The compiled-analytic stratum interval now certifies a legacy tile uniform only when it stays
beyond the detail envelope (and smooth-min slack for solid) with no intersecting primitive,
edit, native field, composer candidate, passage or active transition; everything else fails
closed. The stream layer uses the proof for whole-tile skips; legacy sub-tile skips stay off
(two block proofs failed the falsifier in TN).
Falsifier: full-grid reference vs skip-enabled run, identical mesh/payload/topology hashes on
every tile of all four owner views, Clang and GCC, W1/W8/W12; 20-camera sweep per strate,
0 violations; a too-tight control changed 3/184/233/238 tiles and was caught.
Whole view (8 workers): SW 2.61->2.48, FI 7.18->5.05, TN 5.97->4.63, VS 2.70->2.40 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
FI/TN/VS (legacy compiled-analytic) skip ZERO tiles at every LOD and evaluate ~44k density
points per LOD0/1 tile; SW skips 359/512 at LOD0. The whole-tile skip requires a native
tile with a uniform verdict, and the legacy paths return Mixed/unknown (R39 notes attached).
Opt-in stage timers, sampled field timer, per-LOD classification and composer counters.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
The L70 "2x SurfaceWorld GenerateTile in Unreal" was the instrument: the trace enabled
bCaptureVoxelForgeCoreHashesForValidation, which disables the native heightfield-band
shortcut and whole-tile skips (W1 same keys: 9.00 s hash-off vs 28.46 s hash-on). With
hashing off, Unreal and headless match at W8 within 1.2-2.5% (Core 14.42 s vs 14.26 s) on
identical tile keys. LOD0 p95 tail = real cost of non-empty passage tiles near the spawn mouth
(p50/p95 18.8/105 ms); most LOD0 tiles are empty.
verify.py reads its baselines from Standalone/Tools/verify/baselines/ instead of untracked
Saved/Linux. The 11 .vfmesh control meshes (~78 MB) are not committed yet (to be stored
compressed). Fast tier green.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
- Export commandlet writes the live Core adapter's world_assembly + resolved layout pair;
the headless owner path imports that pair (manifest/layout equal, layout_chunk_width_q
validated). owner_world.json re-exported from the owner's staged assets on the Linux host.
- Adapter SW route: Core clearance/radius floors, 0.5 m support search, outward-facing
tie-break; passage mouth/mid radius floors match Core (36/24 owner voxels).
- FI in Unreal now uses the legacy (compiled analytic) generator unless a plan is imported:
the auto native-FI plan failed the passage audit on the owner plan (1843 blocked samples,
945 open-air violations). Owner decides whether native FI returns.
- In-engine spawn traces per strate (3 each): SW GenerateTile 42.7 worker-s vs 20.1 headless
for equal tile counts; LOD0 p95 45-169 ms in UE.
- Fast tier green; all three passages pass on the shared plan.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
- Capture exports now cover a 90-degree player frustum (radius 3, LOD 0-8, overlap patches);
a 17x9 ray coverage probe aborts before rendering if tiles are missing (truncated-mesh
control: 480 uncovered samples, rejected). Clear colour is sky blue, so un-meshed space can't
pass as dark rock. 16 views of the owner's four-strate plan.
- Explore commandlet: finite XYZ spawn override for traces (Source/, Linux-compiled).
- In-engine four-strate trace NOT obtained: the engine-generated plan fails passage validation
at a solid centerline voxel (~(-43,-34.5,-448.5) m) while the headless owner_world.json
plan passes. Engine and headless plans diverge.
- R36 (read-only): order TN -> VS -> FI for native migration; native FI is reachable from the
adapter but not selected in the owner run, cause unknown.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
The FI->TN wall beat (4.77 cm at 7.06 m) was the lower-mouth flare: the full mouth radius
was assigned to the penultimate control point, a one-segment radius jump. The shared route
builder now blends to the mouth radius over 12 m with a smoothstep. Wall ripple 1-2:
7.27 -> 1.62 cm. All three passages pass support/blocked/grade/ripple/naturalness; controls
detected. Fast tier 269/269 incl. the new-plan stage; W1/W8/W12 equal, Clang and GCC.
Hash pins: FI style 2 re-pinned at v213.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
- Shared composer: a passage's restored support is applied after a physically overlapping
later bore, instead of being carved away by it (0-1: 59 -> 0 unsupported).
- Capsule clearance for support restoration is tested against nearby route segments only
(1-2: 3 -> 0, 2-3: 2 -> 0 unsupported).
- L66 index switches replaced by a physical criterion (both mouths inside their strata bands,
neither a SurfaceWorld landing).
- Audit: floor zero-crossing refined by bisection (the 0.2498 m "step" was the gauge);
wall ripple keeps the authored radius, cubic detrend per 32 m window. Controls detected:
L35 ribs, beads, synthetic 0.25 m step, L36 washboard, L23 terrace.
- Result: 0-1 and 2-3 pass all audits; 1-2 fails wall ripple 7.27 cm (limit 3.5) from an
authored-radius beat at 7.06 m. Fast tier was green before the final gauge change and was
not re-run after it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
- Core caps each passage mouth reach to its own strate's usable content band (owner
DA_Strate1 exports ReachMax=5000 owner voxels, which put the SW->FI lower mouth 79.5 m
below FI, inside TN).
- Unreal adapter: SW route search keeps the old 64 m candidates first, then a bounded
single-mouth search with shorter lengths; logs candidate/rejection counts.
- Internal-chain passage floor/support tweaks (only with >2 strates): ripple 1-2 4.78->2.90 cm,
2-3 3.57->0.10 cm. Interfaces still 0/3 walkable (59/3/2 unsupported).
- verify.py: legacy tiers use Standalone/data/regression/owner_world_2strate_l61.json; new
four-strate stage reads its expectations from owner_world.json (currently red on interfaces).
