Files
VoxelForge/Standalone/Research/WorldAssemblerM3/Linux58/REPORT.md
T
Fr0zkaandClaude Opus 5.5 6c6cc37e1d feat(core): native lattice generator, slice 1 -- helix-ramp spine + small maze (L58); research R27
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
2026-10-25 15:45:32 +01:00

4.9 KiB
Raw Blame History

Linux58 — Native lattice, slice 1

Records and geometry

The fixture uses seed 0, strate 1, and recipe Core.Lattice.PrimitiveField version 1. Room and edge IDs are stable and keyed by (seed, strate, recipe_version, stable_id). The resolved arrival is passage/0-1/0 lower at (9.142, -2.073, -122.500) m; departure is passage/1-2/0 upper at (12.271, -3.565, -149.500) m. The plan resolves 9 rooms and 9 edges before branches: arrival/departure rooms, maze cells 00–04, two dead-end rooms; arrival portal, helix ramp, maze edges 00–03, departure portal, and two dead-end edges. The canonical record digest is 639d8b975024c9824c944f8244fae88d796350da5f4168b85d5ec7b501c5a842.

The mandatory spine descends 20 m through four helix turns: 6 m centreline radius, 3 m clear width, 5 m pitch, and 2.6 m finished floor-to-soffit. Measured grades are 7.555° at the centreline and 10.028° at the inner edge. The cubic smooth-max fillet measures 0.570778 m (required >=0.500 m). The prototype’s accepted outer shaft envelope is 21 m. The maze has five cells and two dead ends. This recipe is opt-in; legacy Maze and VerticalShaft assets remain legacy.

Reachability, intervals, and surface

On the final field, arrival-to-departure reachability passed all 883 stations: 44,150 capsule samples, 4,415 support samples, and 4,415 headroom samples had zero failures. Maximum grade was 10.028° and minimum headroom was 2.200 m. The deliberately disconnected shaft control was rejected (reachable=no; zero connected route stations).

The interval oracle checked 4 boxes against 1,372 brute-force samples with zero misses. It rejected both false all-air and false all-solid proofs; isTileShapeFree blocked the occupied wall-foot tile. The whole-surface audit covered 89,577 LOD0 triangles across the helix edges and maze wall feet: zero flagged triangles and zero clusters. The L45 ramp, L46 entrance, and L47 wall-foot controls failed as expected; the plain-native control passed.

Peers and timing

Clang and GCC builds used FP contraction off. W1/W8/W12 matched for four randomized cameras, including one inside the helix, with matching tile hashes and the canonical record digest above.

Clang measured first tile 0.201 s, near ring 0.345 s, whole view 1.694 s, and moving p50/p95 485.536/977.128 ms. The whole view is above the 0.7–1.0 s target; moving p95 is just under 1 s.

Captures and hashes

All four JPEGs passed the capture gate, were visually reviewed, and are below 300 KB.

View Bytes SHA-256
captures/lattice-helix-top-down.jpg 19,616 0ab87ee1e799ce96752dc3bf9d6803928e4d6aaad8a87400d899ff3f3be06149
captures/lattice-ramp-eye-next-turn.jpg 22,540 d16b0ad577f5ec0e676e18ba62a5de64c71385cd93b53b225ac25f109b184265
captures/lattice-maze-eye-level.jpg 12,361 463ee13e962ecc36de1778abcd09f34f5bd375db5537395e7b1e0e0df3d72331
captures/lattice-cutaway-overview.jpg 28,673 10204ac2b120ad360e31ac589351c31d0c0e9603feb22841860bd4e145613194

Source stamp: 7d181a5b606879ae7277db296e43df11f3fdef41eefa127f5c86d8a431fc4a10. The owner input remains unchanged at SHA-256 7b095f9bcc6b90133d5a463f99f0b571133b2c549ce1ecf2d64fe9175e41e183.

Tiers, trials, and failed runs

verify.py fast passed in 176.4 s; medium passed in 1,170.0 s. The single full run on the clean build reached 1,700 s and timed out during the later GCC motif matrix. Before timeout, its L58 GCC field, orphan control, W1/W8/W12 peers, inside-helix camera, and Clang comparisons had passed. No second full run was made.

The run ledger records 10 verification-tier invocations against the 15-trial limit: 3 fast, 6 medium, and 1 full. Earlier failures were a stale style assertion; capture framing/quality; the initial broad transition ROI and boundary seam; and a floor-profile ripple (0.0887 m, adjacent step 0.1791 m) before blend tuning. One direct surface audit initially had a missing math import, and one diagnostic parse had a syntax error; both were corrected. A post-tuning fast assertion also used stale blend values. The final fast and medium tiers passed. The only final-tier failure was the full-run timeout above.

Orchestrator review (Claude)

Accepted. The stamp matches the final Clang and GCC builds (7d181a5b…). The full tier had timed out at the builder's 1,740 s process wrapper (the tier itself now runs about 38 min). I re-ran it with the tier's own 7,200 s limit: PASS, 2,307 s. Reachability passes at all 883 stations, the orphan control is rejected, and the surface audit is clean. Two items for the next lattice slice:

  • Whole view 1.69 s is above the about-1 s target.
  • It reads as a corkscrew tunnel bored through rock (separate stacked turns with rock between them). It doesn't read as an open vertical shaft with a ramp around its wall, which is what "vertical shaft world" implies. Open the shaft core in slice 2.

Removed the duplicate control SVGs (already in Linux48).