DNA Evidence Credential Authority Failure Through Probabilistic Genotyping Likelihood Ratio Accepted Without Independent Evaluability of the Algorithm That Produced It

Probabilistic genotyping software analyzes complex DNA mixtures — samples containing genetic material from multiple contributors — and produces a likelihood ratio certifying the statistical probability that a named individual contributed to the mixture. The traditional random match probability method requires a clean, single-source sample. Probabilistic genotyping extends DNA analysis to degraded, low-quantity, and mixed samples where traditional methods cannot produce a definitive profile. The likelihood ratio it produces carries significant evidentiary weight: ratios in the millions or billions are regularly presented to juries as near-certain identifications.

The leading probabilistic genotyping systems — TrueAllele, developed by Cybergenetics, and STRmix — are proprietary commercial products. Their source code is withheld as a trade secret. The algorithms that produce likelihood ratios are not publicly available for independent validation. When academic researchers at the University of California attempted to purchase a research license to study TrueAllele, Cybergenetics declined, stating it does not provide research licenses. Representatives of the company have testified under oath in courts across the country that the product is subject to peer review — while using contract law to prevent the independent research that peer review requires.

The credential gap in probabilistic genotyping has been documented across multiple jurisdictions and oversight bodies. The GAO identified a lack of independent review of these tools, noting that most validation studies have been conducted by software developers themselves or by law enforcement agencies. The President's Council of Advisors on Science and Technology concluded that further studies should be performed by independent research groups with no stake in the outcome. NIST observed that for most peer-reviewed articles describing validation experiments, sufficient data are not publicly available to allow independent replication.

In the most consequential judicial confrontation with the credential gap, a federal judge ordered the New York City Office of the Chief Medical Examiner to disclose the source code for its probabilistic genotyping software. When the code was reviewed, accuracy concerns came to light. The software was discontinued. A subsequent state ruling found that prior reliance on the software had been an error and recommended that convictions relying on it be reviewed. The judge noted the software was a "black box" that no independent expert had been given the opportunity to examine — and that error rates for samples with four or more contributors exceeded 50 percent.

The Third Circuit ruled in April 2026 that TrueAllele was admissible without source code disclosure, holding that validation studies and known error rates were sufficient under Federal Rule of Evidence 702. The ruling establishes that a credential whose evidentiary boundary cannot be independently evaluated is nevertheless sufficient to authorize conviction in federal court — provided the developer asserts it has been validated.

The structural condition this case documents is not the absence of an encoded evidentiary boundary. It is the active withholding of the mechanism that produced the credential. The algorithm produces a likelihood ratio. The likelihood ratio asserts a probability of correspondence. The mechanism that produced the assertion is a trade secret. The defendant cannot examine it. The defense expert cannot validate it. The court cannot verify it. The relying party — the jury — receives the credential and is told by its developer that it has been validated, by studies the developer conducted, using data the developer has not made available for independent replication.

The credential resolves the correspondence question in the prosecution's favor. The mechanism that produced the resolution is controlled by the party with the financial interest in its acceptance. This is not the common credential model — possession equals legitimacy, issuance equals ongoing truth, form equals verification. This is that model with an additional structural feature: the developer actively prevents the verification that would challenge the credential's authority.

Probabilistic genotyping produces a credential whose evidentiary boundary is not evaluable at the point of reliance — and requires trusting the issuer because the issuer has structured the system to prevent any alternative. The credential resolves the correspondence question. The mechanism that produced the resolution is withheld. The relying party has no alternative to trusting the developer. A credential whose evidentiary boundary can be evaluated at the point of reliance — without requiring institutional declaration or trusting the issuer — is the structural correction this condition implies.

Courts have largely admitted probabilistic genotyping evidence under Daubert and Federal Rule of Evidence 702, treating developer-conducted validation studies as sufficient. The Third Circuit's April 2026 ruling signals that source code disclosure will not be required as a condition of admission in federal courts that follow its reasoning. Defense efforts to force disclosure through evidentiary motions face significant headwinds. Senate Judiciary Committee hearings in January 2025 examined the legal and ethical implications without producing legislative resolution.

Open-source probabilistic genotyping software has been developed by independent researchers — demonstrating that the transparency condition is technically achievable. The GAO, PCAST, and NIST have all recommended independent validation. None of those recommendations have been implemented as mandatory requirements. The credential continues to authorize conviction in courts that have accepted the developer's assertion of validity as a substitute for independent verification of the evidentiary boundary.

• The probabilistic genotyping likelihood ratio is an evidentiary credential — it certifies a probability of DNA correspondence that courts and juries rely upon to authorize conviction; the evidentiary boundary of the credential — what the algorithm verified, what assumptions it made, what error rates apply to the specific sample — is not available to the defendant, the court, or the jury at the point of reliance
• The credential gap in this case is not the absence of an encoded evidentiary boundary — it is the active withholding of the mechanism that produced the credential, by the party with the financial interest in its acceptance, under trade secret protection
• The GAO, PCAST, and NIST have independently identified the lack of independent review as a systemic condition; most validation studies have been conducted by software developers or law enforcement agencies; independent researchers have been denied licenses to study the systems; the credential asserts validity that has not been independently established
• Documented consequences include wrongful convictions upheld by flawed likelihood ratio calculations, exonerations secured only after competing software produced different results, discontinued software following court-ordered source code disclosure that revealed accuracy concerns, and a 2019 judicial finding that error rates for complex mixtures exceeded 50 percent in the software previously used to convict defendants
• The Third Circuit's April 2026 ruling that source code disclosure is not required for admission establishes that a credential whose evidentiary boundary cannot be independently evaluated is sufficient to authorize conviction — provided the developer asserts validation; the credential resolves the correspondence question; the mechanism that produced the resolution is withheld; the relying party has no alternative to trusting the issuer
• Open-source probabilistic genotyping software exists — the transparency condition is technically achievable; the credential gap is not a technical limitation but a commercial and legal choice; a credential encoding the evidentiary boundary of the likelihood ratio in a form evaluable at the point of reliance — without requiring trust in the developer — is the structural correction this case implies