The Issue
A crime lab hands a jury a number. Twenty-one point four trillion – 21,400,000,000,000 – times more probable. That statistic came out of a piece of software called TrueAllele, and it helped convict a man of a gun crime carrying the death penalty as a possible sentence. His lawyer asked to see how the software actually reached that number. The company said no, the code is a trade secret. Courts around the country are still fighting over whether that answer is good enough.
If DNA evidence comes from a program like this in your case, probabilistic genotyping software, you need to understand what it does, and what it means when the people relying on it refuse to show how it works.
Background: Why This Software Exists
For decades, DNA mixture interpretation relied on a method called combined probability of inclusion, or CPI. It works like a checklist. An analyst looks at a DNA mixture from multiple contributors and asks a simple question at each of the twenty standard genetic locations: can I exclude this specific person as a possible contributor here, yes or no. Add up the answers, and you get a rough statistic.
CPI works fine on a clean, two-person mixture. It falls apart on the messy evidence that actually shows up in real casework, three or four contributors, degraded DNA, tiny touch samples where some genetic markers simply do not show up at all. Faced with that kind of evidence, CPI often produces no usable answer. Labs were sitting on DNA evidence they could not interpret.
For that reason, and because CPI came to be seen as inherently unfair to the accused, its popularity faded. The method is a subjective, one-sided statistic: it looks only at whether a suspect can be included in a mixture and never asks how well any alternative person would fit the same evidence, which tends to work against the defendant. Dr. Mark Perlin, the founder of Cybergenetics and the creator of TrueAllele, has been one of the method’s sharpest public critics. In a 2015 interview with Forensic Magazine he put it bluntly, calling CPI “a random number generator” and warning that convictions had rested on it.
As labs moved away from CPI, they turned to methods that use more of the DNA data and weigh the evidence on both sides. Rather than a simple include-or-exclude answer, analysts began reporting a likelihood ratio, or LR, which compares how probable the evidence is if the suspect contributed to the mixture against how probable it is if only unrelated, random people did. Related probabilities such as the random match probability, or RMP, describe how often a matching profile would be expected to occur in the general population. These measures form the statistical backbone of the probabilistic genotyping systems that eventually replaced CPI for complex mixtures.
Probabilistic genotyping software was built to solve that problem. Programs like STRmix and TrueAllele do not use a yes-or-no checklist. They build a statistical model of the crime scene sample and run it through millions of simulated scenarios, using a mathematical approach called Markov Chain Monte Carlo. Picture being handed a blurry photograph of a crowd and asked who is in it. Instead of guessing once, the software runs millions of different guesses about who might be standing where, how many people are in the shot, and how much of each person’s outline is obscured. It scores every guess by how well it explains the actual blur pattern in front of it, and the guesses that explain the evidence best rise to the top. The output is not a match. It is a likelihood ratio, a number that says how many times more probable the evidence is if a specific person contributed to the mixture, compared to if some random unrelated person did.
Why the Source Code Fight Matters
Here is the problem. That likelihood ratio depends entirely on assumptions built into the software: how many contributors the analyst tells it to assume, how it models degraded or dropped-out genetic material, how it weighs artifacts in the data. Get those assumptions wrong, and the number that reaches the jury can be wrong too, sometimes dramatically so.
This is not theoretical. In 2016, a federal judge in New York ordered the source code released for a different program, New York City’s own Forensic Statistical Tool. Independent reviewers found it had a flaw that tended to overstate the likelihood of guilt. The city retired the software. Separately, the maker of STRmix has acknowledged a coding error that affected results in roughly sixty cases in Australia. Software written by people has bugs. Forensic DNA software is not exempt from that rule, and federal validation work through the National Institute of Justice underscores how sensitive these tools are to the choices made when they run.
So when a defendant asks to have his own expert examine the code behind the software used to convict him, the request is not paranoia. It is the same kind of scrutiny any other piece of scientific evidence has to survive. But the companies that make this software treat the code as a closely guarded trade secret, and courts around the country have not agreed on how to handle that.
Some courts have ordered the maker of a probabilistic genotyping tool to hand its source code and validation materials to the defense under a protective order, so an independent expert could examine it before a hearing on whether the evidence should even be admitted. The reasoning is direct: without that access, a defendant is stuck taking the company’s word for it that the software works as advertised, and that is not how an adversarial system of justice is supposed to function.
Other courts have gone the other way, reasoning that a program’s extensive published validation studies, its use across dozens of accredited labs, and its track record in admissibility hearings were enough, without needing to hand over the underlying code itself. The result, nationally, is a real split. There is no single nationwide rule, and even the scientific working group that publishes validation guidelines for the field has not resolved the disagreement. If you want a deeper, lab-level walkthrough of how these models are built and validated, our team breaks it down over at Ambeau Forensics.
How This Applies to Your Case
If probabilistic genotyping software produced DNA evidence against you, whether it is STRmix, TrueAllele, or a lab’s own in-house program, the first move is finding out exactly which assumptions the analyst fed into it. How many contributors did they tell the software to assume were in the mixture. What settings governed how it handled degraded or low-quantity DNA. Small changes to those inputs can move a likelihood ratio by orders of magnitude, and the report handed to you may not spell out every assumption baked into the number. A qualified forensic DNA expert can pull those inputs apart.
Request the software’s full validation record and, where the developer offers it, its extended output, which shows the intermediate steps the program took to reach its answer, not just the final statistic. The maker of STRmix has said this extended output is often more useful for real scrutiny than the raw source code itself, since it lets an expert see exactly how the program treated the specific data in your case. This is the heart of a real DNA evidence challenge.
Whether it is worth fighting for full source-code access depends on which program is involved and which court you are in. Some developers’ formal access policies mean that fight is often unnecessary; the door is already open if you know to ask. Others are a harder fight, and the law is genuinely split on whether a defendant is entitled to win it. Either way, do not accept a likelihood ratio as self-proving. It is a statistical conclusion built on assumptions a qualified expert can examine and, where appropriate, challenge under Louisiana’s standard for expert testimony at Louisiana Code of Evidence Article 702. If your case runs through a Louisiana court, that is the rule that governs the fight.
Frequently Asked Questions
What is a likelihood ratio, in plain terms?
It is a comparison, not a match. It says how many times more likely the DNA evidence is if a specific person contributed to the sample, versus if an unrelated person did. It is not a statement that the person is guilty.
Why won’t the software companies just release their source code?
They treat it as a valuable trade secret. Courts that have ordered disclosure have generally required a protective order, meaning the code is shown only to a qualified expert under a confidentiality agreement.
Has this software ever been shown to have bugs?
Yes. A coding flaw in New York’s own DNA software led to it being retired after independent review, and the maker of STRmix has acknowledged a coding error affecting around sixty cases in Australia.
Does Louisiana use this kind of software?
Probabilistic genotyping software is used by forensic laboratories across the country, and complex DNA mixture cases in Louisiana courts have involved this kind of analysis. Ask directly which software was used in your case and what assumptions the analyst applied.
Can I challenge this kind of evidence even if I can’t get the source code?
Yes. You can challenge the assumptions the analyst chose, the lab’s internal validation studies, and whether the software has been properly validated for the type of sample in your case.
Conclusion
A number that sounds this precise, trillions to one, can feel unanswerable. It is not. That number came from a program built on human-chosen assumptions, running on code no one outside the company that built it has fully seen. Some courts have decided that is not good enough on its own. If DNA evidence in your case came from software like this, do not treat the printout as the final word. Ask what went into it, and get someone who understands the science to check the math before you accept the state’s number as the truth.