Causal Preservation Under Substrate Transfer: Interchange Intervention Tests for Source-Specific Functional Continuity

This preprint adds a causal-intervention layer to PreservationBench-AST. Earlier PreservationBench papers showed that copied provenance, report continuity, task competence, proxy reward, source-like control, recurrent history dependence, and perturbation response can dissociate after substrate transfer. This paper asks a sharper question: when a target resembles a source from the outside, does it preserve the source's behaviorally relevant causal mechanisms, or only match selected behavioral and report surfaces? The paper introduces causal-preservation tests using interchange interventions. For paired source histories A and B, source activation interventions are first selected for source-side causal leverage. Source-B schema-state or attention-state activations are then patched into source or target context A, and the patched target is compared with the source's own counterfactual response. The bounded conclusion is that causal intervention tests can separate source-specific mechanism preservation from ordinary behavior matching in this toy benchmark. The result does not measure consciousness, personal identity, survival, biological preservation, whole-agent equivalence, or whether Attention Schema Theory is true.