# Reference Coupling Reuse Audit

Phase A reused existing light/cheap CSV outputs. Phase B regenerated only missing row-level future-only/current-state columns from the same seed/sample.

- seed: `1`
- sample: discovery `5000`, validation `5000`
- new sample selection: `false`
- result diff mismatches against existing seed1 original trends: `0`
- overall final label: `persists_beyond_shared_reference`