Early and Efficient Regime Detection in Bell Test Event Streams with Recursive Indexing

Abstract Large experimental event streams typically require full processing before reliable diagnostic decisions can be made. Recursive Indexing (RIX) offers a complementary approach: organize streams into recursive transport structure and test whether stable scaffold and regime signals emerge early. Across four NIST Bell-test HDF5 streams, RIX recovers a stable 16-basin apparatus-level scaffold and reproducible run families (Milestone 15). These signals stabilize by 1% (scaffold) and 1–2% (family classification) of event rows, delivering a 98.75% average reduction in family-classification diagnostic burden while full 16-basin saturation occurs later (10–50%). Local evidence gating further improves reliability, achieving 100% accuracy on 331 of 368 local samples judged classifiable by abstaining on weak-evidence segments. These results establish RIX as an efficient, evidence-aware framework for staged diagnostics in large scientific event streams.