Open methods. Defensible math.
Published methodology, the 111-case fair-use dataset, TQ foundations, the evidentiary hierarchy framework, and 70+ cited methods.
The 111-case fair-use dataset.
Curated, validated, and reproducible. 53 fair-use rulings and 58 non-fair-use rulings, with four-factor coding and post-Warhol re-classification. Downloadable as JSONL with attribution.
Transformation Quotient — full derivation.
11-dimensional feature space: 7 statistical + 4 neural. BCa bias-corrected bootstrap CIs. Adversarial perturbation budget. Reproducible from seed and parameter hashes.
Tier mapping & Daubert decomposition.
SaTML 2025 tier mapping. Daubert factor decomposition. Jurisdiction-threshold table for US (circuit-by- circuit), EU, UK, DE, JP, KR.
70+ academic papers.
- Min-K%++Zhang et al., ICLR 2025
- DE-COPDuarte et al., arXiv 2402.09910
- Cooper PDECooper et al., 2024
- nv-recallStanford Algorithm 1
- TRAKPark et al., ICML 2023
- CSD styleSomepalli et al., NeurIPS 2023
- SSCDPizzi et al., CVPR 2022
- DINOv2Oquab et al., 2023
- D-TRAKZheng et al., 2024
- Collective Dataset InferenceCVPR 2025
- MUSEShi et al., 2024
- MelodySimInternal — based on Smith et al.
Three redacted PDFs.
Open call for research collaborations.
If you're working on attribution, MIA, fair-use NLP, or diffusion forensics — talk to us.
Open methods. Defensible math.
Every detector is published. Every score is reproducible from the chain-of-custody hash.