[ajavanma]

PCW now uses PointCloud Core for array validation, class mapping, voxel grouping and label voting, and prediction lifting. A small compatibility adapter retains PCW's error context, artifact schemas, and divide-before-sum feature means so existing prepared data and checkpoints keep their numerical contract.

Core 0.1.0 is installed from an immutable source revision with a SHA-256 hash while PyPI publication is pending. The PCW publishing workflow rejects source URL dependencies until they are replaced with tested index releases. A standalone synthetic example runs with only Core and NumPy; documentation explains the integration and numerical boundary.

Validation:

• 212 local tests passed: 125 unit, 32 CLI, 49 model, and 6 legacy artifact/runtime compatibility tests.
• All 206 unit/CLI/model tests passed against the built wheel outside the checkout; dependency checks passed.
• All five preprocessing arrays were byte-identical to pre-integration PCW for 128 seeded scenes.
• New compatibility tests and the standalone example passed on NumPy 1.26.4 and 2.4.6.
• Source/wheel builds, strict Twine checks, archive inspection, and strict MkDocs build passed.

CI additionally runs the installed-wheel adapter tests, Core-only example, and CPU container workflow. No GPU host execution is part of this change.