- Unreal trace L14 vs HEAD, same assets (16-chunk FlatPlain): 12.20 vs 12.56 worker-s; no x2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Linux UE builds failed on the Core wrapper module: the Linux-only PerModuleInline.gen.cpp saw Standalone/Compat
shadow headers instead of the engine's, and WITH_EDITOR was redefined under -Werror. Compat headers now forward
to the engine headers with include_next for generated Linux module TUs (canonical Core TUs opt into the Compat
definitions via CoreNamespacePreamble.h); CoreMinimal.h undefines WITH_EDITOR first. VoxelForgeCore.Build.cs
enables that forwarding and AVX2 on Linux only; VoxelForge.Build.cs disables Unity on Linux only. Portability:
std::int64_t consistently in VoxelContentManager.cpp; grouped &&/|| in VoxelRedefinedDensityField.cpp. Linux-only
trace fields split Core generateTile / mesh copy / RMC stream build. Standalone Clang and GCC whole-view hashes
are unchanged vs 3202193. The Unreal spawn trace is blocked by the owner's new assets (SW route sweep fails).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Per-round bisect of a176d07..3202193 (56 builds; timings inflated by concurrent UE builds) plus a clean
quiet-machine endpoint re-measure: L14 12.62 vs L61 14.65 worker-s (+16%), LOD0 3.99 vs 4.99 ms/tile. The
owner's Windows x2 is therefore mostly on the Unreal side or not like-for-like. Four-strate U2 plan: all
strates generate and are deterministic; passage 0-1 lower mouth is out of FI's band, 0/3 interfaces pass,
FI/TN/VS use legacy evaluators. R33: ranked Unreal-side suspects and Windows experiments.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Owner re-saved strates 2026-10-26: DA_Strate1 SurfaceWorld, 2 FloatingIslands,
3 TunnelNetwork, 4 VerticalShafts (default params). BuildHost copies and the
enforced SHA-256 tables updated (old U1 set kept in the report); the export
commandlet handles the four-strate plan; Standalone/data/owner_world.json
regenerated (SHA-256 F42F96AD...). One normal BuildHost gate passed (64/64
tiles) but its top-strate window produced 0 triangles, and a missing
NewDataAsset reference warns -- both to check. No measurements this round.
Report: Standalone/Research/WorldAssemblerM3/Unreal2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Windows MSVC catch-up after abyss L15-L61 (no Unreal there): 64-bit clipmap
shift (C4334), qualified VoxelForgeCore::WorldAssemblySubvoxelsPerVoxel, UHT
metadata and duplicate-signature fixes. Optional CoreGeometry section on
UVoxelStrateDefinition (native caves, lattice, profiled heightfields, named
stamps, rooms/tunnels; off/empty by default), resolved after passage planning.
VoxelForgeAssetVersion 2 converts strate height/transition/gap to metres on
load (exact, 0 m change on the owner's 5 assets; packages never saved).
Gates pass (SurfaceWorld, FloatingIslands, Core-off, M2, W1/W8/W16).
KNOWN REGRESSION vs L14 (2026-10-20, same command, same PC): full set 3.9 ->
5.37 s, worker-seconds 15.8 -> 30, LOD0 1.11 -> 1.69 s, ring 0.50 -> 0.80 s;
not yet attributed (Core L15-L61 vs Unreal side).
Report: Standalone/Research/WorldAssemblerM3/Unreal1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New verify.py sanitize tier (Clang address,undefined,float-cast-overflow; GCC address,undefined;
-fno-sanitize-recover) runs 19 cases per compiler at LOD0-8 across every generator, passages and the
crossing: 0 runtime reports; a planted signed-overflow control is caught by both. A static scan fixed
INT_MIN negation in floor division, unbounded signed LOD shifts and coordinate-times-extent overflow in
VoxelClipmapDesiredTiles.h and VoxelTypes.h (plugin and standalone), and VoxelWorld.cpp, VoxelContentManager.cpp
and VoxelDensityVolume.cpp (plugin only, uncompiled here). Supported-input output unchanged (version 211).
fast/medium/full PASS. Research R31 (self-checked): far-LOD 4-8 cut plan.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
VoxelHash::Cell/Pair, legacy Cellular3D and the shared Core cellular hash added 0x7FFFFFFF to a signed
int32 before casting to uint32; overflow is undefined, so Clang and GCC placed cave passage 1 differently
(first differing sample at LOD1 tile (0,64,-512)). The bias is now done in unsigned arithmetic, in both the
standalone copy and the Unreal plugin header (Source/VoxelForge/Public/VoxelCaveMorphology.h). Native
caves, motifs and lattice now stream LOD0-8 with skirts and cross-level transitions. New verify.py
--speed-matrix runs every generator on one asserted-identical view (5 interleaved runs): motifs 1.33,
profiles 1.30-1.32, caves 1.41, natural SW 1.67, lattice 1.70, FI 4.93 s whole view. fast/medium/full PASS,
Clang/GCC peer hashes equal. Research R29 (open-edge ledge) and R30 (compiler-divergence suspects).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Core.Lattice.PrimitiveField v1, opt-in per strate (legacy Maze/VerticalShaft stay legacy): resolves the real
arrival and departure mouths first, plans a mandatory spine, then maze cells and two dead ends (9 rooms,
9 edges, stable IDs). The spine descends 20 m through four helix turns (6 m radius, 3 m clear width, 5 m
pitch, 2.6 m clearance; max grade 10.0 deg); each feature is one filleted void (0.57 m measured). Final-field
reachability passes at 883 stations; an orphaned shaft control is rejected; exact intervals (0 misses) and
a whole-surface audit with 0 flags. Whole view 1.69 s (above the 1 s target). fast/medium/full PASS.
Research R27 (self-checked): static stream-to-pool water design.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Sampled cycle attribution of FI LOD0 field work: grid fill 83%, boundary 13%, skirts 4%; island roots,
columns and helix dominate, the column cache path is real but secondary (no thrashing). FI-only Z-inner
traversal at LOD0 (same lattice samples, exact) cuts FI whole view 5.35 -> 5.03 s and moving p50 by 12%;
global Z-inner and <=4-island culling were measured and rejected. The meshing band now includes the planned
passage support column's conservative bounds, closing a real gap where the arrival shaft meets the floor
(WorldAssemblyEvaluatorVersion 208 -> 209). fast/medium/full PASS, Clang and GCC.
Research R26 (self-checked): native lattice (maze/shaft) generator design.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Motif vocabulary v3: version-1 records keyed by (seed, strate, stable ID) with route keepouts, band
clearance and pairwise spacing. A rounded capped-cylinder monolith (r 1.4 m, 24 m tall) and a rounded-box
plinth, smooth-unioned into a flatField floor with a 0.55 m geometric rounding (measured fillets 0.73-1.17 m).
Exact radial/box interval bounds, a BVH, isTileShapeFree aware (0 brute-force misses, false-proof controls
rejected). Final-field route passes; whole-surface audit covers foot, cap and platform edges. Cold whole
view 0.524 s. Owner world unchanged. fast/medium/full PASS (Clang and GCC).
Research R25 (self-checked): FloatingIslands LOD0 column cache conflicts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Opt-in --instrument-tile-drain records per-tile queue/dispatch/finish timelines, stage spans and counters
(proof skips, X8 lanes, GradAt/midpoint memo, skirts, FI cache stats); <2% when off. Measured: SW 13.3 and
FI 50.2 worker-s across LOD0-8, tails are work-saturated. FI sign-edge repair now computes in parallel and
publishes in sorted serial order (exact output): sign-edge 535 -> 47 ms, FI first tile 1.28 -> 0.85 s, near
ring 1.46 -> 1.04 s, whole view 6.09 -> 5.65 s. Proof block size, lane batching, SW midpoint memo and helper
cap 12 measured and left off. fast/medium/full PASS, Clang and GCC. Research R24: slab + motif generator.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Core.Caves.PrimitiveField v1, opt-in per strate (legacy TunnelNetwork assets stay legacy): solid host rock
and one smooth union of two elliptical spine tubes and an ellipsoid room, each intersected with its rising
floor by cubic smooth-max (W 2.4 m, measured fillet 0.70 m, no trailing floor cap). Bound in the L40 field
table and L42 capabilities, BVH feature candidates, isTileShapeFree aware, exact interval boxes (0 misses).
