Each passage records upper and lower landing chambers (floor, gradient, radius, 4.5 m clear height) composed with the
tunnel shell; cave route endpoints bind to the chambers; envelope audit permits only the named chamber joints.
Seed 2: 0 solid in 319,579 passage+chamber capsule samples, 0 shell violations, 0 envelope conflicts, controls fire.
Open: tunnel-to-chamber floor blend leaves 19.1/19.0 deg grades (limit 15); sweeps/timing not run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Column-cache census matches assembler/world (the L77 share rise was relative, not more calls). Reuses the TN
predicate and skips relief/mouth/shell work for stamped samples outside the 1.053 m influence. vs assembler/world:
whole view -0.5..-4.0% on all layers; near-ready Surface World +0.3/+1.7, Floating Islands -4.3/+2.9, Tunnel Network
-2.5/+2.9, Vertical Shafts +4.7/+3.3 (A/A ±1%). 16/16 seeds, fast tier, Clang/GCC green.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Identical builds differ <=0.6% median; residual near-ready vs assembler/world is real: Surface World +2.9..+3.6%,
Vertical Shafts +4.0..+4.3%, Tunnel Network -3.6/+6.8%. Profile points at column-cache reads and the sweep lookup.
Removed an unused sample field and per-lookup shared_ptr refcounting. 16/16 seeds, fast tier, Clang/GCC green.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Passage visits -78..-89% near-ready; composer worker time back to baseline on Tunnel Network; v239 is 2-4% faster
than v236 in 7/8 cells but still +4.3..+13.3% near-ready vs assembler/world. Surface World composer time equals
baseline yet near-ready is +4-6%: unexplained. 16/16 seeds, fast tier, Clang/GCC green.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Mouth floor/profile carried into the first/last edges (fixes seeds 1 and 2); fixed a Q16/Q32 scale bug in the mouth
clearance scorer; shortest grade-safe handles; bounded cave AABBs and hoisted edge-BVH normalisation. Seeds 0-15 pass
all composed audits and negative controls; fast tier, Clang/GCC W1/W8/W12 green. Near-ready still +3.9..+13.7% vs
assembler/world (Surface World +3.9..+8.2% with no caves). Not for merge.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
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
Arrival edge floor aligned to the passage mouth, 0.75 m backbone floor-support budget, anchors on the planned floor:
0 unsupported, 0 grade failures. Open: 196/50,967 capsule samples blocked where a higher descending edge's floor
backing meets the next edge (route-edge-6/7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
One current-state page (shipping behaviour, references with sources, gaps, verification, closed lines), linked from
CLAUDE.md. Removes the L86/L88 passage/boundary ablation switches and the unused legacy route-grade audit; inventory
of every kept switch in the report. Output identical: 24/24 + 18/18 hashes, fast tier, walk/seam audits, UE host
build; timing unchanged (-0.2%, CI spans 0). No Source/ changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Reach = 0.75 shell + 0.21 relief + 0.50/6 smooth-max support + 0.01 = 1.0533 m, used by proof, feature, fallback,
off-grid bounds and the sweep cutoff. Removes the dead 4 m term (v225 added metres to core-voxel coordinates).
Captures and mouth close-ups byte-identical; proof audit 0 misses over 25,728 shell/crest samples; floor 7.884 mm;
seams, walkability, fast tier, Clang/GCC W1/W8/W12 green; timing equal within noise.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Conservative route slabs with an outside sentinel, tight expanded segment bounds; DensityGrid exact fallbacks
479k -> 0 per run. Near-ready 2-14% faster than L119, still +4.7..+10.2% vs L109; whole view within -0.5..+8%.
Same look (floor offset 7.884 mm, captures checked), seam audit bit-identical, fast tier + walk audits,
Clang/GCC W1/W8/W12 determinism green. BVH and 25 cm tolerance trials rejected. Not for merge yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Chord error 2.5 -> 10 cm, shell work skipped outside the 0-0.75 m band. Sweep pairs -39%, prepare -60%,
near-ready 3-8% faster than L115 but still +13..+22% vs L109. Floor offset 7.88 mm; captures match L115
by eye; fast tier, walk audits, Clang/GCC W1/W8/W12 determinism green. Not for merge yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Floor height from the planner profile h(s) by horizontal projection + arc-length interpolation;
adjacent fine segments share endpoint values. Floor zero-crossing offsets max 7.8 mm (L114: up
to 73 mm); rock-floor close-ups show no ribs. The 2 L114 shell misses were oracle false positives
(points inside an overlapping bore); the oracle now checks every overlapping span; no gate relaxed.
