Methodology
Built for how banks evaluate vendors
How classifications are defined, validated, and disputed — documented for model-risk and vendor evaluation.
Our approach
- Classifications are factual (“operates sweepstakes wagering”), never accusatory.
- Three orthogonal axes: confidence, severity, and legal status.
- Entities are deprecated, never hard-deleted, with one-level lineage.
- Every pattern declares what weakens it — the false-positive discipline this product owes its subjects.
Validation
Evaluated against a synthetic, labeled-set of persona streams labeled by construction, including clean controls, measured as adversarial-weighted precision and recall per feed and model version.
Synthetic evaluation
precision .9333recall .9618F1 .947[NEEDS: model version]
Real-world validation follows the first pilots.
Dispute process
A documented dispute and descriptor-submission process.
[NEEDS: dispute-resolution SLA]
[NEEDS: link to the full methodology documentation, once cleared for external sharing]