Files
VoxelForge/Standalone/Research/WorldAssemblerM3/Linux171/REPORT.md
T
Fr0zkaandClaude Opus 5.5 9d088ef96c research: UE arrival A/A could not run (L171)
Two identical ca0bf76 VFHost builds staged together crash at BeginPlay after snapshot parsing
(surface_strates=858993460 = 0x33333333, std::length_error). No arrival pairs measured.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
2026-11-02 18:08:43 +01:00

1.3 KiB
Raw Blame History

L171 — UE arrival measurement

Result: Not measured. The A/A warm-up aborts before the first streaming sample, so this round cannot say whether L170’s +1.28% arrival difference exceeds UE noise or identify its cause.

Two assembler/world ca0bf76 VFHost Development builds completed and produced identical VoxelForge, VoxelForgeCore, and VoxelForgeEditor binaries. With that matched module set staged, the A/A warm-up reaches world_assembly_plan with one surface stratum, then logs surface_strates=858993460 after snapshot parsing and terminates with std::length_error: vector during AVoxelWorld::BeginPlay. The trace has no scheduler readback or streaming sample. This points to a Core snapshot/module compatibility failure; the exact cause is unproven.

Valid arrival pairs: A/A 0/15, switches-on 0/15, switches-off 0/15. No medians, paired spread, or slower counts are available. The A/B batches were not started because the A/A control failed validation. Failed launches are logged under Saved/Linux/l171/ue/arrival/.

The baseline and L170 HEAD-off VFHost builds passed. The runner compiled and its CLI loaded; git diff --check passed. No Unreal Source/ files changed. The full arrival matrix and fast verification did not run: the UE control never reached measurement.