{
  "context": "Evaluates how well the solution uses numpy-stl to load meshes, apply configurable duplicate handling policies, filter degenerate triangles, and persist cleaned output. Emphasis is on leveraging the library's built-in enums, helpers, and I/O rather than reimplementing geometry logic. Only package-specific usage affecting duplicate and degenerate handling is scored.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Policy mapping",
      "description": "Maps the spec's keep/keep-one/drop-all options to the numpy-stl enum values in stl.mesh.RemoveDuplicates (NONE, SINGLE, ALL) rather than custom flags.",
      "max_score": 20
    },
    {
      "name": "Library deduplication",
      "description": "Invokes mesh.Mesh.remove_duplicate_polygons or constructor/from_file remove_duplicate_polygons=RemoveDuplicates.* to enforce the chosen policy instead of manual duplicate detection.",
      "max_score": 25
    },
    {
      "name": "Degenerate filtering",
      "description": "Uses mesh.Mesh.remove_empty_areas or remove_empty_areas=True during mesh creation to drop zero-area triangles when requested, and disables it cleanly when remove_degenerate=False.",
      "max_score": 20
    },
    {
      "name": "Mesh I/O via package",
      "description": "Loads input meshes with mesh.Mesh.from_file (or BaseStl.load) and writes cleaned meshes with the package's save support rather than manual STL parsing/serialization.",
      "max_score": 20
    },
    {
      "name": "Counts from mesh data",
      "description": "Derives total/kept/removal counts from numpy-stl mesh structures (len(mesh.data), mesh.vectors/normals) instead of maintaining parallel custom structures.",
      "max_score": 15
    }
  ]
}