Reproducibility Materials for "Benchmark Integrity in Three COVID-19 and Mpox Misinformation Datasets: Duplicate Overlap, Cross-Dataset Contamination and Exploratory Label-Quality Signals."

This reproducibility release accompanies the revised manuscript, Benchmark Integrity in Three COVID-19 and Mpox Misinformation Datasets: Duplicate Overlap, Cross-Dataset Contamination and Exploratory Label-Quality Signals.

The archive provides Python implementations and analytical outputs for evaluating exact and near-duplicate content, evaluation-partition overlap, cross-corpus textual contamination, cross-dataset classification transfer, repeated-seed cross-validation and exploratory out-of-fold label-quality indicators across COVID-19-FNIR, Constraint and an mpox veracity-based analytic subset.

The revised release includes five-seed model evaluation, bootstrap uncertainty estimation and a 2,000-replicate group-aware bootstrap analysis that accounts for observations sharing identical normalised texts.

The archive also contains saved predictions, machine-readable statistical results, figure files, software requirements and instructions for reproducing the analyses. Original third-party datasets are not redistributed. Data-source information and input-file checksums are provided to support independent verification.