Capsule flood fill arrival -> departure passes (headroom >= 0.3 m, grade <= 15 deg); the generalized
whole-surface audit passes with its controls failing. Cold whole view 0.727 s, near ring 0.19 s. Graph
ordering/serialization changes are scoped to native caves (the flagged L47 demo is unchanged); verify
tiers write captures under --out only. fast/medium/full PASS, Clang and GCC.
Research R23 (deepened): per-tile LOD0 kernel speed plan.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
WorldAssemblySurfaceProfileV1 on the native SurfaceWorld recipe and plan; version 0 'natural' is omitted
from serialized identity, so the owner world is unchanged. periodicHills (seeded cosine hills, plan-time
grade <= 15 deg), mirroredRelief (cap mirrors the floor by lambda, clearance checked) and flatField share
one profile function for points, lattice classification and exact cosine-extrema intervals (0 interval
misses in brute-force audits). Style-override fixtures with planned passages. Cold whole view about 1.33 s
for each profile vs 1.74 s natural SW (Clang). Owner hashes identical to L52; fast/medium/full PASS.
Research R22: SurfaceWorld critical path to about 1 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
The shared marching-cubes mesher applies its 8-cell block proofs at step 2 for standalone shape contexts
and skips density-grid fills outside certified-uniform blocks (plus a one-sample normal halo); mixed and
unknown blocks still sample exactly. FI certified-air boxes 1,959 -> 14,119; FI LOD1 samples -63%. FI whole
view 7.08 -> 5.61 s (-21%), moving p50 about -33%; SW unchanged. Geometry hashes identical (no version
bump); proof audit 36.9 M samples, 0 violations, Clang and GCC. Source stamp now covers the standalone
mesher copy. Research R21 (deepened): native Core cave generator design.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Measured first: 45% of FI tiles are all-air, yet FloatingIslandFieldV1 returned unknown for any box near
an island AABB (25,536 unknown boxes, 0 tile skips). Conservative ranges for island bodies, root lobes,
buttresses, connectors, surface detail and smooth-min budgets now let the shared box classifier prove FI
air/solid (216 tile skips; 28.6 M audited samples, 0 violations). The X8 path reuses its FI preflight
samples. LOD7-8 far-shell and column work is skipped outside per-tile island/connector influence bounds
(farHorizonVersion 4 -> 5). FI 16.74 -> 7.36 s (-56%), FI moving entering-tile p50 halved; SW unchanged.
Peer determinism (Clang/GCC W1/W8/W12, random cameras) PASS; fast/medium/full PASS.
Research R20: world-type catalogue and dreamcore/liminal concepts (revised after a deepening pass).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Versioned capsule vein records (stable ID, resource ID, endpoints, radius, seeded irregularity) resolved
into the snapshot only when requested, placed beside the route and away from passages, landings and
landmarks (forced-passage control rejected). The bound field table gains materialAt (host + two resource
IDs + Q16 weights) behind a per-tile BVH; mesh export v2 carries per-vertex materials; a measurement-only
yield computes removed resource volume (0 on already-air space). With veins on, every density, mesh,
payload and topology hash is unchanged; demo off: owner hashes identical to L47. fast/medium/full PASS.
Research R18 (cavern-graph planner) and R19 (why FloatingIslands costs ~10x SurfaceWorld).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
AuditWholeSurfaceL48.py scans every LOD0 triangle in the route corridor + 3 m and the chamber box + 3 m:
welded-edge normal variation, high-angle edge density, max jumps and sliver area in 1 m cells, clustered
and located, with thresholds from native patches in the same export, plus a detrended full-width floor
profile. L45 ramp, L46 entrance and L47 wall-foot meshes fail it; plain native terrain passes. Red-triangle
SVG maps included. Not wired into the tiers yet; the attempted Core fix is parked on wip/l48-surface-audit.
Research R15: mesh artifact metrics.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
The demo tunnel's hard floor cap used a vertical Z - floorZ on a sloped Bezier floor, which is not a signed
distance, so marching cubes produced a staircase through the first 7 m (40-51 mm ripple, 82 mm steps). The
cap now uses the grade-normalised signed distance with a floor-adjacent bore re-application fading over the
entrance; mesh tile Z bounds include cavern bounds. New three-rail entrance audit: L46 fails, L47 passes.
verify: the cost baseline exports only build inputs into Saved/Linux/<out>/baseline-<commit> and removes it;
/tmp unchanged after a full tier. Demo off: owner hashes identical to L46. fast/medium/full PASS.
Known: sawtooth at the tunnel wall foot (next round). Research R14: Windows/MSVC readiness review.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
The flagged SW demo's ramp (57 carves + 107 support pillars, visibly ribbed) becomes one cubic-Bezier
sweep: closest point by 17-point seed plus 5 Newton steps, a cubic floor profile, a rounded 2.6 m bore,
and a hard floor cap applied after unions and relief that only adds support where the field is air.
Optional resolved cavern-graph records (one room, one edge) enter canonical identity only when present;
their bounds extend World::isTileShapeFree. Reachability passes (L43 control fails), headroom 2.8 m,
floor ripple 1.8 mm (L45 ramp control fails at 17.9 mm), shape limits met; demo off: owner hashes
identical to L45. fast/medium/full PASS. Research R13 (read-only): continuous SDF sweeps along curves.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Stamp vocabulary v2; the flagged SW demo opens through a side-facing portal and a bent ramp (max 14 deg,
two landings) into a 17 x 15 m chamber with a rock dais. New reachability audit: a capsule surface search
from walkable ground near the arrival mouth to the chamber floor (66 samples, 0 blocked/unsupported); the
buried L43 layout fails it. Demo off: owner hashes identical to L44. fast/medium/full PASS (Clang+GCC).
Research R12 (read-only): composite cavern-graph generator design and slices.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
L27 added passage support columns, but World::isTileShapeFree only bounded the route and bridge, so tiles
touched by a column could be approved for early native-only generation; depending on scheduling they used
that incomplete field or the compiled one (six LOD0 tiles differed between W1 and W12 at an SW camera near
the arrival mouth). The shape-free proof now includes a conservative envelope for every possible column.
