Files
cad-building-showcase/scripts/audit_showcase_extraction.py
zhengwr a6ac166049 Initial commit: cad-building-showcase skill
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>
2026-06-12 16:16:04 +08:00

506 lines
21 KiB
Python
Executable File

#!/usr/bin/env python3
"""Audit extraction coverage so the agent can judge what may be missing."""
from __future__ import annotations
import argparse
import json
import math
import sys
from collections import Counter, defaultdict
from pathlib import Path
from typing import Any
sys.path.insert(0, str(Path(__file__).resolve().parent))
import derive_showcase_spec as derive # noqa: E402
HIGH_RISK_TYPES = {
"INSERT",
"HATCH",
"SOLID",
"3DFACE",
"REGION",
"ACAD_PROXY_ENTITY",
"DIMENSION",
"TEXT",
"MTEXT",
"IMAGE",
"WIPEOUT",
}
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--dxf-dir", required=True, help="Sanitized DXF folder")
parser.add_argument("--config", required=True, help="Extraction config JSON")
parser.add_argument("--spec", required=True, help="Generated building_spec.json")
parser.add_argument("--out", default="work/cad_showcase/extraction_audit", help="Output path stem or folder")
parser.add_argument("--vendor-dir", default="work/vendor")
parser.add_argument(
"--no-strict-gates",
action="store_true",
help="Report gate failures without a nonzero exit code",
)
return parser.parse_args()
def iter_block_contents(entity, max_depth: int, depth: int = 0):
"""Yield every leaf entity inside an INSERT, regardless of extraction config."""
if depth >= max_depth:
return
try:
from ezdxf import explode as ezdxf_explode
virtual_entities = list(ezdxf_explode.virtual_block_reference_entities(entity))
except Exception:
return
for virtual_entity in virtual_entities:
if virtual_entity.dxftype() == "INSERT":
yield from iter_block_contents(virtual_entity, max_depth, depth + 1)
else:
yield virtual_entity
def output_paths(output: Path) -> tuple[Path, Path]:
if output.suffix:
return output.with_suffix(".json"), output.with_suffix(".md")
output.mkdir(parents=True, exist_ok=True)
return output / "extraction_audit.json", output / "extraction_audit.md"
def segment_length_source(segment: tuple[tuple[float, float], tuple[float, float]]) -> float:
(sx, sy), (ex, ey) = segment
return math.hypot(ex - sx, ey - sy)
def inspect_entity(
entity,
patterns: dict[str, list[str]],
config: dict[str, Any],
max_chord: float,
fallback_layer: str = "",
block_name: str = "",
) -> dict[str, Any]:
layer = derive.entity_layer(entity)
dxftype = entity.dxftype()
category = derive.category_for_entity(
entity,
patterns,
config,
fallback_layer=fallback_layer,
block_name=block_name,
)
segments = derive.entity_segments(entity, max_chord)
supported_length = sum(segment_length_source(segment) for segment in segments)
return {
"layer": layer,
"type": dxftype,
"category": category,
"segment_count": len(segments),
"supported_length": supported_length,
"high_risk": dxftype in HIGH_RISK_TYPES,
"insert_name": block_name or (str(entity.dxf.name) if dxftype == "INSERT" and entity.dxf.hasattr("name") else ""),
}
def source_files_from_config(dxf_dir: Path, config: dict[str, Any]) -> list[Path]:
drawings = config.get("drawings")
if drawings:
return [dxf_dir / drawing["file"] for drawing in drawings if (dxf_dir / drawing["file"]).exists()]
return sorted(dxf_dir.glob("*.dxf"))
def spec_counts(spec: dict[str, Any]) -> dict[str, dict[str, int]]:
categories = [key for key in spec.get("categories", {}) if key != "slab"]
result = {}
for floor_key, floor in spec.get("floors", {}).items():
result[floor_key] = {
category: len(floor.get("segments", {}).get(category, []))
for category in categories
}
return result
def audit(dxf_dir: Path, config: dict[str, Any], spec: dict[str, Any], vendor_dir: Path) -> dict[str, Any]:
ezdxf, recover = derive.import_ezdxf(vendor_dir)
patterns = derive.layer_patterns(config)
max_chord = float(config.get("curve_chord_source_length", 750.0))
scale = float(config.get("scale", config.get("scale_to_meters", 0.001)))
file_reports = []
global_entity_types = Counter()
global_layer_counts = Counter()
global_high_risk = Counter()
global_insert_names = Counter()
layer_lengths: dict[str, dict[str, Any]] = defaultdict(
lambda: {
"entity_count": 0,
"supported_length_source": 0.0,
"classified_length_source": 0.0,
"categories": Counter(),
"entity_types": Counter(),
"high_risk_types": Counter(),
}
)
total_supported = 0.0
classified_supported = 0.0
captured_from_blocks = 0.0
excludes = derive.exclude_layer_patterns(config)
max_block_depth = int(config.get("expand_blocks_max_depth", 2))
block_layer_lengths: dict[str, dict[str, Any]] = defaultdict(
lambda: {"supported_length_source": 0.0, "entity_count": 0, "category": None}
)
for path in source_files_from_config(dxf_dir, config):
doc = derive.read_doc(path, ezdxf, recover)
file_total_supported = 0.0
file_classified_supported = 0.0
file_entity_types = Counter()
file_high_risk = Counter()
file_layers = Counter()
file_insert_names = Counter()
for source_entity in doc.modelspace():
if source_entity.dxftype() == "INSERT":
insert_name = str(source_entity.dxf.name) if source_entity.dxf.hasattr("name") else ""
global_high_risk["INSERT"] += 1
file_high_risk["INSERT"] += 1
if insert_name:
