In March 2026, the U.S. Court of Appeals for the Third Circuit handed prosecutors a significant victory in DNA mixture litigation. In United States v. Anderson, No. 25-1223, the court ruled that TrueAllele (a proprietary DNA interpretation software used by crime labs across the country) is reliable enough under Federal Rule of Evidence 702 to go before a jury, even though the defendant could not access its underlying source code. The case involved a likelihood ratio of 11.5 trillion to one. The jury convicted.

The decision will be cited in suppression and Daubert hearings nationwide. In Texas, where courts apply a parallel reliability standard, the ruling’s influence will be felt in any case where the prosecution’s DNA evidence rests on probabilistic genotyping software. This blog explains what that software is, why defense attorneys fight so hard to challenge it, and where the arguments that actually work are found.

What Is Probabilistic Genotyping? A Plain-English Explanation

Traditional DNA analysis works when the sample is clean and comes from a single person. The lab extracts a DNA profile, compares it to the suspect’s profile, and if the markers match, reports a statistic: something like “this profile would occur in 1 in 30 quadrillion people.” That number reflects a relatively straightforward calculation.

Real crime scenes are messier. DNA is often found as a mixture: two, three, four, or more contributors whose genetic material is combined on a single swab. A bloodstain on a steering wheel may contain DNA from the driver, a passenger, a previous owner, and a crime victim. A gun grip may carry DNA from multiple handlers. Traditional analysis struggles with mixtures, particularly when contributors are in unequal proportions or when the sample is degraded.

Probabilistic genotyping software was developed to address this problem. Instead of looking for a simple match, the software runs thousands or millions of statistical simulations to calculate what genotype combinations could have produced the observed mixture. It then reports the results as a likelihood ratio: a number expressing how much more probable the observed DNA evidence is if a specific person contributed to it versus if a random person of the same ancestry did.

What “11.5 trillion to one” actually means: In the Anderson case, TrueAllele reported that the DNA mixture from the gun was 11.5 trillion times more probable if the defendant had contributed to it than if a random Caucasian person had. This sounds definitive. But the number is a model output, not a measurement. It depends on assumptions the analyst made before running the software, and different software programs fed the same sample have produced results differing by more than a thousand-fold.

 

The Third Circuit’s Decision in United States v. Anderson

In Anderson, Hunter Anderson was charged with being a felon in possession of a firearm. The state crime lab could not conclude with scientific certainty whether his DNA was present in the mixture found on the gun. The prosecution turned to TrueAllele, a product of Pittsburgh-based Cybergenetics. TrueAllele produced the 11.5 trillion to one likelihood ratio. Anderson’s lawyers moved to exclude the evidence under Daubert.

The district court held a two-day hearing with experts on both sides and admitted the evidence. On appeal, the Third Circuit panel affirmed. Walking through the standard Daubert factors:

  • Testability: TrueAllele can be tested. Five independent probabilistic genotyping programs produced comparable results on the same sample. The court found this sufficient.
  • Peer review and publication: 42 validation studies, 8 of which were peer-reviewed.
  • Known error rate: A Virginia study found TrueAllele’s false-positive rate to be 0.005%, compared to 2–6% for human-reviewed comparative DNA analysis.
  • General acceptance: Broad use in forensic labs and admission in courts across the country.

On the source code question, the court held directly: “Daubert is not a criminal discovery device.” The defendant’s right to challenge the methodology does not extend to a right to inspect the proprietary source code of a private company’s software product. The defense got a full hearing. That, the court said, was what Daubert promised.

The court’s bottom line: The issues the defense raised (hidden source code, analyst judgment calls, and the probabilistic nature of the output) go to weight, not admissibility. They are arguments for the jury, not grounds for exclusion.

 

What the Court Got Right and What It Glossed Over

What the Court Got Right

The Daubert framework asks whether a method is scientifically valid and reliably applied but not whether it is perfect. By that measure, probabilistic genotyping has a legitimate scientific foundation. The underlying statistical approach is sound. The validation studies exist. The false-positive rate, at 0.005%, is genuinely lower than human review. The court’s analysis of the formal Daubert factors is defensible.

