometry_visible