Two vote-killers found on 13号地块 towers: axis geometry sits OUTSIDE the
wall-tight clip (stubs beyond the facade), and min_grid_len = pitch*0.25
discarded 0.8m axis-number ticks outright. Axis candidates are now
collected in a pitch*0.16-expanded window, and strictly axis-parallel
stubs (pitch*0.004..0.05) count alongside full-span lines — same-position
ticks cluster into one vote. Synthetic tick fixture goes WEAK -> full votes.
Honest finding logged: the 13号 tower frames genuinely contain NO axis
geometry (all 249 ticks belong to other frames), so votes=0 is the correct
verdict there — modeling proceeds via clip-corner fallback plus mandatory
per-floor overlay verification; wall-position correlation noted as the
future alignment source when axes are absent.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
13号地块 prefixes every layer (A_DOTE/A_AXIS/A_WALL); the exact-token
axis defaults silently zeroed grid-correlation votes for all 44 frames.
Defaults now *AXIS*/*DOTE*; rowframe fixture gains a layer_prefix="A_"
variant asserting the grid still votes (no WEAK).
Logged open item: the tower sheets still vote 0 after the pattern fix —
axis stubs vs full-span grid lines or clip-window semantics, to debug next
round; plus the plot-13 storey table and the finding that per-floor
multi-clip stitching already works via repeated floor keys.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ODA auto-fallback validated at scale: plot-13 18/18 files ok (8 rescued),
plot-27 13/13 ok (9 rescued) — previously 9 failed + 8 empty-modelspace
main sheets. Documented the next structural gaps found in plot-13's tower
sheets: multi-wing (座) title prefixes stored as separate text entities,
zoned per-level plans needing multi-clip stitching (one clip per floor today),
and typical-floor (标准层) replication; elevations and site plan are unbound
XREF shells, so storey heights need another source.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Conversion/encoding chain (libredwg on Chinese TArch DWGs):
- run_dwg2dxf: capture bytes, decode leniently (GBK console output crashed
the batch under subprocess text=True)
- sanitize_dxf_comments: byte-level, code/value pair walk — the UTF-8
errors="ignore" round-trip mangled every GBK CJK string ("一层平面图"
-> "一平图") and eating XRECORD "999" VALUE lines corrupted files
- read_doc: recover fallback also on ValueError; scrub_surrogates keeps one
corrupt drawing from killing the whole report
- decode_mif in report + frame detector: \M+5XXXX escapes (whole 黄桥 set)
restored so plan-title detection works at all
- ODA auto-fallback: reconvert DWGs whose libredwg DXF fails inspection or
reads an empty modelspace (黄桥 plans: 0 entities via dwg2dxf, 21k via
ODA); env ODA_CONVERTER override, fake-converter regression test
Frame detection:
- detect_sheet_frames expands INSERTs (memoized, depth<=3) when collecting
axis/wall lines — 黄桥 wraps each frame's grid in anonymous A$C... blocks
(msp had 8 DOTE lines, blocks held 700); all 7 frames went from WEAK
clip-corner fallback to grid-aligned votes x=14
Extraction judgment rules (both loud, both default-on, both switchable):
- drop_isolated_clusters: tiny segment clusters far from the main mass are
sheet-legend content drawn on real layers (title-block window symbols made
every floor's footprint grow a sliver fin)
- drop_unsupported_roof_lines: roof outlines with no footprint/cap plate
within 2m beneath are hatch-only-support canopies that extrude into
floating ribbons (F3 SW 坡屋面, B1 ground canopy at -5.85m)
Tests: 64 -> 75 checks, incl. GBK/MIF synthetic DXFs, block-wrapped grid
alignment (non-uniform bays — uniform grids make subset correlation
ambiguous), legend-cluster and unsupported-roof fixtures.
references/generalization_test_log.md tracks the test protocol, per-project
gate table, and every gap with its regression coverage.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Agent-led CAD-drawing → 3D web showcase pipeline. Inspects CAD reports,
maps layers/floors, extracts a normalized building spec (with block-probed
glazing and audit gates), builds Blender/trimesh GLB variants, and assembles
a Three.js web viewer. Includes XREF-shell detection and a regression test.
Large inputs/outputs (work/, outputs/, source_cad/, renders) are gitignored.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>