Artefacts and scripts for "A pre-registered tail-risk evaluation methodology for variability-aware multilevel MFTJ compute-in-memory"

Analysis artefacts, frozen configuration and scripts for the manuscript "A pre-registered tail-risk evaluation methodology for variability-aware multilevel MFTJ compute-in-memory" (submitted to IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems). SIMULATION - NOT DEVICE EVIDENCE. Every number in this archive comes from a simulator; no device was fabricated and no figure is a measurement. The archive contains the analysis artefacts (JSON/CSV/YAML) that every table and figure of the manuscript reads, the frozen configuration, the CIM runner and mapping policies, and the generators themselves. From the archive alone, scripts/paper_tables.py and scripts/paper_figures.py rebuild the table source and all ten figures; this was verified by extracting the archive into an empty directory and comparing SHA-256 hashes against the submitted versions - they match byte for byte. Deliberately excluded, with reasons in README.md: the per-cell tensor dumps (5.7 GB), the trained checkpoints (223 MB), run logs, and test fixtures. The archive reproduces the analysis exactly; it does not re-run the simulation from scratch, which needs the checkpoints and the training sets.