Fast tier PASS; Clang/GCC W1/W8/W12 deterministic. Speed regression grows: near-ready +17..+31% vs
L109 (fine-segment/sample pairs 49.0M vs 19.8M). Not merged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Floor ribs are present in L109/L112/L113 rock tunnels too: the floor height comes from the
nearest segment, so it jumps (up to 5.9-7.3 cm) where the nearest segment switches. v222 reduces
them; faint ribs remain. Quiet-machine timing (load 0.64): near-ready 9-17% slower than L109
(FI W12 0.95 -> 1.10 s); candidate totals ~5x L109. Fast tier FAIL: 2 SW wall-shell probes of
31,381 read host air. Walk/grade audits pass; Clang/GCC deterministic; UE builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Open-air walls: 0.75 m rounded shell, relief fades over the top 0.25 m - the needle spikes are
gone. Every owner plan passes the 15 deg forward-grade check (max ~10.7 deg) and all walk audits.
Floor bands: median 1.25 m interval, ~0.21 m drop on a ~10.4 deg floor = one 0.25 m voxel row
projected along the grade (not control points, not slab thickness) - not yet fixed. Exact
segment-to-tile filter did not reduce candidates; removed. Timings this round are unreliable
(a gitea process used ~43% CPU throughout). Fast tier PASS; Clang/GCC deterministic; UE builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Dense centripetal spline in XY, height piecewise-linear on the planner's certified profile,
radius/floor depth blended; open air = floor slab + 1 m side walls, no roof; audits sample the
generated centreline. All walk audits pass (grades <= 15 deg, headroom, shell, mouths, ripple);
Clang/GCC W1/W8/W12 deterministic; fast tier PASS; Linux UE builds. Chevron rings gone.
Open issues: jagged spikes along the half-pipe wall tops, visible step lines on the trough
floor; near-ready +7..+10% vs L109 (candidate list p95 4 -> 22). Not merged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Route resampled to ~1 m (310 -> 2,790 segments); open air writes a 0.75 m floor slab plus 1.0 m
side walls, no roof. The 3D spline overshoots between planned route points, raising floor grades
above 15 deg; a grade-limited variant left 2 unsupported samples on passage 0-1, and the shell/
ripple audits still model the old chord route. Stopped after two designs. Not merged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Owner design: flat world-horizontal floor (0.50 m corner blend), 0.75 m solid shell written only
where the host field is air (interiors union first and beat every shell), 5 m cubic mouth ramps.
Removed route support restoration, deferred support, support columns, mouth-support restoration
and their capsule checks (-5,129 / +1,641 lines). Near-ready @W12: FI 1.39 -> 0.86, TN 1.29 ->
0.90, VS 0.85 -> 0.74, SW 0.70 -> 0.76 s (more shell/skirt surface). All walk audits pass;
Clang/GCC W1/W8/W12 deterministic; fast tier PASS; Linux UE builds; 6 UE arrival traces pass.
OPEN ISSUE: open-air shell (e.g. passage 0-1, 1037-1489 m route span) shows a regular stepped
banding pattern - not merged into assembler/world until fixed/approved.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Owner shapes compile per stratum on the Core parallel scheduler and merge in owner-layout order;
far-parent ellipsoids/connectors generate in parallel and merge canonically; 512-unit far bins.
Base compile 239 -> 203 ms, far-parent build 128 -> 33 ms. Near-ready @W12: SW 0.84 -> 0.75,
FI 1.51 -> 1.42, TN 1.38 -> 1.30, VS 0.92 -> 0.85 s. All hashes equal L104 (Clang/GCC W1/W8/W12).
Fast tier PASS; Linux UE builds; 6/6 Unreal arrival traces pass.
R45 (read-only, owner's "are we overcomplicating?"): next simplifications = tunnel floor/mouth support
inside the bounded passage field; boundary bands once per XY column; certified early SW tiles.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Near tiles legitimately need floor/ceiling boundary patches, but every hit query scanned the
tile's whole XY column. Direction-specific 3D cell bins conservatively index triangle bounds; the
existing predicates still decide every hit. Candidate visits per query -88..-99%; identical query
and hit counts. Hit-function cost ~1.7-1.9 -> ~0.35 s per FI/TN run. Near-ready @W12: FI 1.63 ->
1.52, TN 1.51 -> 1.38, VS 1.00 -> 0.92 s. All tile/whole hashes equal L103 (Clang/GCC W1/W8/W12).
