Tool Defaults and Declared-License Contradictions on the Hugging Face Hub: replication package

Replication package for the article "Tool Defaults and Declared-License Contradictions on the Hugging Face Hub". This is version 2.0.0 of the record; version 1 (doi:10.5281/zenodo.22305960) holds the original snapshot of the Hub retrieved on 3 September 2026, which this version reuses unchanged. The package covers 3,038,885 public model repositories and 884,673 declared derivation links. Under the corrected classification, the baseline rule flags 34,669 of 533,570 decidable links. It includes the raw metadata, the corrected classification, the full-population toolkit analysis, the common-sample channel contrasts, the junction bootstrap, pinned source revisions of the toolkit's model-card templates (unsloth/save.py and the Studio export code), the archived model-card samples, and the manuscript sources. All four archives extract into one directory; readme.txt gives the commands, requirements and checksums. The analysis measures contradictions between declared license metadata, not legal noncompliance. A complete text archive of the 500-card historical calibration is not claimed; the recorded observations, the 160 audit coding-sheet texts and the 230 toolkit sample texts are included. No model weights are included. Compiled metadata and outputs are released under CC BY 4.0 and project code under MIT; third-party material keeps its original rights.