Individuality, Identity and the Logic of Comparison

The principle of individuality sounds grand but its physical basis is mundane. Every manufacturing process operates within tolerances. A drill press does not cut the same thread twice at the atomic scale. A pair of leather shoes from the same production run will accumulate different scratches, wear patterns, and surface irregularities the moment their owners start walking. A block of wood from a tree develops grain patterns during growth that no other block replicates exactly. These deviations are not defects. They are the inevitable product of a physical world where matter moves under variable conditions.

In forensic science the argument runs in two directions. The first is descriptive: objects that are manufactured identically diverge as they age and are used. The second is probabilistic: even before use, the chance that two separate objects share every measurable characteristic decreases rapidly as the number of characteristics measured increases. At some point in the measurement space, two objects that agree on all examined features are effectively unique, or at least the probability of a coincidental match is vanishingly small.

Kirk's claim is an empirical one. It needs to be demonstrated rather than assumed. For some disciplines, like friction-ridge fingerprints, a large body of casework and controlled research supports the claim that no two individuals share the same detailed ridge arrangement. For others, like bite-mark analysis, the claim has been much harder to substantiate, and several wrongful convictions have followed from overconfident identifications in those disciplines.

Not all physical features are equal in the work they can do for a forensic comparison. The discipline divides them into two broad categories, and the distinction determines how far a conclusion can go.

Sub-class characteristics are the trickiest. They look like individual features because they vary from one manufacturing run to another, but within a batch produced by the same die or tool, several items may share the same sub-class marks. An examiner who mistakes sub-class characteristics for individual characteristics will incorrectly individualize an item to a specific source when in fact many items from the same batch share those marks.

A forensic comparison always involves two things: a questioned item whose source is unknown, and a known sample whose source is established. The examiner studies each independently before comparing them, to avoid letting the result of one bias the observation of the other. This two-stage protocol is standard across fingerprints, firearms, documents, footwear, and DNA.

1. Analysis
   The examiner studies the questioned item in isolation, noting all observable characteristics, assessing quality, and determining whether comparison is feasible. If the item is too degraded or partial, the process may stop here.
2. Comparison
   The questioned item is placed alongside the known sample and the examiner systematically checks for agreement and disagreement across all features noted in the analysis phase. The comparison is feature-by-feature, not a holistic impression.
3. Evaluation
   Based on the comparison, the examiner reaches one of several conclusions: individualization (common source), exclusion (different sources), or inconclusive (not enough information to decide). Some disciplines add intermediate categories, such as 'consistent with a common source' or 'cannot be excluded'.
4. Verification
   A second examiner independently reviews the comparison to check the conclusion. This step is required in some disciplines and recommended in others. It catches errors before a report reaches investigators or a court.

The four-step ACE-V framework (Analysis, Comparison, Evaluation, Verification) became explicit in fingerprint science in the 1990s and has since been adopted or adapted by other pattern disciplines. Its value is that it separates observational steps from interpretive ones, making the reasoning more transparent and more reviewable.

The principle of individuality has a philosophical dimension that forensic practitioners sometimes pass over too quickly. The claim 'no two objects are exactly alike' is, strictly speaking, an empirical generalisation about the physical world. It cannot be proved by examining every pair of objects that has ever existed. Instead, it is supported inductively: we have examined many pairs and found them to differ, and we have good physical reasons to expect differences to arise. That is a solid epistemic foundation, but it is different from a mathematical proof.

The statistical side tries to quantify what 'unique' means in practice. If a fingerprint examiner finds that a latent print and an inked exemplar agree on seventeen friction ridge minutiae in the same spatial arrangement, what is the probability that a different finger would produce the same pattern? Studies of large fingerprint databases have attempted to estimate this frequency. The numbers are typically very small, but they are not zero, and the question of how many minutiae are 'enough' to individualize has never been answered with a universally accepted threshold. Different agencies in different countries use different numerical standards, or none at all.

The 2016 report from the President's Council of Advisors on Science and Technology (PCAST) sharpened the NAS critique in a specific and uncomfortable way. It separated two questions that forensic experts had sometimes conflated: (1) whether a feature set in a discipline is sufficient to discriminate between sources at all, which PCAST called foundational validity, and (2) whether examiners in practice apply the method at a known, acceptable error rate, which it called applied validity. A discipline can have the first without having established the second.

• Foundational validity: does the feature set carry enough information to distinguish sources? This is a question about the physical world and can be studied with known-source datasets.
• Applied validity: do human examiners applying the method actually reach correct conclusions at a measured rate? This requires large-scale black-box proficiency studies where examiners work on samples of known ground truth.
• Error rates: every method has both false-positive and false-negative rates. PCAST argued these must be measured and disclosed to courts, not assumed to be near zero.

PCAST reviewed fingerprint, firearms/toolmark, bite-mark, footwear, and hair analysis, as well as DNA. It found that DNA analysis of single-source samples and latent fingerprint analysis were the disciplines with evidence of both foundational and applied validity, though it called for additional black-box studies on fingerprints and noted that the fingerprint false-positive rate was higher than many expected. Firearms and toolmark analysis had some foundational support but limited applied validity data. Bite-mark analysis, it concluded, lacked even foundational validity and should not be used as evidence of identity.

These findings have not shut the disciplines down. Fingerprint evidence is still admitted in courts around the world, and firearms examiners still testify. But the PCAST report has changed the language careful examiners use. The shift is from 'I have made a positive identification' (an absolute uniqueness claim) to 'the features of the questioned item are consistent with originating from the known source, with a false-positive rate in controlled studies of approximately X percent.' That is a probabilistic statement, and it sits closer to what the evidence actually supports.

Because the comparison process can end at several different conclusions, it matters precisely what each conclusion means and how strongly it can be defended. The terminology is not fully standardised across disciplines or jurisdictions, but the major categories are stable and worth knowing.

A court, and especially a jury, will interpret 'consistent with common source' and 'positive identification' very differently, even when the examiner means the former to be a qualified, probabilistic statement. This is one reason several forensic bodies have moved toward likelihood-ratio framing, where the examiner quantifies how much more likely the evidence is under a prosecution hypothesis than under a defence hypothesis, rather than giving a categorical conclusion. The approach forces the probabilistic logic into the open.