Reproducibility package for "Attention Is Not All You Need for Diffraction"

Paper-focused reproducibility package for the AI diffraction study. The draft contains a main archival tarball together with a compact reviewer-assets archive for the Jupyter notebook walkthrough. The main archive contains the final-stage ViT checkpoints used in the main paper narrative, the retained supplemental VT/RT/CNN checkpoints that back the current supplemental figures and tables, the training and evaluation configs and launchers needed to interpret those checkpoints, compact JSON result artifacts used to support the benchmark/topology/positional-ablation claims, and short provenance notes and manifests.
The release does not redistribute the full RRUFF benchmark HDF5s or raw RRUFF mirrors; instead it documents a reconstruction-oriented workflow in the companion repository at github.com/scattering/paper-ai-diffraction.