The 24-28% "idle" is in the serial base-shape (~0.24 s) and far-parent (~0.13 s) phases; the
LOD0-1 drain is >= 99.4% busy. Fast tier PASS; Linux UE builds; FI Unreal arrival 1.96-1.98 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Off-grid census by caller (MC midpoints, skirts, far shells); all 63.0M sweep misses accounted
(1.9M exact-route fallbacks, 61.1M counted bound skips). Exact MC midpoint queries memoized by
float bits (Unreal enables it too); feature bounds expanded. L100/L101 whole-view hashes equal in
all four strata (v216 unchanged); 0 added/removed crossings; 15 capture pairs pixel-identical.
Near-ready @W12 L100 -> L101: FI 1.77 -> 1.65, TN 1.66 -> 1.54, SW 0.85 -> 0.84, VS 1.03 -> 1.03 s
(L98 2.28/2.23/1.02/1.11). Fast tier PASS; Linux UE host builds. Unreal arrival traces not obtained
(the runner wrongly applied the 6 GiB address-space cap to Unreal).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
L99 treated a missing exact-grid sweep sample as "no route here", so off-grid queries exposed the
FI/TN boundary band inside passage 1-2 (flat ceiling/floor). In-bound misses now evaluate the
exact route; a conservative route-feature bound still skips out-of-extent misses. Composition
order unchanged (graph -> bands -> passage -> stamps -> floor caps).
Comparison now counts added/removed crossings and triangles: 0/0 crossings on all four strates
(FI +56 triangles). 1-2 mid-tunnel: L98->L100 changes 32k px (L99 changed 1.14M).
Near-ready @W12 (L98/L99/L100): FI 2.62/1.78/2.05, TN 2.36/1.55/1.80, SW 1.17/0.95/1.04,
VS 1.16/1.03/1.08 s. All walk audits pass; 24/24 Clang/GCC W1/W8/W12 determinism; fast tier PASS.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Route plan unchanged. Per tile, route segments whose bounded influence (tube + 4 m) touches the
tile write min-combined tube distance (base sweep); relief/floor/mouth shaping only in the shell
(detail sweep); 0.5 m edge blend into the strate field. Near-ready @W12: SW 1.01->0.80,
FI 2.23->1.43, TN 2.18->1.37, VS 1.09->0.98 s; whole view FI 3.38->2.37 s. All passage walk
audits pass (21,425/21,425 canonical, 3/3 interfaces); 24-run Clang/GCC W1/W8/W12 determinism;
0 sign flips. Change: mean 0.02 cm / 0.05 deg; max 25.9 cm (passage walls/roofs ~3.5 m from
route, LOD2) and 49.8 deg (rock walls 13-25 m from routes, LOD2). Captures with diffs in
Linux99/captures. Fast tier PASS (v215 pins).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Per-world prepared chord/length/radius/Catmull-Rom/floor-profile data (310 segments, ~66 KB,
<0.04 ms); same expression grouping, divisions and square roots, so output is bit-identical
(zero density/tile mismatches, Clang vs GCC equal). No measurable readiness gain - the cost is
the number of samples running passage work, not per-sample arithmetic. Note: the report's
world labels are shifted by one row (the 2,850-tile row is SurfaceWorld). Fast tier PASS;
Linux UE host builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Explicit carve reach R (diagnostic, default unbounded): R=2.5 m -> 1.56 cm max displacement,
106 deg max normal change; R=4 m -> 0.63 cm but 53 deg normals. Fails the bar; readiness gain
small and inconsistent (TN -5%, VS +6-7%). No world change (v214). Fast tier PASS.
R43: Unreal SW bReady 2.16 s = serial sign-edge repair (1.94 s, 10.9M corrections;
parallelSignEdgeRepair exists but off); ~0.45 s of SerializeResolvedWorld = adapter SW route search.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
- Startup apply boost (24 meshes / 14 ms) now ends on initial LOD0-2 + collision readiness or
the 30 s timeout, not on the first desired-epoch change: FI LOD0-2 settled 3.49 -> 3.40 s.
