← hub·evidence

Source classification, spotcheck

Complete source archives, one per bucket, so you can judge the lenient classifier: real authored projects vs single-file uploads vs library dumps vs machine-generated SDK garbage.

Lenient rule under review: vendored = a file inside a dependency dir (vendor/, node_modules/, site-packages/) only, NOT by cross-repo recurrence. Each repo is kept WHOLE. Files inside vendored dirs are tagged [vendored], machine-gen headers [sdk-dump]. Read a whole repo and judge: is it a real authored project, a lone script, a library dump, or SDK garbage? Buckets scale to ~188k clean / ~106k single-file / ~18k library / ~5k SDK across the lake.

collapse all expand all sort:

select a file

Pick a repo on the left, then read any file in it, top to bottom.