Fingerprints are unique. This has been the foundational assumption of fingerprint identification for over a century: no two fingers, even on the same person, produce the same ridge pattern. Identical twins have different fingerprints. The individuality is established at the level of minutiae — ridge endings, bifurcations, dots — which are positioned by stochastic developmental noise during fetal growth.
The macro-geometry is not unique. The overall shape of the ridge flow — the curvature, the orientation field, the positions of cores and deltas — is shared across fingers of the same person. An AI system trained on fingerprint images can match prints from different fingers of the same individual with 77% accuracy. The system identifies the person, not the finger.
This contradicts nothing about minutiae-based uniqueness. The minutiae remain individual to each finger. But the large-scale pattern — the forest, not the trees — carries a signal that is person-specific rather than finger-specific. The ridge orientation field appears to be influenced by the geometry of the fingertip, the volar pad shape, and the timing of ridge formation, all of which are shared across digits of the same hand to some degree.
The forensic implication is direct: a print from one finger can be linked to a print from a different finger of the same person, without ever matching minutiae. This was previously considered impossible. The macro-geometry was treated as the classification system (arch, loop, whorl) but not as an identification feature.
The individuality of fingerprints was always true at the scale used for identification. At a different scale — coarser, geometric, developmental — the prints carry a different kind of information: not which finger, but whose hand. The same data, at different resolutions, answers different questions.