- Known-empty LOD0 collision classifications persist across unloads within a modification epoch
and are invalidated on edits; commandlet readiness accepts CollisionNotRequired tiles. The 30 m/s
movement trace now passes (no re-requests of the 4 support tiles).
- Startup trace events: SerializeResolvedWorld ~466 ms synchronous in BeginPlay (unchanged);
SW bReady at 2.74 s vs FI 0.75 s (to investigate). Redundant second parse removed.
- Core output identical. Fast tier PASS; Linux UE builds. R42 notes attached.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Clamping the passage-composition source density at -c (c=2..6 voxels, diagnostic, default off)
changes the surface far beyond the bar: c=6 still 65 sign flips, up to 165 cm displacement, 120 deg
normal change (downstream composition consumes deep values). Source gradient at crossings is
distance-like at the median but has huge tails (FI p90 |grad| 512, max 894 vox/vox), so the
density-dependent passage reach is inflated by non-distance values. No world change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
54,324 sparse 4 m bricks, uncapped carve/tread/clearance/mouth lists, built in 26-28 ms,
~12 MiB. Weighted over 30.7M near-LOD sample hits: active carve p90 13, mouth p90 13; reach
fallback keeps 18.3% of passage time (FI 12%, TN 19%) - all because the density-dependent reach
(-d*Q + 0.24 m) exceeds 10 m for deeply negative source samples. Not wired into production.
R41 notes (Unreal LOD0-2 gap: ~0.7 s pre-request in BeginPlay, LOD2 tail, apply boost reset;
L80's 4 missing tiles = re-requested empty collision-support tiles) attached.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Verified ablations (zero passage work / zero band calls). LOD0-2 @W12: exact FI 2.27 / TN 2.25 s;
passages off FI 0.81 / TN 0.85; bands off FI 2.11 / TN 2.06. Passages cost ~1.4-1.6 s of near
readiness, bands ~0.1-0.2 s. A tile-local 4-voxel trilinear passage-delta cache increased carve
segment visits (FI 34.5M -> 56.2M) and slowed LOD0-2 (FI 2.58 -> 3.77 s); removed. No world change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
L86's switch left ApplyResolvedStratePassageCarves running (near-tile service unchanged at ~21
worker-s), so its "no change" ceiling was wrong. Corrected (bands + carves off): FI LOD0-2
2.34 -> 0.66 s, whole 3.47 -> 1.49 s @W12; TN 2.27 -> 0.74 / 3.28 -> 1.52 s. Near-tile service
15.6-22.5 -> 2.6-3.8 worker-s. Critical chain: shape/base compile -> LOD0-1 -> far-parent build
(107-153 ms, pipeline thread) -> LOD2. Workers 81-85% busy in the direct window.
No production scheduling change; hashes equal L85. Fast tier PASS.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Environment-only diagnostic hook disables passage composition (never a world setting).
Cold LOD0-2 / whole @W12 WITHOUT passages: SW 1.12/2.25, FI 2.18/3.34, TN 2.15/3.15, VS 1.12/2.03 s
- the same as with passages (FI 2.16 s). Passages are ~58% of sampled CPU but not the
readiness bottleneck. A lazy exact passage-result cache was +47% slower (24% hit rate, ~700 MB)
and was removed. No world change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Tiles with passage-overlap score >= 9 split grid fill over up to 4 helper lanes (disjoint rows,
same per-sample order, join before MC) when workers are idle. FI junction tile fill 173 -> 99 ms;
estimated serial floor ~0.55-0.60 s. Readiness essentially unchanged (FI/TN LOD0-2 ~2.15 s @W12):
the limit is now aggregate work, not single tiles.
Unreal: LOD0-2 ordered by passage overlap before travel/distance (L80 modes kept); far parents
built after LOD0-1, LOD2 gated on them; helper budget via ParallelFor. Unreal null-RHI FI
arrival: LOD0-2 resident at 3.65 s.
All tile hashes identical to L84 (four whole views). Fast tier PASS; Linux UE host builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Exact through LOD2 (all near tile hashes equal L82). W8 LOD0-2: SW 1.09->1.06, FI 2.47->2.34,
TN 2.39->2.43 (noise), VS 1.20->1.18 s. Fast tier PASS.
Scaling W4/W8/W12/W16 (LOD0-2, FI): 3.59 / 2.48 / 2.33 / 2.34 s - flattens past W8-W12.