What Deserves Harder Scrutiny

The number of contributors problem. Every probabilistic genotyping analysis begins with an analyst decision: how many people contributed to this mixture? This is not a measured fact. It is a judgment call, often based on a count of allele peaks in the sample. If the analyst assumes three contributors but there were actually five, the software’s entire model is built on a false premise and the output, however precise in its expression, is precisely wrong.

In United States v. Ortiz, 736 F. Supp. 3d 895 (S.D. Cal. 2024), STRmix was excluded from evidence entirely when a defense expert showed that the number of contributors was likely six or more, not the number the analyst had assumed. The court found the model’s foundation so compromised that the results could not be admitted. This is exactly the kind of challenge that can succeed even after the Third Circuit’s ruling in Anderson.

The program-to-program variance problem. A 2023 study published in the Journal of Forensic Sciences (Thompson, W.C., “Uncertainty in Probabilistic Genotyping of Low Template DNA”) compared TrueAllele and STRmix on the same DNA evidence in a federal criminal case. The results were strikingly different: STRmix reported a likelihood ratio of 24 in favor of the non-contributor hypothesis. TrueAllele reported a likelihood ratio ranging from 1.2 million to 16.7 million in favor of the contributor hypothesis, depending on the reference population used. That is not a minor discrepancy. That is a difference of more than three orders of magnitude meaning more than a thousand-fold difference in the reported strength of the DNA evidence. Both programs are scientifically validated. Both cleared Daubert. They cannot both be right about the same sample.

The 0.005% false-positive rate in context. TrueAllele’s reported false-positive rate of 0.005% is lower than human error. But 0.005% of millions of processed samples is still thousands of false positives. And the Virginia study from which this figure derives was conducted on a specific set of cases with particular characteristics. Whether that rate generalizes to the degraded, low-template, high-contributor-count mixtures that are often encountered at real crime scenes is a contestable scientific question.

The Confrontation Clause question. The Sixth Amendment guarantees a defendant the right to confront the witnesses against them. When the prosecution’s DNA evidence is a number generated by a proprietary algorithm running on a privately held server, who is the witness? The analyst who ran the software? The creator of the software who designed the model? The Third Circuit did not reach the Confrontation Clause question in Anderson, because it was decided on Daubert grounds. That question remains open and will be litigated in future cases.

 

Texas: The Kelly Standard and What It Means Here

Texas is not bound by the Third Circuit’s ruling in Anderson. Federal circuit court decisions are persuasive authority in Texas courts, not binding authority. Texas applies its own reliability standard for novel scientific evidence, established in

Kelly v. State, 824 S.W.2d 568 (Tex. Crim. App. 1992), and developed through subsequent cases including Nenno v. State and Hartman v. State. The Kelly framework asks three questions:

  • Is the underlying scientific theory valid?
  • Is the technique applying that theory valid?
  • Was the technique properly applied in the case on trial?

The first two prongs mirror Daubert’s reliability inquiry. The third prong (proper application in this specific case) is where probabilistic genotyping challenges in Texas will focus. Even if TrueAllele is generally valid and the technique is sound, the question of whether the analyst correctly applied it in this case remains entirely open for challenge.

The analyst’s assumptions are the battleground. In Texas, the defense can challenge the specific analyst decisions made in the case on trial: the assumed number of contributors, the stochastic threshold settings, the reference population selected for the likelihood ratio calculation, and the quality control steps followed in the lab. These are not abstract challenges to the software’s general validity. They are challenges to the application of the software in this particular case, which is exactly what the third Kelly prong addresses.

A Texas Connection: Lydell Grant

Texas courts have already encountered probabilistic genotyping on both sides of the ledger. In the case of Lydell Grant, a Houston man convicted of stabbing a man outside a bar in 2010, the Texas Innocence Project asked Cybergenetics to run the DNA evidence through TrueAllele. The software indicated that the DNA under the victim’s fingernails was far more consistent with another person’s profile than with Grant’s. Grant was exonerated in 2021 after spending more than a decade in prison.