The compiled world is unchanged (the old W1 variant), so no version bump. verify tiers gain a seeded
(SplitMix64 0x4c3434) random-camera W1/W8/W12 sweep across both recipes and several clip settings; it fails
before the fix and passes after. fast/medium/full PASS; cost within noise. Research R11: suspect review.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Stamp vocabulary v1 (carve ellipsoid/arch, fill pillar/dais) with smooth CSG and seeded relief; the
director resolves records (stable ID, strate, micrometre position, conservative bounds) from
hash(seed, strate, ID) with up to 512 candidates and rejects any that touch a passage mouth, walk corridor,
content face or ceiling/bedrock band; a stamp BVH skips tiles outside all bounds. No requests: canonical
JSON and owner hashes unchanged. --landmark-chamber-demo adds a SW chamber (buried; see review).
fast/medium/full PASS on clean builds. Research R10 (read-only): ore/resource channel design.
Review also records a pre-existing W1/W12 determinism mismatch at a non-spawn camera (L37+, not S14).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
ResolvedLayoutSnapshotV3::finalize maps each strate's generator and interface once into capability
records (FI gradient air-gap, surface/FI landing support, passage-anchor columns, SW-to-TN floor);
ApplyResolvedStrateBoundaryBand and passage composition read those instead of generator kinds. Derived
records stay out of canonical JSON, so layout identity is unchanged. Owner hashes identical to L41;
fast/medium/full PASS on clean Clang and GCC builds; cost within noise.
Research R9 (read-only): player-visible roadmap refresh.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
UVoxelStrateManager::Initialize resolves the layout width to integer micrometres before passages or
scheduling; each slot stores physical [bottom, top) bounds, transition mode/width and passage settings;
chunk indices remain a floor-based compatibility projection. The Core adapter derives interiors and L29
bands from those resolved values. verify: full-tier cost baseline is HEAD (or --baseline), tier timeout
7200 s with 1800 s per process; the source stamp now includes the plugin Source tree compiled on Linux
(it previously missed VoxelStrateManager.cpp and friends). Owner hashes identical to L37; fast, medium
and full PASS on a clean rebuild. Unreal manager/adapter edits are uncompiled on Windows.
Research R8: inventory of unit-bearing UPROPERTYs and a custom-version asset migration plan.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
SurfaceWorld, FloatingIslands and legacy evaluators sit behind one immutable densityAt adapter held in
a variant (no virtual call per sample); BindResolvedStratumFields builds a sorted strate table at world
setup, so DensityAt, native SW/FI sampling and ClassifyBoxDetailed no longer ask the strate manager per
sample. No arithmetic change: owner hashes identical to L37/L39 at W1/W12. Cost vs L39: FI -0.66%,
SW -1.31% (noise). verify fast/medium/full PASS (full re-run alone: 407 gates).
Research R7: disk cache design for UE5/Windows/multiplayer and a zlib probe (about 1/3 size, 0.1 ms/tile).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
BuildLinux.sh hashes a sorted manifest of compiled inputs into VOXELFORGE_SOURCE_STAMP, printed first by
every executable; Standalone/Tools/verify/verify.py checks it before and after each tier and records
binary/input hashes and the resolved recipe, style map and layout. Fast (163 s) runs on the canonical owner
world only and asserts FI Style 2 / SW Style 1; medium (8 min) and full (26 min) add the style matrix,
GCC, determinism and cost. Controls: stale source, mislabelled style-0 owner, all-black capture, L30/L31,
L35/L36 and the L38 step are all rejected. No geometry change; owner hashes identical to L37.
Research R6: leak table, generator contract and a 10-step metre-native migration plan.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Inventory of every audit gate with controls and wall times; a fast tier (~3-4 min, real owner styles)
per trial, medium (~8-10 min) per candidate, full (~32-34 min) once. A compiled-in source manifest stamp
that every runner checks, so evidence provably comes from the committed tree. Lists duplicate runners.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Offline replay of the L37 and L38 tube fields on the FI Style 2 helix shows no 7.8 m ripple from the
tube distance itself (< 0.3 cm). The real signal is a step: the upper coil's 1 m fallback floor strip,
composed independently, puts a support root 0.44 m from the lower bore's centreline at head height;
the 7.8 m "period" is a sinusoid fitted to that step. Proposes pairing the floor cap with its bore.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Read-only study for L38: a synthetic FI Style 2 helix (3 m radius, 3.31 m pitch) shows Euclidean
closest-centerline ownership is wrong when bore radius > pitch/2; a floor-near (+-0.125 m, within bore
radius) nearest-floor-point owner gives 2.9 cm p-p on helix and style-like paths. The L38 0.94 m ripple
period equals the 20-segment chord. L38's continuation applied it: floor ripple 0; wall ripple open.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
ShapeWorldInternal.h kept an extra ')' after a diagnostic edit was restored post-build, so the L37
commit did not compile; the fixed build reproduces L37's verified owner-world hashes exactly.
Unreal (uncompiled here): SetHardTransitionFallbackForAssembly is const over a mutable runtime flag
(the adapter calls it through a const manager); the legacy fallback clears the flag; VoxelStrateManager
includes <cmath> for std::llround. Adds research R2 (read-only Unreal compile/behaviour review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
The passage mouth fade multiplied the signed carve distance by zero, which turned native rock into an
open surface (the reason earlier rounds stamped walk pads at mouths); it now blends between the
uncarved field and the full signed carve. The FI style-0 route followed a lifted grade envelope over
the crater (the "bridge"); it now runs on the rim and follows the native surface. New built-shape and
crease audits with thresholds frozen before the fix (L36 and box controls fail). 24/24 passage rows,
30/30 determinism, 16/16 invariants, 18/18 interface/ring. Unreal: one recipe setter (uncompiled).
Research R1 (read-only, web): ranked proposals for natural walkable mouths, floor-limited carve first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
The walk tread was built per segment as a one-sided density clamp, so its crossing with the native
field moved at segment joins (L35: 2.25-2.48 cm washboard). Treads now overlap up to 2 m as bounded
signed-distance slabs (0.06 cm residual). The FI mouth's crater detail and tread met without a smooth
join; v23 filters the local field, drops thin isolated solids and smooth-unions the tread with the
terrain. New floor ripple and mouth fragment/crease audits with L35 as failing controls. 24/24
passage rows, 30/30 determinism, 16/16 invariants, LOD0-8 rings; cost within noise.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
The shared passage carve took the minimum of piecewise-linear capsules (scallops at joins) and added
wall relief on regular 8 m / 4 m octaves, which read as evenly spaced ribs. Interior sweep joins are
now Catmull-Rom smoothed and the relief uses irregular rotated 43/29/19 m scales; the natural wobble
remains. New axial ripple audit (0.125 m radius samples, 0.5-8 m periods): beaded control and L34
geometry fail, all 24 passage rows pass. FI mouth tread apron tapered. 30/30 determinism, 16/16
invariants, ring audits LOD0-8 pass; cost within noise.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Adds a density-field floor/wall profile audit along passages near mouths (0.125 m spacing; L23
terraced negative control detected) and a reusable 12-view passage capture set (80 deg FOV, eye
height, LOD0, skirts off, capped clip planes). The old blurry mouth shot was camera framing; the
L33 entry ledges were tile skirts. Owner world hashes unchanged; 24/24 passage rows, 30/30
determinism, 16/16 invariants; cost within noise.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Route planning for an upper SurfaceWorld failed on long candidate chords over steep relief; chords
are now capped to the arrival/departure radii on an 8 m centre grid with an origin-proximity cost
(21.5 m walk, about 4 deg). The departure passage starts at the SW ground and tunnels down through rock.
