- scripts/build_solids.py: B-rep solids from standard section 03-001 (rev2, approved 2026-10-21)
- lining per 지보패턴 run (independently selectable), excavation band per run
- road surface (계획고, 280 mm) with per-station superelevation; base/invert ASSUMED
- side drains (box minus pipe), H/I boxes = 공동구 (assumed); 시설한계 50 mm as reference wires only
- staged run (4 GB RAM / 180 s per call); STEP and GLB from the same B-rep
- model/tunnel_solids.glb, model/tunnel_solids_meta.json, data/verify_solids.json
- viewer/index.html: solids, pattern and component toggles, click readout, no development text
- STEP (84.6 MB) and model/parts/ are git-ignored; the STEP is generated locally by the script
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
Independent re-derivation from the workbook and GLB vertex colours: 42/42 rows,
735/735 stations decoded to the workbook pattern, no mismatches. Boundary blend
colours (39) noted as a limitation. V14 remains partial (lining report self-checked).
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
Source: RAG 지보패턴현황_20260918.xlsx, sheet 참고_지보패턴현황, 당진방향 table
(4+740..5+500, 42 contiguous rows). Lining pattern (M column) per station now
drives the lining colour and the viewer readout. Daesan table kept as reference
only (WL-4 there, not applied). Drawing patterns E-1, E-2, RP-2, WL-3 are in
neither table and stay [미확정].
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
Viewer loads with no page errors; clicking the lining with other layers hidden
returns station, placeholder pattern and lining spec. V12 (original fingerprint)
and V14 (independent verification) recorded as not performed.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
19 checks: 6-sheet common values, 3-sheet temperature-exempt detail, 9 pattern
groups vs report rows, R1 6.982 vs 03-001 raw texts, thickness arithmetic,
output naming and version, meta content, no writes into RAG.
Report fix found by the check: SD400 label was missing; added.
Decision recorded: R1 = 6.982 m kept (user, 2026-10-21).
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
Extracted per-sheet pattern groups (입출구부 50 m / 온도미고려), note-5 scope,
fck 27 MPa, SD400, aggregate 25 mm, rebar label counts and printed radii.
Model keeps R1 = 6,982 mm; lining sheets print 6.892 m -> [CONFLICT DETECTED]
for user check. Sheet-to-station and pattern-name mapping left [미확정].
Viewer shows the lining spec on click.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
No pattern-to-station data exists in the drawings. Per user decision
(2026-10-21) the modelled range 4+745..5+480 is split into 2 m segments with
arbitrary pattern names, marked placeholder in CSV, meta and viewer, to be
replaced later. Ring step reduced to 1 m so that the 2 m segments align.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
build_skeleton.py sweeps the confirmed standard section (R6982 inner, 300 mm
lining, R7402 excavation, crown 6.615 m above road) along the centreline for
4+745..5+480, road = 계획고, ground = 지반고 band. Walls and invert are
approximations and recorded in meta. Verified: inner radius identity, crown
height, chord length 799.9 m for 4+700..5+500. Viewer uses three.js 0.160.0.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
Table 1 (R1..R3', phi1/phi2) and table 2 (A..K, 라이닝 두께 300) are parsed by
S(%) rows +3..-3. R1=6982, R3'=4699, R2'=3910 match the arc list; lining
thickness 300 mm matches the concrete arc band. Units of a..f and the
cross-slope station ranges need HITL confirmation.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
Ruler 4+740..5+480 (38 stations, 20 m). Two decimal series per station:
A (|across|~33 m) varies with terrain; B (|across|~16 m) rises ~0.498 m per
station (2.49% grade). Series naming and the ruler scale (chord ~21.27 m per
20 m station) are left to HITL. Output is a draft; no value is adjusted.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
extract_alignment_plan.py takes the midpoint of each station's two side
ticks (CA-ALGN-STSM) as the centreline point. Sheet 001 gives 4+700..5+500
with consecutive residual <= 0.02 m. Sheet 002 gives identical output, so
the duplicate question is left to HITL. Profile EL is not yet extracted.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
Scans 21 main-line DXF files (02 평면·종단, 03 본선굴착, 04 본선기타) for
header, layers, entity counts, extents, 3D flag and text samples. Strict
load failures fall back to ezdxf recover mode and are recorded.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A5Cun3RHQKna9qxE6ybZ4A