global_insert_names[insert_name] += 1
file_insert_names[insert_name] += 1
# Config-independent accounting of geometry locked inside blocks,
# so missing block expansion cannot hide behind a high coverage number.
for inner in iter_block_contents(source_entity, max_block_depth):
inner_length = sum(
segment_length_source(segment)
for segment in derive.entity_segments(inner, max_chord)
)
if inner_length <= 0:
continue
inner_layer = derive.entity_layer(inner)
record = block_layer_lengths[inner_layer]
record["supported_length_source"] += inner_length
record["entity_count"] += 1
if record["category"] is None:
record["category"] = derive.match_category(inner_layer, patterns, excludes) or ""
for entity, fallback_layer, block_name in derive.iter_model_entities(source_entity, config):
item = inspect_entity(entity, patterns, config, max_chord, fallback_layer, block_name)
layer = item["layer"]
dxftype = item["type"]
length = float(item["supported_length"])
category = item["category"]
global_entity_types[dxftype] += 1
global_layer_counts[layer] += 1
file_entity_types[dxftype] += 1
file_layers[layer] += 1
layer_lengths[layer]["entity_count"] += 1
layer_lengths[layer]["supported_length_source"] += length
layer_lengths[layer]["entity_types"][dxftype] += 1
if category:
layer_lengths[layer]["categories"][category] += 1
layer_lengths[layer]["classified_length_source"] += length
classified_supported += length
file_classified_supported += length
if block_name:
captured_from_blocks += length
total_supported += length
file_total_supported += length
if item["high_risk"]:
# Top-level INSERTs were already counted above.
if not (dxftype == "INSERT" and entity is source_entity):
global_high_risk[dxftype] += 1
file_high_risk[dxftype] += 1
layer_lengths[layer]["high_risk_types"][dxftype] += 1
if item["insert_name"] and entity is not source_entity:
global_insert_names[item["insert_name"]] += 1
file_insert_names[item["insert_name"]] += 1
file_reports.append(
{
"file": path.name,
"entity_types": dict(file_entity_types.most_common()),
"top_layers": dict(file_layers.most_common(20)),
"high_risk_types": dict(file_high_risk.most_common()),
"insert_names": dict(file_insert_names.most_common(20)),
"supported_length_meters": round(file_total_supported * scale, 3),
"classified_supported_length_meters": round(file_classified_supported * scale, 3),
"supported_length_coverage": round(file_classified_supported / file_total_supported, 4)
if file_total_supported
else None,
}
)
unclassified_layers = []
classified_layers = []
for layer, data in layer_lengths.items():
record = {
"layer": layer,
"entity_count": data["entity_count"],
"supported_length_meters": round(data["supported_length_source"] * scale, 3),
"classified_length_meters": round(data["classified_length_source"] * scale, 3),
"categories": dict(data["categories"].most_common()),
"entity_types": dict(data["entity_types"].most_common(10)),
"high_risk_types": dict(data["high_risk_types"].most_common()),
}
if data["classified_length_source"] > 0:
classified_layers.append(record)
else:
unclassified_layers.append(record)
unclassified_layers.sort(key=lambda item: (-item["supported_length_meters"], -item["entity_count"], item["layer"]))
classified_layers.sort(key=lambda item: (-item["classified_length_meters"], item["layer"]))
block_locked_layers = []
block_locked_total = 0.0
block_locked_classifiable = 0.0
for layer, data in block_layer_lengths.items():
length = float(data["supported_length_source"])
block_locked_total += length
if data["category"]:
block_locked_classifiable += length
block_locked_layers.append(
{
"layer": layer,
"category": data["category"] or None,
"entity_count": data["entity_count"],
"supported_length_meters": round(length * scale, 3),
}
)
block_locked_layers.sort(key=lambda item: -item["supported_length_meters"])
report = {
"dxf_dir": str(dxf_dir),
"coverage": {