FI -> SW -> TN fixture: all 4 styles pass every passage audit on both compilers. FloorProfileBlend
now checks endpoint height/tangent seams (step and tangent-step negative controls detected). Owner
world hashes unchanged; 30/30 determinism; cost within noise.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Band faces now use signed distance to the displaced solid intervals, continuous across bands, so
MC interpolates the interface instead of terracing at density levels. The Gradient amplitude is
restored to width/8 (about 4 m peak-to-peak at 16 m). Ring/terrace audit at LOD0-8 on the FI floor and SW
ceiling passes; negative controls: L30 (rings) and L31 (flat) both fail. Hole/interface audits,
16 passage rows and 24/24 determinism pass. Evaluator 20. Cost vs L31 SW +1.9%, FI +1.5%.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
On the real FloatingIslands recipe, column caps sit inside anchor-island rock and native bedrock;
the old check sampled the route foot. The audit now samples each planned cap against the native
field (negative control: a cap moved into air is detected). The audit runner records requested vs
actual recipe and exits on mismatch. All 16 passage rows (4 styles x 2 recipes x 2 compilers) pass.
The readable shader uses a smooth tint. Interim: FI Gradient displacement capped at width/256
(flat floor, no contour rings); restoring real undulation is L32. Evaluator 19.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
The Gradient interface left native solid density in the air-side gap and snapped the face to
voxel levels (concentric ring steps on the FI floor). The owner Gradient path now writes signed
distance both sides; match error RMS 0.36 -> 0.11 m. Band and fade-tile culling cut transition
cost: whole view vs L24 SW +2.7%, FI +0.2%. Evaluator 16. Deterministic across compilers and
workers; interface and passage audits pass. Found: on the real FloatingIslands recipe, the
rock-column connected-ends audit fails (all styles); earlier matrices ran SurfaceWorld twice.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
EVoxelStrateTransition from the upper strate shapes the bedrock(N)/ceiling(N+1) interface in the
shared composition layer: Hard flat; Gradient = smooth 2D hash surface (16 m owner width, about 4 m
measured undulation); Interleaved = 3D domain-warped fingered overlap within the width. Each band
keeps >= 2 Core voxels solid; faces return to the seam near passage mouths. Interface audit over
512 m: 0 hole columns for all modes; wider-than-declared and too-thin controls rejected. All passage,
route, proof and validator audits pass; deterministic Clang = GCC. Evaluator 15. Cost +4% SW,
+1.3% FI vs L28. Open: ring-like artifact on the FI floor capture; transition look unverified.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
GenerateMesh scopes passage queries to the tile's sample bounds with a feature AABB BVH plus
per-passage segment BVHs; L9 proof boxes use a passage-bound BVH. Whole view vs L24: FI ~0%, SW
+3.7% (reviewer interleaved). All hashes equal L27; all passage/route/proof audits pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Unsupported passage runs > 12.5 m become a solid relief-bearing column (ceiling/island to
ground, flared base) with the helix inside, for every recipe; FI anchor uses the same path.
All 4 styles x both recipes: 0/0/0, 0 open-air, naturalness pass; no-long-bridge audit (L26
fails it). Deterministic, Clang = GCC. Look approved; cost +35% SW / +19% FI blocks merge (L28).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
All 4 styles x both recipes pass capsule, open-air and naturalness audits, and full-view hashes are
deterministic. But the SW passage crosses 355 m of open SurfaceWorld sky as a ~2 km fragmented rock
ramp, which is not natural. L27 replaces long bridges with a shared ceiling-to-ground rock column.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Fixes the L24 ribbon (owner request). NOT merged: the open-air support gate leaves ~4,000 unsupported
samples on every SurfaceWorld-recipe passage style (TN -> SW), a walkability regression. L26 fixes that
on this branch. See Linux25/REPORT.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Core plans inter-strate passages from each upper strate's FStratePassageConfig for every
generator: Straight / Worm (the new natural default: meander, variable radius, rough walls) /
Spiral / Cascading. A naturalness audit (no >8 m flat runs, >=0.20 m wall relief) fails on L23's
smooth tube. FloatingIslands selects Spiral, with a deterministic unique anchor island (ordinal 2)
joined to the bedrock floor by a rock root. The SW arrival no longer stamps flat shelves.
Capsule audits 0 blocked / 0 unsupported / 0 grade failures, all styles. Full-view hashes
deterministic W1/W8/W12, Clang = GCC (reviewer re-run). Evaluator 10, FI geometry 5.
KNOWN: the FI helix renders as a spiral ribbon in open air around the pillar instead of a tunnel
enclosed in rock; fix next round. Unreal edits uncompiled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
After L22 the FI -> SW passage no longer connected the strates (legacy owner-voxel coordinates,
lower end 23 m above SW), and Core never carved passages through the bands. Now the director
plans passages from the resolved layout (PlanResolvedLayoutPassagesV1; deterministic hash of
seed/strates/connection; placement disk from PassageConfig, 8-24 m around (0,0)); shared
composition applies generator field -> bands -> bounded passage carve and support shelf, opening
the bands only along the tunnel. Mouths bind to the strate routes. Capsule audit: 0 blocked /
0 unsupported (325,962 samples FI, 284,319 TunnelNetwork), negative controls detected; all 8
generator kinds exercised. SW away from the passage is bit-identical. Layout V3, evaluator 9.
Open: look not yet natural (stepped tube in FI void, terraced SW arrival); SW-as-upper route
planning fails; Unreal adapter uncompiled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
StrateLayoutChunkSizeMeters (world setting, default 8 m; owner decision) is resolved once in
the director/adapter and standalone loader; every generator reads the resolved width (no
CHUNK_SIZE / Core-tile inference). Every strate gets solid ceiling and bedrock bands from its
BoundarySealThickness in the shared composition path (diggable by default); faces come from the
shared boundary path. Owner world: FloatingIslands +8..-120 m (128 m, now with a bedrock floor),
SurfaceWorld -120..-632 m (unchanged; full-detail and exact-query hashes equal L21); the 60 m gap is
gone. The strict layout validator passes on the real world, with 0 endpoint dispatch
disagreements. Spawn/benchmark camera derived from route.front. Ledges under the islands gone
(reviewer capture). Evaluator 7, recipe schema 2, layout snapshot 2, FI geometry 4.
