Turn floor plan drawings into geometry your application can measure.

DWG / DXF
     ↓
sheets         one per layout — extents, area, perimeter, preview
     ↓
layers         names, colours, per-layer entity counts
     ↓
entities       typed geometry with coordinates in drawing units
     ↓
your app       measurement, routing, indexing, display

Nothing in that chain is inferred. Each level is a structure that already exists in the drawing, made explicit and queryable. The bundled sample shows exactly this shape for a real DWG file — 4 layers, 2,096 entities, extents in mm.

Every entity arrives in the same envelope regardless of type, so your code has one shape to handle:

• geometry — spatial data only, at the original CAD precision. Vertices for polylines; centre, radius and angles for arcs; insertion point and rotation for text.
• bbox — minX, minY, maxX, maxY. Always present, so spatial indexing and hit-testing need no preprocessing pass.
• metrics — length, area, perimeter, vertexCount, with null where a value does not apply rather than a misleading approximation.
• layer, category, properties — layer name, one of Geometry / Annotation / BlockReference / Hatch / Other, and display attributes such as colour index and line type.

Sheet-level boundingBox, area, and perimeter come back too, which is often all a floor-area estimate needs.

CADLens deliberately does not guess what a polyline represents. What it does give you is the layer structure the draughtsman used, with per-layer entity counts, so you can route geometry by the conventions your industry already follows — AIA layer names, a client's in-house standard, or whatever your customers actually send.

That boundary matters for the same reason type safety does: a parser that quietly labelled a line "wall" would be wrong on some fraction of drawings and you would have no way to tell which. A layer name is a fact from the file. Your mapping from layer names to meaning is a decision you can test, version, and correct.

Plenty of tools convert CAD to PNG or PDF. That is genuinely useful for display, and CADLens renders a preview per sheet for exactly that reason. But a raster image cannot tell you which polyline is on A-WALL, how long it is in millimetres, or how many door blocks sit on level two. Those questions need entities.

The distinction is the whole point of the product: CAD → PNG is a rendering problem, while CAD → entities / layers / geometry / metrics is a data problem. See DWG to image vs DWG to JSON for the longer comparison.