VoxelForge/Standalone/Research/WorldAssemblerM3/Linux141/REPORT.md

L141 — first liminal layer

The standalone fixtures and six captures are ready for review. The flat field has one oversized capped-cylinder monolith. The road is a 65 m fillDais platform, 6 m wide, with its deck 22 m above a flat field. Fog supplies the road's cloud-sea backdrop in these standalone images.

The periodicHills fixture uses the validated 80 m periods and 2.2 m amplitudes on both axes. Its field repeats exactly, but these captures show one broad rise rather than hills repeating to the horizon. The validated period is longer than the 56 m eye-level capture radius; the wider full-detail capture exceeded Core's 160-million-node interval bound. Smaller periods fail the existing profile validator. A capture extent that trims vertical tiles independently from horizontal tiles is the next useful step; this round leaves the Core profile limits unchanged.

Changes

• Reused Core's flatField, periodicHills, fillMonolith, and fillDais support. Added standalone fixture inputs and runners in BuildL141Fixtures.py and CaptureL141.py.
• Added liminal_road_override as a standalone adapter to the existing free-placement fillDais stamp, plus --audit-liminal-road-walk. The walk audit samples the Core field with the existing capsule sampler and checks width and grade.
• Added optional distance fog to LinuxOffscreenCapture for the screenshots. The fog is capture-only; Unreal cloud rendering remains Unreal-side.
• No evaluator change or Unreal Source/ change. Owner output hashes match their pinned L130/v227 canonical rows in the fast tier.

Measurements

• Determinism: each fixture matched across Clang and GCC at W1, W8, and W12 (six cells per fixture). The matrix used LOD0, clip radius 1; each cell compares recipe/plan hashes, resolved-layout identity, and per-tile hashes. Individual logs are in Saved/Linux/l141/logs/determinism-*.log.
• Raised-road capsule walk: 127 stations, 8,001 capsule samples, zero blocked samples, zero unsupported stations, 6.000 m width against a 0.680 m capsule diameter, and 0.000° grade. See Saved/Linux/l141/logs/road-walk-audit.log.
• Near-ready: one W12 LOD0–2 sample after one warm-up, clip radius 1, 139/139 tiles. The measured cold_ready_s was 0.173886 s. The window was one run after one warm-up; uptime showed a 1-minute load average of 1.31 and no other build was running. See Saved/Linux/l141/logs/near-ready-measured.log.
• Each capture is 1600×900 and passed capture_quality.py. Eye views place the camera 1.6 m above the field or road deck. Camera coordinates and image hashes are recorded in the capture runner output; export and render logs are under Saved/Linux/l141/logs/.

Checks

• Clang 19 and GCC standalone builds passed with FP contraction off, four jobs, and source stamp 0a2828c398aa62870690667901d44f2daf5b1a5547b9640da6dbb01df721af8b.
• python3 Standalone/Tools/verify/verify.py fast --out Saved/Linux/l141 --overall-timeout 1740 passed in 285.7 s. Owner SurfaceWorld and FloatingIslands W1/W12 canonical hashes matched the L130/v227 baselines.
• Fixture determinism, dedicated road walk audit, six standalone captures, and all six image-quality checks passed.
• Python bytecode compilation and git diff --check passed.
• An exploratory broad --audit-resolved-layout-passages invocation also ran unrelated passage ripple checks and reported 2/3 passage ripple records passing, with passage/1-2/0 failing. The focused liminal road audit passes; no legacy threshold was changed.
• No Unreal build was run because no Unreal Source/ file changed. No timing matrix was requested or run.

Remaining risk

The hill screenshots do not yet show multiple full repeats to the horizon. The profile itself is exactly periodic; the standalone capture volume is too short at the current full-detail safety bound. The raised road is geometric and walk-audited, while its clouds are represented by fog in the capture renderer.