Open: SW cap profiles exceed the ceiling band (owner call); transitions still Hard; Unreal
setting/manager/adapter edits uncompiled here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Core ResolvedLayoutSnapshotV1 (canonical JSON, stable identity): per strate stable ID, generator,
recipe, interior bounds in metres, ceiling/bedrock bands as Core really applies them, transition
mode/width, dispatch provenance. Validator: compat mode reports today's deviations; strict mode
enforces ceiling -> content -> bedrock -> next ceiling (owner rule), two separate bands, nominal
contiguity, interpenetration only within the declared EVoxelStrateTransition width; 7 negative
controls rejected. Findings: SW built at -120..-632 m although the params say -60 m; a 60 m unowned
gap filled by analytic FI fallback; no active bedrock bands; transitions ignored by Core.
All L19 hashes unchanged (28 + 14 crossing runs). The Unreal adapter exports resolved_layout
(uncompiled here).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Verdict: direction right; preserve today's bounds as compatibility data but reject overlaps in new
layouts; approve the LOD2 boundary fix with a precise affected region; one director-owned vertical
partition (strate interior / shared bedrock band / explicit void); bedrock = solid volume + explicit
LOD-stable surface; first step = versioned resolved-layout snapshot + validator, no output change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Standalone/DESIGN_STRATA_UNITS_AND_LAYERS.md: maps today's code to the three layers, ranks 9
violations (top: no single physical stratum-bounds/interface record; SurfaceWorld 8 m vs
FloatingIslands 4 m per chunk), defines the director-resolved generator request/result contract,
moves seals/caps/voids/interfaces into shared composition, and proposes an explicit LOD-stable
interface surface for the L19 ledges. 6-step migration plan with "worlds don't move" checks;
owner decisions listed. No code or world change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
The shared boundary-cap path now emits a cap wherever the field is solid just below a strate's
configured top (it previously required a solid-to-air crossing there) and checks a 5x5 perimeter
grid before suppressing a fallback cap. Strict audit: 0 misses / 0 edge gaps, LOD0-8.
farHorizonVersion 3 -> 4. SurfaceWorld LOD0-2, full detail and exact queries unchanged; 28/28
matrix plus 14/14 crossing; route parity and proofs pass.
Ledges: the lower stratum's terrain at the FI/SW interface, seen through the FI void, sampled at
0.5/1/2 m at LOD1-3 (up to 1.5 m / 7.7 m ring steps). Fix deferred to the units/layers round.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
FI geometry v3 drops the all-XY lower boundary seal that created false floors (3.5/8 m); the
ceiling comes from the shared VF_ApplyUpperBoundarySeal and AppendStratumBoundaryCeilings path
for every generator, at each strate's own configured upper boundary (owner: every strate has a
ceiling). Far-LOD ceiling continuity: 0 gaps at LOD4-8. SurfaceWorld bit-identical to L14
(14 configs + 24 m crossing); FI deterministic (28-row matrix); route parity and proofs pass.
Unreal: FI geometryVersion 2->3 and the shared upper-seal helper extracted (UBT compile needed on
the owner PC). Open: small ceiling holes; lower-stratum LOD ring steps seen through the FI void.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
The capture tool gains --shade=readable (directional light, soft ambient, height tint) and an
optional --tile-edges overlay; display only. The L17 Core fix is NOT merged: its owner/Core chunk
scale change (8 m -> 4 m) moves every stratum and leaves the SurfaceWorld spawn camera over a void
(reviewer before/after renders). Kept on wip/l17-fi-scale. The FloatingIslands full-XY
boundarySeal cap is the likely real cause of the false floors and needs a separate minimal fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Lossless zlib halves the size; loaded tiles are exact; corrupt/stale files are rejected and
regenerated; zero cost when off. Code is on wip/l16-tile-cache until it is made Windows-safe.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Refine x2 at LOD4-8 would save 13.6% of whole view but smooths far crests and opens holes
(gate and captures agree); connector shortcuts open gaps. Default far output is unchanged; the
code is on side branch wip/l15-far-budget.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Region pages fell back to the legacy GetDensityAt evaluator while the base shape used the Core
manifest field, so a tile's density (and sometimes mesh) depended on streaming history. Pages
are now bound to their world's base FStandaloneShapeWorld evaluator before publication; density,
lattice, detail, classification and solid queries all delegate to it. WorldAssemblyEvaluatorVersion
5 -> 6. Route-parity audit: fails on L12 (5/5); after the fix 0 mismatches over 87 tile pairs
(base/page, base/native-only, edited, two histories), 56 exact-query pairs, 2.3M page samples,
4 seeds/recipes. Page-tile proof audit: 1,274 blocks, 0 violations. The spawn and island worlds
are unchanged; crossing output changed as intended. Reviewer interleaved spawn: +1.1% (noise).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Analytic region pages now retain the immutable shared manifest, so LOD1 page tiles get the L9
uniform-block proofs (71.7 -> 5.6 ms per tile). Reviewer W12 no-hash: 24 m crossing 0.898 ->
0.166 s; Luna: 1 km 5.89 -> 1.25 s. Spawn unchanged; all L1/FloatingIslands hashes and the 24 m
crossing hash unchanged (W1/W12).
KNOWN BUG (pre-existing, measured here): the same tile key yields different density from the base
shape (manifest field) than from a region page (legacy field), sometimes different mesh too, so
output depends on streaming history. To be fixed in L13.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Crash: crossing tiles were coalesced into one huge rectangle, and BuildSpatialBins tried to
resize to ~4.3e17 bytes (std::bad_alloc; the L7 32 m failure had the same cause). Analytic
regions now skip unread dense bins and compile in world-aligned 128-voxel pages, published
together; out-of-range keys error before allocation. 24 m, 32 m and 1 km crossings pass
(1 km: 72 pages, peak RSS 700 MiB).
Spawn: World::isTileShapeFree proves LOD0 SurfaceWorld tiles that never read the cave shape;
those start before shape compile, and all others wait. Reviewer W12 no-hash: near ring 0.337 ->
0.159 s, whole view +2.1%. All SurfaceWorld/FloatingIslands hashes unchanged (W1/W8/W12, both
compilers, forced-overlap run). Unreal VoxelWorld.cpp adapter changed but NOT built here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
First tile starts ~300 ms after cold start: plan 51 ms (serial), shape compile 223 ms (caves,
signs, detail band, far parents). The first SurfaceWorld tiles use the native heightfield and do
not read that shape, but generateTile requires the full base shape. 8/16 m crossings repeat no
compile (61 ms for 288 tiles); a 24 m crossing died with std::bad_alloc under the 6 GiB
address-space cap (cause unknown). Telemetry not committed (+1% when off).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Step-2 (LOD1) tiles of native SurfaceWorld now prove 8-cell blocks uniform from exact heightfield
column extrema over the padded block (cells + normal halo) and skip their density fill; features,
edits, mixed strata and Unknown verdicts still fill. LOD1 samples 42,875 -> 2,063 per tile.