The Grant case illustrates something important: probabilistic genotyping is not inherently a prosecution tool. When applied to exclude a contributor, it can be a powerful exoneration tool. The fight over source code access, analyst assumptions, and validation methodology applies equally in both directions. The same challenges that limit a prosecution’s ability to use TrueAllele inculpatorily also enable its use exculpatorily.

Source Code: Brady and Due Process After Anderson

Anderson held that Daubert does not require source code disclosure. But the Third Circuit was careful to note that other legal mechanisms may require disclosure. The government’s Brady obligations require disclosure of material exculpatory evidence in the prosecution team’s possession. The defendant’s due process right to a meaningful opportunity to present a complete defense may independently require access to information necessary to challenge the prosecution’s most powerful evidence. In Texas, where the exclusionary rule under Article 38.23 has no good faith exception, a successful challenge to the admissibility of TrueAllele evidence on due process or confrontation grounds could result in suppression of the DNA evidence entirely.

 

What Defense Attorneys Should Do When Facing Probabilistic Genotyping Evidence

  • Demand the complete analyst file. Every input assumption the analyst made when configuring the software run: the assumed number of contributors, the stochastic threshold, the reference population, the software version, and the quality control documentation. These are not protected by the software company’s trade secret interests. They are the analyst’s work product and must be disclosed.
  • Get an independent expert. The defense expert’s job is not to attack probabilistic genotyping generally (courts have rejected that argument). The expert’s job is to evaluate whether the analyst’s specific assumptions and inputs in this case were scientifically justified, and to assess what the likelihood ratio would have been under alternative reasonable assumptions.
  • Challenge the number-of-contributors assumption. This is the single most productive challenge in cases involving complex mixtures. If the lab assumed three contributors and there were four, the entire analysis is wrong. Demand the raw electropherogram data that the analyst used to make this determination and have an independent expert evaluate it.
  • Run the sample through an alternative program. The Thompson 2023 study found that TrueAllele and STRmix produced results differing by more than a thousand-fold on the same sample. Having the same evidence analyzed by a different validated probabilistic genotyping program is a legitimate defense strategy that has succeeded in suppression hearings. If the programs produce materially different results, that itself is powerful cross-examination material.
  • Challenge the likelihood ratio’s presentation to the jury. Jurors who hear “11.5 trillion to one” will almost certainly overweight that number. The defense expert can explain that the number is a model output, not a measurement; that it is sensitive to input assumptions; and that different reasonable assumptions would produce a dramatically different number. This does not defeat the evidence, but it contextualizes it for a jury that would otherwise treat it as scientific certainty.
  • Kelly prong three: proper application. Even in Texas courts that accept TrueAllele’s general validity, challenge whether the specific analyst followed the software’s own protocols, whether the lab had conducted internal validation on this version of the software, and whether the quality control documentation shows any deviation from the established protocol.

 

The Bigger Picture

Probabilistic genotyping is not the last algorithm that will show up in a Texas criminal courtroom. Gait analysis software, voice identification tools, AI-assisted image enhancement, and cell phone location inference programs are already being used in investigations and are approaching courts as evidence. Each of them presents the same fundamental challenge: a black box that processes inputs and produces an output, where the defendant cannot examine the process by which the output was generated.

Anderson is a data point in a longer argument about what scientific reliability means in criminal cases. The Third Circuit answered the question within the Daubert framework and found TrueAllele reliable. The harder constitutional questions (such as confrontation, due process, and the defendant’s right to meaningfully test the evidence that may send them to prison) remain open. Texas courts, applying the Kelly standard with its independent focus on proper application in each case, have room to develop those questions in ways federal courts have not yet reached.

 

Sources

 

If you are facing criminal charges in Texas in which the prosecution’s DNA evidence involves probabilistic genotyping software, call (214) 225-7117 for a free, confidential consultation. The analyst’s assumptions, the software version, and the quality control documentation are the starting points for this challenge.