Estimated infinite-worker floor ~1.0 s FI/TN = shape compile (0.35-0.41 s, not overlapped:
no shape-free tiles in these views) + slowest single LOD0 tile service (0.51-0.59 s).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
- Wall relief noise is skipped when its hard-bounded contribution cannot change the final max;
mouth surface/tread noise only on paths that use it.
- Exact active BVH-node masks cached per tile/passage/segment range; no per-sample lower_bound.
- Full-view audit imports the exported Core plan (not the stale 26-point root route): 21,425/21,425
walkability samples pass.
Cold (L80 protocol, medians) LOD0-2: SW 1.16->1.11, FI 2.86->2.51, TN 2.88->2.54, VS 1.23->1.22 s.
Passage share FI 64.6->58.7%, TN 65.4->58.4%. Hashes identical (v214). TN "slowdown" from L81
not reproduced (5 pairs, 3.08 -> 2.75 s). Fast tier PASS; Linux UE host builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
The carve is max(source, -roughSDF/Q), so its reach grows with -source; the outer test uses
-d*Q + 0.24 m (wall relief 0.21 + margin) beyond the route radius, disabled in mouth-support
and column-overlap tiles. Skips 16% of passage iterations on the highest-activity tiles and
all of them on ordinary tiles. 84 candidate/full pairs identical in every hash; 0 changed samples.
Evaluator 214 for cache invalidation only. Cold LOD0-2 (L80 protocol): SW 1.30->1.25,
FI 3.59->2.93, TN 3.18->3.51 (noise?), VS 1.33->1.22 s. Fast tier PASS.
Note: the export's legacy 26-point root route owner_world.passages[0] fails a full-view
pre-audit (65 solid samples, unchanged since L70); the canonical 211-point route used by the
plan passes. Stale data to clean up.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
- Headless and Unreal schedulers drain the collision-critical ring, then LOD0-2, then far LODs
at low priority. LOD0-2 ready (W8 medians): SW 4.36->1.76, FI 6.89->5.26, TN 4.88->3.22,
VS 2.71->1.56 s. Whole view roughly unchanged (FI/VS slightly later).
KNOWN RISK: far tiles wait until every near tile is ready, so far LODs may starve while
moving fast (to address next).
- Exact composer trims: tile-level upper-mouth candidate list, segment-index min/max rejection
before lower_bound, 64-entry thread-local value-noise corner cache. Passage share 65 -> 63%.
- World exact: 24/24 hash matrix + pinned L75 owner hash. Fast tier PASS. Linux UE host builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Opt-in (L77 build + --linux77-profile; compiled out by default, 0.02% handler cost).
FI/TN cost account closes against direct grid-fill timing within 0.15% (LOD0-2 bins within 1.4%).
Passage carving = 66% of LOD0-2 CPU (52-54% whole view), passage support 10-14%, base packed
+ analytic field ~3%. Hotspots: value noise 17%, lower_bound in VisitNearestPassageSpatialNode
14%, ApplyResolvedStratePassageCarves 6.5%. LOD0/1 complete LAST (4.1/4.6 s) while LOD2 is
ready at 2.2 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
Opt-in at build time (VOXELFORGE_PROFILE_LEGACY_TERMS=1; compiled out by default, ~2% when on).
Per legacy sample: passage/support composer ~1.05 us vs primitive scan ~0.19 us, boundary
band ~0.06 us, noise ~0.01 us. Resolved-passage constants are cached at field binding
(arithmetic order unchanged); whole-view effect within noise. Per-tile hashes identical to L75
in all 24 compiler x worker x view comparisons; fast tier PASS; Linux UE host builds.
R40 researcher notes attached.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
L74's passage bound omitted the 0.21 m wall relief and 0.20 m floor relief, so its whole-tile
proof was unsound in principle (no tested view hit it). Passage support is now bounded per route
segment by four parameter-subinterval boxes using Catmull-Rom derivative extrema, local max
radius and relief padding. Stratum slicing only at LOD0-2 in FI/VS. Legacy sub-tile block
certificates are audit-only (runtime version was slower). TN regression (-80,-20,-3218) kept.
Exact full-grid vs skip hashes on every tile, 4 views, Clang/GCC, W1/W8/W12; control caught;
80-camera sweep 0 violations. Whole view vs c67b200: SW -1.3%, FI -7.6%, TN -2.3%, VS +0.2%.
Compiles in the Linux UE host (VFHostEditor, 27/27 actions).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
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
- 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
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
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
Fr0zkaandClaude Opus 5.5