Audit: 80,502 skipped blocks (3 seeds) + 9,216 island blocks force-filled, 0 violations. All
SurfaceWorld and FloatingIslands hashes unchanged at W1/W8/W12, Clang off and GCC default.
Reviewer interleaved W12, no hashing: whole view 1.439 -> 1.409 s. L8 phase probes removed;
--profile-lod-work and --audit-uniform-block-proofs are opt-in.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
W12 L1 view: tile work 17.3 worker-s finishes 4 ms from ideal 12-way scaling; the remaining
~0.28 s is startup (shape compile 215 ms). Largest lever: LOD1 density-grid fill (3.35 worker-s;
proof blocks are gated to steps 1 and 8). LOD6's 4x columns = far connector-shell repair (1.38
worker-s). No code committed: the diagnostic instrumentation stays uncommitted for L9.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
SharedColumnCache.h: a world-owned, bounded 4-way second-tier cache of exact column values
(heightfield and floating islands), keyed by recipe/plan identity plus content hash, XY and island
id. Lock-free seqlock reads; set-local spin lock on misses. The FLazy thread-local tier stays
first. 83.6% of repeated column work was cross-worker; formula evaluations -60.9%.
W12 L1 view: Luna 2.165 -> 1.754 s; reviewer interleaved 2.059 -> 1.722 s. 8 m crossing new tiles
82 -> 52 ms. +66 MiB peak RSS. All hashes (whole view, per-LOD, exact queries, islands seed 0)
identical at W1/W8/W12, Clang off and GCC default. MSVC/UBT not compiled here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX
Linux offscreen capture (EGL/GLES pbuffer; llvmpipe in the sandbox) and an opt-in screen-space gate
for LOD>=3 (normal distance projected at each LOD's closest display distance, 1 px at 1440p =
0.61 mrad). The gate rejects the S13 octave-drop and stride-2 controls, fault injections, and a
vertical-vs-screen control. Finding: today's far meshes differ from the field by up to 92 px, and
LOD swaps snap mesa rims by tens of px (p50 0 px), a silhouette effect of plain LOD switching. The
octave cutoff was dropped: nothing qualified, and octaves are a small share of far time. Off-path
cost: reviewer interleaved W12 within noise vs L4 (+0.6% median). Hashes = L4.
Largest far lever measured: column work outside noise, 34% of far worker time (duplicate XY across tiles).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ED33VnqepD4YY6gDAAA832
VoxelForgeCoreThreadLocal::FLazy<T> (new public header ThreadLocalStorage.h) makes every large
thread_local cache a pointer that heap-allocates on that thread's first use (still worker-private,
lock-free). Static TLS 17.33 MiB -> 43 KiB (initialised image 5.8 MiB -> 21 KiB); worker start
5.2 -> 0.12 ms; idle worker RSS 17.7 MiB -> 52 KiB. Linux workers back to 8 MiB stacks.
World unchanged (L1 hashes Clang off/GCC default, W1/W12; FloatingIslands seed 0).
Reviewer interleaved W12: 2.256 -> 2.091 s. MSVC/UBT not compiled here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ED33VnqepD4YY6gDAAA832
CoreNamespacePreamble.h applies per-compiler contraction-off pragmas (Clang, MSVC, GCC; unknown
compiler = #error) and a fused-multiply-add self-test that World::compile runs before building.
Clang -ffp-contract=on and GCC's default (fast) now reproduce the L1/Windows hashes (they diverged
in L2); Clang -ffp-contract=fast ignores the pragma and is refused at startup (negative control).
Seed sweep with every cube on real terrain: 48 seeds x 2 recipes x 3 builds x W1/W12, 576 runs,
0 mismatches, 0 empty meshes. Cost +0.5% (2.326 -> 2.337 s, interleaved). World unchanged.
MSVC/UBT not compiled here: Windows to-do list in Linux3/REPORT.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Clang -ffp-contract=on/fast and GCC's default (fast) all change the density hash on the L1 view;
contraction off reproduces L1 (= Windows S13) at W1/W12. Seed sweep: 48 FloatingIslands seeds and
14 effective SurfaceWorld seeds (34 of 48 fell on a seed-independent cube) match across Clang/GCC
and W1/W12. Standalone gains --seed/--recipe/--render-distance-chunks (viewer + headless) and a
Windows replay script. Linux standalone workers get 32 MiB pthread stacks (TLS is 17.3 MiB/thread),
so no external stack limit is needed. No Core or world-output change; MSVC compile unverified.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Headless Linux benchmark/hash tool (Clang 19 / GCC 14) sharing the viewer's plan builder and
streaming pipeline (moved verbatim into StreamBenchmarkPipeline.h / WorldAssemblyPrototype.h).
Owner Round17 view: density/mesh/payload/topology hashes equal the Windows S13 reference at
W1/W8/W12 (Clang, 3 runs each) and W8 (GCC), with -ffp-contract=off. No output change.
Baseline (i9-12900H, CPUs 0-11, powersave/balance_performance): whole view W8 3.562 s median,
W12 3.512 s, W1 16.555 s; peak RSS < 605 MiB.
Review: the default-stack crash is 17.3 MiB of static TLS per thread (glibc puts it in the
thread stack; HeightfieldColumnCache 6.8 MB), not recursion -- open, after S14.
MSVC compile of the viewer refactor not verified (no MSVC here).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Exclusive per-stage tile timers and probes. Findings: LOD0 is MC-exclusive
(10.4 of 14 ms p95, 243 ns/sample); LOD4's 3x cost is column-cache misses
(8,224 vs ~1,600-1,900 at LOD3/5), cap refine secondary; per-tile feature lists
cost ~0.001 ms; cross-worker column repeats common (277k at LOD0). Global octave
drop and stride-2 interpolation probes fail the 2 cm gate. Worker cap 1 kept
(cap 0 worse 40 m/s tick tail). Report: Standalone/Research/WorldAssemblerM3/Round17.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
voxel.StreamingReadyQueueDepth=8 admits one extra worker-set of prioritized
tasks; voxel.StreamingBackgroundWorkerCap=1 sizes slots to the 8 TaskGraph
background workers actually running tile tasks. Full set 5.96 -> 4.91 s,
LOD0 1.40 -> 1.09 s, ring 0.49 -> 0.51 s, 40 m/s ring 75/75. Tick p95
+0.6 ms spawn / +0.26 ms walk, max flat. No geometry change.
Report: Standalone/Research/WorldAssemblerM3/Round16.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
Startup apply mode on by default (12 meshes / 14 ms per frame, ends when the
initial set and walkable ring are ready, 30 s cap); normal play keeps the
authored budget. Backlog peak 252 -> 10, LOD8 apply wait 1.1 s -> 4 ms.