"supported_length_meters": round(total_supported * scale, 3),
"classified_supported_length_meters": round(classified_supported * scale, 3),
"supported_length_coverage": round(classified_supported / total_supported, 4)
if total_supported
else None,
"block_locked_length_meters": round(block_locked_total * scale, 3),
"block_locked_classifiable_length_meters": round(block_locked_classifiable * scale, 3),
"block_captured_classified_length_meters": round(captured_from_blocks * scale, 3),
},
"spec_counts": spec_counts(spec),
"files": file_reports,
"classified_layers": classified_layers,
"unclassified_layers": unclassified_layers[:80],
"block_locked_layers": block_locked_layers[:40],
"high_risk_entity_types": dict(global_high_risk.most_common()),
"top_insert_names": dict(global_insert_names.most_common(50)),
"entity_types": dict(global_entity_types.most_common()),
"top_layers": dict(global_layer_counts.most_common(50)),
"agent_questions": agent_questions(unclassified_layers, global_high_risk, global_insert_names),
}
report["gates"] = evaluate_gates(report)
return report
GATE_GLAZING_MIN_METERS = 5.0
def evaluate_gates(report: dict[str, Any]) -> list[dict[str, Any]]:
"""Hard checks that must pass before 3D generation.
A gate FAIL means the extraction is known-incomplete in a way that high
coverage numbers cannot excuse; fix the config (block expansion, layer
mapping, source re-export) and re-run instead of accepting the model.
"""
gates: list[dict[str, Any]] = []
spec_glazing_segments = sum(
counts.get("glazing", 0) for counts in report["spec_counts"].values()
)
locked_glazing_meters = sum(
item["supported_length_meters"]
for item in report["block_locked_layers"]
if item["category"] == "glazing"
)
glazing_named_blocks = [
name
for name in report["top_insert_names"]
if derive.match_patterns(name, derive.GLAZING_BLOCK_PATTERNS)
]
glazing_evidence = locked_glazing_meters >= GATE_GLAZING_MIN_METERS or bool(glazing_named_blocks)
if not glazing_evidence:
status, detail = "SKIP", "No curtain-wall/glazing evidence in blocks or block names."
elif spec_glazing_segments > 0:
status = "PASS"
detail = (
f"Spec contains {spec_glazing_segments} glazing segments; "
f"~{round(locked_glazing_meters, 1)} m of glazing-layer geometry found inside blocks."
)
else:
status = "FAIL"
detail = (
f"Glazing evidence exists (~{round(locked_glazing_meters, 1)} m on glazing layers inside blocks; "
f"block names: {', '.join(glazing_named_blocks[:3]) or 'none'}) but the spec has 0 glazing segments. "
"Enable block expansion / classify_block_contents or fix glazing layer mapping, then re-run."
)
gates.append({"name": "glazing_captured", "status": status, "detail": detail})
annotation_classified = [
item["layer"]
for item in report["classified_layers"]
if derive.match_patterns(item["layer"], ["*-IDEN", "*ANNO*", "*-DIMS"])
]
gates.append(
{
"name": "no_annotation_layers_classified",
"status": "WARN" if annotation_classified else "PASS",
"detail": (
f"Annotation/identification layers classified as geometry: {', '.join(annotation_classified[:5])}. "
"These are tags, not elements; exclude them."
if annotation_classified
else "No annotation/identification layers classified."
),
}
)
coverage = report["coverage"]
locked_classifiable = float(coverage["block_locked_classifiable_length_meters"])
captured = float(coverage["block_captured_classified_length_meters"])
if locked_classifiable > 0 and captured < 0.5 * locked_classifiable:
gates.append(
{
"name": "block_locked_geometry_captured",
"status": "WARN",
"detail": (
f"~{round(locked_classifiable, 1)} m of classifiable geometry sits inside blocks but only "
f"~{round(captured, 1)} m was captured. Review block expansion settings."
),
}
)
else:
gates.append(
{
"name": "block_locked_geometry_captured",
"status": "PASS",
"detail": (
f"Captured ~{round(captured, 1)} m of ~{round(locked_classifiable, 1)} m "
"classifiable block-locked geometry."
