THREATBYTESTM research
Reproducibility
Reproducibility requires versioned datasets, experiment configurations, model artefacts, container definitions, and published processing scripts when licensing permits.
Artefacts01
Platform version, dataset version, sandbox image digest, and experiment identifiers appear together on every published result summary.
Reproducibility checklist02
- Specimen hash lists for every dataset version.
- Feature extraction code and version.
- Training configuration, seeds, and hardware notes.
- Model artefacts where licensing and safety allow.
- Evaluation scripts and the exact data splits.
Constraints03
Live malware is not redistributed. Datasets are published as hash lists and derived features, so other researchers can obtain the same specimens from the original repositories under those repositories' terms.