Ring 0.49 s, LOD0 1.40 s, 40 m/s ring 75/75. Remaining floor: frame-paced
admission (~8.3 submits/frame, 8 running workers max). No geometry change.
Report: Standalone/Research/WorldAssemblerM3/Round15.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
Planar far horizon samples the LOD8 interior at the MC cell spacing, keeping
128-voxel boundary strips; far recipe v3. LOD8 p95 59.8 -> 13.9 ms, spawn
worker-seconds ~41 -> 26.9, full set 7.8 -> 7.1 s, ring 0.49 s, LOD0 1.43 s,
40 m/s ring 75/75. S3/S4 2 cm agreement, 100% cap coverage, W1/W8/W12 and
W1/W8/W16 determinism pass. Report: Standalone/Research/WorldAssemblerM3/Round14.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
voxel.StreamingCollisionLookahead (default on) predicts the support tile 350 ms
ahead along full 3D velocity and prioritises its 9-tile collision ring. Owner
plan, BuildHost actor: 40 m/s arrival-ready 25/74 -> 74/74 (CVar-off control
15/74); 6/12/25 m/s stay 100%; -Z 50 m/s 93/93. Spawn unchanged (~7.7 s).
Critical LOD0 path at 40 m/s: generation 67 ms p50, RMC body update 38 ms p50.
LOD8 cost counted (native horizon 48 ms of 60), optimisation deferred.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
voxel.CoreRegionConcurrentAdmission (default on): tiles independent of the
entering region keep submitting while a region page compiles; pages are
immutable shared_ptr owners so running readers survive eviction. 6 m/s: 11
whole-pass skips -> 0, walkable ring ready at every centre change.
voxel.StreamingResultBacklogCap=256 bounds finished results (slow-apply worst
case 125 -> 76 MB). Spawn unchanged at ~7.7 s: LOD8 is 46% of worker time.
At 60 m/s + teleport the collision ring is outrun (102/113 centres).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
Two admission blockers found by per-frame decline codes: the near reserve held
four slots idle after LOD0 was done (coarse capped at 6), and finished-but-
unapplied results counted against the task cap. Admission now counts queued +
running tasks; near reserve releases once the walkable ring is ready while
stationary (voxel.StreamingDynamicNearReserve, default on). Walkable ring
0.65 -> 0.58 s, full LOD0 2.28 -> 1.44 s, busy workers 2.2 -> 5.6 of 10.
Old behaviour behind voxel.StreamingLegacyPendingResultCap. RMC source reviewed
(read-only): batching/fast-cook ranked, not yet worth it on measured apply cost.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
Direct per-tile trace on the BuildHost actor (owner plan, 2,578 keys): full set
17.4 s legacy; gated coarse-first (after the 9-tile walkable ring) 11.7 s with
the ring still at 0.63 s, but full LOD0 slips 2.3 -> 10.2 s. Startup apply
boost changed nothing (0.4%). 37.6 worker-s over 17.4 s on 10 slots = ~3.3
busy: the floor is admission/work supply. S5's 251 ms staircase was the trace
sampler. Fixed CVar parsing in headless BeginPlay and the 60 m/s crossing
instrument. All three knobs stay opt-in.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
S3's far cap sat up to 8.2 m above the near roof and read as floating pieces.
The far render cap now projects onto the owning strate's upper boundary
(density unchanged), on a global XY lattice, LOD4 16-voxel grid + LOD5-8
128-voxel. Cap agreement 0.05 cm (100 cm control detected), 2520/2520 upward
rays hit roof on the owner walk. Far recipe v2; LOD0-3 hashes equal S2/S3.
Tile p95 flat; cold clipmap +4.4%. Far-cliff fluting left open (coarse MC step).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
Far-shell repair (81-96% of S2's LOD7/8 tile time) is skipped for certified
SurfaceWorld tiles; ground and cap come from the shared planar heightfield.
Near LOD0-3 hashes equal S2; far recipe version 1 salts LOD5+ identities.
Cold first view 6.6 -> 3.4 s (aim 1.57 s). Not a visual sign-off: distant cap
shows floating pieces, sprint streaming arrives late.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
Island-only worker context replaced by a shared worker-local candidate domain
(WorldAssemblyFeatureTileListV1 / ScopedWorldAssemblyFeatureQueriesV1) serving
SurfaceWorld, FloatingIslands and far-shell queries; a per-domain cost hint keeps the
full scan on small plans. Content placement prepares the same queries and skips proven
air: owner island cell p95 7.07 -> 0.24 ms; populated fixture 572 exact placements,
3.27 s -> 29 ms. Full/candidate hashes equal with dropped-feature controls on both
families; S1 default W1/W8/W16 hashes reproduced; Core-off and M2 fixtures pass.
Default contract for future families in ARCHITECTURE 8.23. Run on gpt-6.1-sol medium.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
AVoxelWorld::CoreVoxelSizeCm (default 25, 32 voxels/tile) flows through plan, field,
mesher, content, water, atmosphere, edits and collision. Staged-owner BuildHost passes at
20 cm (SurfaceWorld + islands, W1/W8/W16, anchors, BP edits, collision) with a fixed-25cm
negative control; default-size hashes match round 4; Core off stays legacy 12.5 cm.
Reusable BuildWorldAssemblyFeatureTileListV1; wired into FloatingIslands only so far.
256-island stress: tile p95 112.5 -> 2.0 ms, 0 mismatches, dropped-feature control fails.
Content placement measured (no populated cells): island ground search 7.25 ms/cell vs
legacy 0.29 ms -- open. Outside-review evaluation in Round5/REPORT.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
SurfaceWorld and FloatingIslands plans generate at startup from the resolved seed,
strate types and heights (hashed, cached). Imported JSON is an opt-in override (off by
default); an invalid override shows an on-screen notice and falls back to legacy.
BuildHost on staged owner assets: SurfaceWorld cross-system checks, islands at owner
height (route, collision, anchors, BP edits), seed 137501, W1/W8/W16, Core-off legacy.
XY-column cache verified (hashes kept); island LOD0 p95 91 ms vs 133 ms sprint deadline.
Not exercised: season-seed path (no Season in staged settings), rendered notice pixels.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
Islands read as separate (91.89% open sky at mid-height), underside cavity closed,
rim teeth were capture shading. Island manifest import + BuildHost proof passed on the
pre-cache tree; M1 SurfaceWorld and M2 V1/V2 gates passed.
Known open: final XY-column cache compiled in the owner build but is unverified;
island LOD0 p95 ~155 ms misses the 133 ms sprint deadline (legacy ~43 ms); the island
manifest only fits the BuildHost fixture layout (owner's world: Core init fails,
empty world) -> round 4 builds plans inside Unreal.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ
Fr0zkaandClaude Opus 5.5