# GCORF-01 Extraction Protocol v0.1

1. Normalize source identifiers without rewriting source meaning.
2. Extract Evidence Units before any person-level profile is generated.
3. Record observability and evidence grade separately.
4. Build provenance dependency graph and trace graph.
5. Cluster traces by transformation structure, not by famous name.
6. Compress clusters into candidate kernels.
7. Atomicize relative to the current operator grammar.
8. Search existing operator library before creating a new atomic operator.
9. Run transfer and multi-observer tests.
10. Preserve negative evidence, unknowns, rejected branches, and observer disagreement.