),
}
)
return gates
def agent_questions(unclassified_layers: list[dict], high_risk: Counter, insert_names: Counter) -> list[str]:
questions = []
if unclassified_layers:
top = [item for item in unclassified_layers if item["supported_length_meters"] > 0][:8]
if top:
names = ", ".join(f"`{item['layer']}`" for item in top[:5])
questions.append(f"Review unclassified geometry layers with significant length: {names}.")
if high_risk.get("INSERT", 0):
questions.append("Decide whether INSERT/block geometry must be recursively expanded before modeling.")
if high_risk.get("HATCH", 0):
questions.append("Check whether hatches represent walls/slabs/columns that line extraction missed.")
if insert_names:
names = ", ".join(f"`{name}`" for name, _count in insert_names.most_common(5))
questions.append(f"Inspect frequent block names: {names}.")
questions.append("Compare SVG overlays against CAD intent before accepting the model.")
return questions
def markdown(report: dict[str, Any]) -> str:
coverage = report["coverage"]
lines = [
"# Extraction Audit",
"",
"Use this as evidence for Agent judgment. High coverage does not prove correctness; low coverage or high-risk entities require review.",
"",
"## Gates",
"",
]
for gate in report.get("gates", []):
lines.append(f"- **{gate['status']}** `{gate['name']}`: {gate['detail']}")
lines.extend(
[
"",
"## Coverage",
"",
f"- Supported vector length: `{coverage['supported_length_meters']}` m",
f"- Classified supported length: `{coverage['classified_supported_length_meters']}` m",
f"- Supported-length coverage: `{coverage['supported_length_coverage']}` "
"(modelspace linework only — block-locked geometry is NOT in this denominator)",
f"- Block-locked geometry: `{coverage['block_locked_length_meters']}` m total, "
f"`{coverage['block_locked_classifiable_length_meters']}` m classifiable, "
f"`{coverage['block_captured_classified_length_meters']}` m captured into the spec",
"",
"## Agent Questions",
"",
]
)
for question in report["agent_questions"]:
lines.append(f"- {question}")
lines.extend(["", "## Block-Locked Geometry By Layer", ""])
for item in report.get("block_locked_layers", [])[:15]:
lines.append(
f"- `{item['layer']}` (category: {item['category'] or 'unclassified'}): "
f"entities `{item['entity_count']}`, length `{item['supported_length_meters']}` m"
)
lines.extend(["", "## Spec Counts", ""])
for floor_key, counts in report["spec_counts"].items():
parts = ", ".join(f"{key}: {value}" for key, value in counts.items())
lines.append(f"- `{floor_key}`: {parts}")
lines.extend(["", "## High Risk Entity Types", ""])
for key, value in report["high_risk_entity_types"].items():
lines.append(f"- `{key}`: {value}")
lines.extend(["", "## Top Unclassified Geometry Layers", ""])
for layer in report["unclassified_layers"][:20]:
lines.append(
f"- `{layer['layer']}`: entities `{layer['entity_count']}`, "
f"supported length `{layer['supported_length_meters']}` m, "
f"types `{layer['entity_types']}`"
)
lines.extend(["", "## Classified Layers", ""])
for layer in report["classified_layers"][:30]:
lines.append(
f"- `{layer['layer']}`: `{layer['categories']}`, "
f"classified length `{layer['classified_length_meters']}` m"
)
lines.extend(["", "## Files", ""])
for item in report["files"]:
lines.append(
f"- `{item['file']}`: coverage `{item['supported_length_coverage']}`, "
f"high-risk `{item['high_risk_types']}`"
)
return "\n".join(lines)
def main() -> int:
args = parse_args()
config = derive.read_json(Path(args.config))
spec = json.loads(Path(args.spec).read_text(encoding="utf-8"))
report = audit(Path(args.dxf_dir), config, spec, Path(args.vendor_dir))
json_path, md_path = output_paths(Path(args.out))
json_path.parent.mkdir(parents=True, exist_ok=True)
md_path.parent.mkdir(parents=True, exist_ok=True)
json_path.write_text(json.dumps(report, ensure_ascii=False, indent=2), encoding="utf-8")
md_path.write_text(markdown(report), encoding="utf-8")
failed_gates = [gate for gate in report.get("gates", []) if gate["status"] == "FAIL"]
print(
json.dumps(
{
"json": str(json_path),
"markdown": str(md_path),
"gates": {gate["name"]: gate["status"] for gate in report.get("gates", [])},
},
ensure_ascii=False,
)
)
if failed_gates and not args.no_strict_gates:
for gate in failed_gates:
print(f"GATE FAIL [{gate['name']}]: {gate['detail']}", file=sys.stderr)
return 3
return 0
if __name__ == "__main__":
raise SystemExit(main())