How Burr Works

Burr is design-rule checking for CAD-as-code.

The basic loop is:

design.py
  -> generated STEP
  -> burr-design-data.json
  -> burr check
  -> burr-receipt.json
  -> burr explain

The CAD source is still normal build123d. The difference is that important mechanical features use Burr helpers instead of anonymous raw cuts.

Python Source

from build123d import Box, BuildPart, Locations, export_step
from burr_build123d import BurrDesignData, DESIGN_DATA_FILE, m3_clearance_hole

design = BurrDesignData(
    artifact_id="actuator-mount",
    artifact_type="actuator_mount",
    units="mm",
    process={"kind": "FDM", "material": "PETG", "nozzle_mm": 0.4},
)

design.source("design.py")
design.artifact("actuator.step")
design.part("housing", bbox_min=(-42, -16, 0), bbox_max=(42, 16, 26))

with BuildPart() as housing:
    with Locations((0, 0, 13)):
        Box(84, 32, 26)

    # This both cuts the CAD hole and records Burr metadata.
    m3_clearance_hole(
        design,
        feature_id="m3_lower_left",
        part="housing",
        center=(39.5, -8, 8),
        axis=(1, 0, 0),
        role="loaded_mount",
    )

export_step(housing.part, "actuator.step")
design.write(DESIGN_DATA_FILE)

The helper creates CAD geometry and records design intent into burr-design-data.json.

Counterbore helpers record both the smaller bore and the larger head recess, so Burr can check STEP presence and edge material around the recess itself. Bearing-seat helpers record the seat diameter and shoulder, so Burr can check that the bearing seat exists in STEP and still has enough host material around the loaded support envelope.

CLI Loop

uv run python design.py
burr check .
burr explain .

burr check . reads burr-design-data.json, checks source and artifact freshness, runs the selected rulepack, and writes burr-receipt.json.

burr explain . reads the receipt and returns human-readable repair guidance. Agents can use burr explain --json . for structured repair packets.

Boundary

Burr is not image verification, FEA, or general STEP understanding. It is unit-test-style design-rule checking over declared CAD intent, with measurable receipts that humans and agents can use to repair generated CAD.