Open-source research preview
Forensic marks, weighed as evidence.
Compare real specimens — or upload X3P scans — and get an auditable likelihood ratio, uncertainty, and a map of the regions that drove it. Never a black-box “match.”
Verity reports the weight of evidence. It never makes the decision, and it does not claim an examination error rate.
No file needed for the example · Read the method
Validation at a glance
See the methodology- Mark types, one pipeline
- 3
- Source-disjoint AUC
- 0.95+
- Reference pairs
- 170,001
- Per-modality re-tuning
- 0
Compare two marks
Choose two real specimens — no upload required — and follow the evidence from matched regions to the reference population to a calibrated likelihood ratio. Every example is a real, precomputed comparison across bullets, cartridge cases, and toolmarks.
Comparison updated for Hamby-252 · barrel 1 · bullet 1 and Hamby-252 · barrel 1 · bullet 2. moderately strong support for same source. Likelihood ratio 146.
Hamby-252 · barrel 1 · bullet 1vsHamby-252 · barrel 1 · bullet 2
Curated specimen — precomputed real comparison (Hamby-252, same-source)Likelihood ratio
146
The observed comparison is 146× more probable if the marks share a source than if they do not.
Reference & calibration details
This is a calibrated weight of evidence on the pooled bullet-land reference population. It is not a verdict: it is one input to an examiner's judgment, alongside case context. It is not a claim about the error rate of striated examination, which remains unknown. The reported LR is bound-limited: the calibrated value exceeded the empirical bound this reference can support and was capped at it.
Forensic mark comparison is used in criminal courts yet has no characterized error rate. Verity returns a calibrated likelihood ratio with a quantified cost and region-level attribution — one method across striated, impressed, and toolmark marks.
Why this exists