DNA has been advised the gold accustomed of argumentative affirmation for more than 30 years, even as assorted types of junk science have fallen out of use. But in recent years, police and crime labs have continued and broadcast how they use abiogenetic actual to pin suspects—from borer into clandestine ancestor websites to creating police sketches of suspects’ faces.

The latest convenance to come under assay is an abstruse technique, “probabilistic genotyping,” that takes abridged or contrarily ambiguous DNA left behind at a crime scene, often in atomic amounts, and runs it through a software affairs that calculates how likely it is to have come from a accurate person. One such program, TrueAllele, has been used in more than 850 bent cases over the past 20 years. The problem? No one knows whether it works—the code, developed by a clandestine aggregation called Cybergenetics, is proprietary.

Government crime labs that use the software don’t get access to the program’s source code. Employees of Cybergenetics don’t get access. Even the authors of the peer-reviewed studies of TrueAllele have never had access to the code.

But now, two bent cases—one in the U.S. District Court for the Western District of Pennsylvania and accession in the Appellate Division of the Superior Court of New Jersey— may give the world a first peek into TrueAllele’s backstairs algorithm. Last month, the New Jersey judge ordered prosecutors to hand over the source code for TrueAllele, and a few weeks later, the federal judge in Pennsylvania did the same.

Experts say the affairs is so circuitous and so hidden from assay that software bugs are inevitable.

“It is around assertive that there are flaws in the TrueAllele software,” wrote Mats Heimdahl and Jeanna Matthews, two computer science experts, in a acknowledgment to the federal court. “On average, there will be six flaws for every 1,000 lines of code, and TrueAllele has 170,000 lines of code.”

While the orders do not make the source code about available, they allow the defense  to bring in experts to appraise the software for abeyant flaws and inconsistencies—under strict nondisclosure agreements. Should the experts find any problems, the aegis attorneys could try to get the DNA affirmation thrown out, abrasion the cases adjoin their audience and potentially arch to a larger reckoning for TrueAllele.

[Read: How do you build a pet-friendly gadget? We asked experts and animal owners]

The software program’s apparatus are far more circuitous and difficult to carbon than the acceptable action of DNA testing, which matches a doubtable to a robust type of evidence, like blood.

TrueAllele doesn’t a doubtable to concrete evidence; it calculates the statistical that a person’s DNA is present in a complicated admixture of assorted people’s DNA, or in a atomic amount of DNA left behind—for instance, after accession merely afflicted something.

In the Pennsylvania case, federal prosecutors used the software to parse a complicated admixture of DNA found on a blaster left in a car—and specifically, to account the likelihood that any of it belonged to a Pittsburgh man named Lafon Ellis. Cybergenetics experts testified that the sample likely absolute DNA from four people and that it was “21.4 abundance times more probable” that Ellis’s DNA was in the mix than another, random African American person’s. Ellis’s attorney, however, says the gun isn’t Ellis’s and disputes the DNA affirmation abutting him to it.

In accession to prosecutors, aegis teams have also used the software to help prove their clients’ chastity at trial or absolve them later. Cybergenetics has made the affairs accessible to either side in any case for testing.

According to Cybergenetics, TrueAllele’s calculations have already been accustomed into affirmation in 14 states, with 20 bootless attempts in recent years by bent aegis teams to gain access to the program’s source code. Prosecutors have long argued that the affairs was built on widely accustomed algebraic concepts and that aperture up its source code to public assay would abuse the company’s trade secrets.

The software’s lack of scrutiny

But experts and civil rights advocates have long been anxious about the lack of assay of the software. Their arguments may be accepting traction, as apparent by the two rulings last month.

“Our amends system cannot permit aesthetics based on secret evidence,” wrote the Cyberbanking Frontier Foundation and the American Civil Liberties Union of Pennsylvania in an amicus brief to the federal court. “There is a long history of junk science active under the guise of abstruse advance in bent cases—and of public access to and assay of such affirmation as the means to its closing invalidation.”

From bite-mark arrangement affirmation to blood-spatter analysis, and even ballistics testing and fingerprint matching—many argumentative methods that were once commonplace in crime labs and courts have later been questioned or even absolutely debunked after facing alfresco scrutiny.

Life-or-death software programs—like those used in medical accessories or airplanes—must abide absolute validation and assay processes, but not so DNA software. There are few federal regulations about how police or crime labs acquaint new technologies and methods into crime-solving. Every state has its own rules about how a new type of argumentative technology gets accustomed for use, and judges tend to follow prior decisions when chief whether or not to allow new accurate techniques into evidence.

Reviewing TrueAllele will likely be complicated—its artist estimates doing so would take a person, account 10 lines of code an hour, about eight and a half years to complete (which aegis attorneys dispute)—and how to go about it has been accession source of abrasion amid prosecutors and aegis attorneys.

Originally, aegis teams in the two cases were afraid to learn their access to the source code would be belted to a read-only iPad and that they would be accustomed to use only a pen and paper to take notes. They have since lobbied for and won access to an cyberbanking copy they can run and test themselves.

Though TrueAllele has never faced assay like this before, two of its competitors have.

The competitors

When a federal judge of the U.S. District Court for the Southern District of New York ordered source-code access to New York City’s proprietary DNA assay software Argumentative Statistical Tool (FST) in 2016, the assay appear a austere flaw that “tend[ed] to aggrandize the likelihood of guilt.” The judge in that case eventually lifted the careful order on the software, and ProPublica acquired the code and appear it on GitHub, acceptance advisers and the accepted public to analyze it themselves.

The New York City Office of Chief Medical Examiner replaced FST with STRmix, a widely used adversary to TrueAllele. It so happens that the makers of STRmix have also accustomed software bugs that afflicted 60 cases in Australia.

“Software errors are common and argumentative software has no appropriate amnesty from the bugs and mistakes that plague software in other fields,” wrote Khasha Attaran, the abettor federal public apostle apery Ellis, in an emailed statement. “Constitutional attempt and candor demand access because an accused must be accustomed to appraise and claiming the accurateness of software-generated evidence relied on by the federal government.”

The federal prosecutors may appeal the accommodation to the U.S. Court of Appeals for the Third Circuit but haven’t adumbrated yet whether or not they will. The New Jersey case may head to the state’s Supreme Court. The U.S. Attorney’s office prosecuting the Ellis case for the Western District of Pennsylvania and the Hudson County Prosecutor’s Office, which brought the case in New Jersey, both beneath to animadversion on advancing cases, as did Mark Perlin, the architect of Cybergenetics and TrueAllele’s inventor.

Read next: Adventures with AI: Here's what happened when I ate a three course meal advised by bogus intelligence