Code and data for: When Do Rank-One Knowledge Edits Merge? A Gain-Screened Two-Regime Law of Edit Federation

Reproducibility archive for the edit-federation operating map: experiment and analysis code (native ROME rank-one editing, merge cross-talk measurement, gain screen, admission rule), the complete per-cell result artifacts and edit vectors for all 22 model-layer cells across seven architecture families (1-20B) plus the editor- and dataset-generality waves, the frozen pre-registration documents, and the figure-generation pipeline. Every table and figure in the accompanying manuscript is reproducible from this archive; the included pre-registration ledger maps each frozen prediction to its outcome. Code is MIT-licensed; results and pre-registration documents are CC BY 4.0. The SHA256SUMS file carries the tarball and member checksums.