Reproducibility package for "When Localized Causal Effects Do Not Imply Functional Regions in Spatial Recurrent Neural Networks"

Reproducibility package: event-local RMT assay outputs for seeds 42-51 (8 validation batches, 1,000 null draws, four validation data seeds), window-specificity controls, per-seed resolved configs and analysis files, assay logs, leave-one-seed-out predictor audit and split table, code snapshot, and the frozen analysis rules. NE definitions and both conditional and full-flow denominators are documented in Analysis_Freeze_v2.md.