On March 1, a Hartford judge officially dismissed a decades-old murder charge against Alfred Swinton. Swinton, who served 18 years of a 60-year sentence, was released last June after DNA testing exculpated him, and a forensic expert who testified at his trial recanted.

But it was not until this month, when the state’s attorney confirmed that her office would not retry Swinton, that the miscarriage of justice that claimed nearly two decades of a man’s life was rectified. Swinton’s story provides an opportunity for Connecticut to critically examine the ways in which forensic evidence can threaten the integrity of the criminal justice system—and how, in the process, we can look for answers.

At Swinton’s 2001 murder trial, the most damning evidence came from Gus Karazulas, the chief forensic dentist for the Connecticut State Police. Karazulas, who claimed to be an expert in analyzing bite marks, testified that bite marks found on the murder victim’s body had been made by Swinton. His opinion rejected any possibility of error; Karazulas testified, “I believe that with reasonable medical certainty without any reservation that these marks were created by [Swinton’s] teeth.” That sentence was quoted by the Connecticut Supreme Court in upholding Swinton’s conviction.

Bite-mark analysis relies on two premises: first, that human dentition, like DNA, is entirely unique; and second, that human skin can record a dental impression with enough sensitivity to be accurately matched to an individual. The problem is that neither premise has been proved. In the years since Swinton’s conviction, bite-mark analysis has been almost entirely debunked.

In 2009, the National Academy of Sciences published a congressionally commissioned report on the state of forensic science in the courtroom. The report was critical of a wide range of forensic specialties, but it singled out bite-mark analysis as particularly dubious. It found “no evidence of an existing scientific basis for identifying an individual to the exclusion of all others.” In 2016, a report by the United States President’s Council of Advisors on Science and Technology affirmed this conclusion and went a step further, declaring that the odds of transforming bite-mark analysis into a scientifically valid method were nil.

Acting on these developments, Swinton’s attorneys asked Karazulas to re-evaluate the state of forensic dentistry and his own opinions. To his credit, Karazulas did so wholeheartedly. He filed an affidavit on Swinton’s behalf that discredited bite-mark analysis as a valid forensic method and retracted his opinion about Swinton. Approximately 30 other defendants across the country have been exonerated in similar circumstances, but many more remain incarcerated based on bite-mark testimony.

The judicial branch’s role as gatekeeper when it comes to scientific evidence is inherently limited. It is only as good as the prevailing view in the scientific community at the time. When our Supreme Court affirmed Swinton’s conviction in 2004, for example, it was able to marshal compelling evidence that bite-mark analysis was methodologically valid and employed in courtrooms nationwide. This raises a difficult but important question: how can courts protect the rights of criminal defendants when the reliability of a forensic method may shift radically over time?

One answer may lie in imposing stricter limits in criminal cases on how expert forensic opinions are given. According to the latest research, many forensic methods—even if generally reliable—carry meaningful error rates. And most, with the exception of DNA, are not able to identify a perpetrator to the exclusion of others. Yet across all fields, experts elide these critical limitations, overstating the probative value of their evidence and expressing a confidence level that far exceeds what the relevant science can justify.

Opinions rendered with empirically unfounded certainty are not a matter of expertise or professional judgment, they are scientifically invalid. And the consequences are alarming. A three-year study by the Department of Justice and the FBI of 3,000 criminal cases involving microscopic hair analysis revealed that FBI examiners had provided scientifically invalid testimony in more than 95 percent of cases in which testimony was used to convict a defendant.

Courts may not be able to foresee how science will change, but they can—and should—limit forensic experts from overstating the probative value of their opinions. The 2016 report from the President’s Council provides concrete recommendations in this area, urging courts to preclude forensic experts from using a laundry list of common phrases that are scientifically indefensible. One of those phrases is “to a reasonable degree of scientific certainty,” almost precisely the language that helped convict Swinton.

Turning a critical eye to forensic science and seeking avenues for reform will not atone for what was done to Swinton, but it is an essential first step.