Long after he retired from solving murders in rural East Texas, Brian Wharton looked back on one of his biggest cases with unease. A father named Robert Roberson had shown up at an emergency room with his two-year-old daughter, Nikki Curtis. She was unconscious and turning blue. The father speculated that she’d fallen out of bed, but a pediatrician concluded that she had been shaken “very forcefully.” As a detective with the Palestine Police Department, Wharton deduced the father was to blame and testified at the 2003 trial, where Roberson was sentenced to death. But Wharton never could make sense of the man’s demeanor throughout these events. “He’s not getting mad, he’s not getting sad, he’s just not right,” Wharton recalled.

Twenty years later, a defense lawyer showed up at the detective’s door, explaining that Roberson’s affect could be explained by autism spectrum disorder. But that wasn’t all. Many in the medical community had turned against the diagnosis at the heart of Roberson’s conviction: “shaken baby syndrome.” Wharton quickly came to believe he’d helped send an innocent man to death row. “I took a deep breath and said, ‘Okay, now we begin to make this right,’ ” he said. “Fortunately, he’s still alive when science comes to his rescue.”

Roberson’s lawyers are now seeking review from the U.S. Supreme Court, which will decide whether to hear his case in the coming weeks. Scientists, doctors, judges, and death row exonerees have filed briefs framing the case as an opportunity to stop wrongful convictions based on what some call “junk science.” 

Their arguments point to a tension in the U.S. legal system: Convictions are supposed to be final, but science changes. In recent years, prisoners and their lawyers have challenged a number of forensic disciplines: from hypnosis and roadside drug tests to testimony about blood splatter, hair, photographs, burn patterns, bite marks, tire tracks, and speech patterns. Since the 1980s, nearly a quarter of overturned convictions have featured “false or misleading forensic evidence,” according to the National Registry of Exonerations, a collaboration of several universities that tracks such cases. 

“We believed anyone in a lab coat with letters after their names,” said M. Chris Fabricant, a lawyer for the New York–based Innocence Project, which works to overturn wrongful convictions across the country, and who wrote the 2022 book Junk Science and the American Criminal Justice System. But “these methods were developed by law enforcement to solve possible crimes, not in laboratories.” Often a police officer could become qualified as an “expert” with a few days of training, and portray their subjective analysis as impartial science on the witness stand. 

There’s no way to know how many convictions—or even executions—have been based on flawed or misinterpreted evidence, and there is little logic to who wins relief and who does not. Shortly before Roberson lost in the Texas courts, a woman named Jennifer Del Prete was released from prison in Illinois after making similar legal claims about “shaken baby syndrome” in that state’s courts. 

Since 2013, seven states have passed laws explicitly directing courts to entertain claims from prisoners based on changing science. Texas was the first, and the state’s top criminal court has overturned around a dozen convictions. But none were death penalty cases, and defense lawyers say it remains too hard to win relief in the state. “You have a law, but no matter what you do, you can’t satisfy the burden the [Texas] court has created,” said Roberson’s lawyer, Gretchen Sween. “The judicial system has totally failed him.”

The Supreme Court ruled in 1993, in Daubert v. Merrell Dow Pharmaceuticals, Inc., that federal judges had to play a “gatekeeping” role and decide whether scientific testimony was “reliable” and “relevant.” (This came at a time when corporations were claiming it was too easy for plaintiffs to enlist scientists to make claims about their products’ harms.) While some state courts adopted this approach, others focused on whether experts’ methods were accepted by their fellow scientists. In the meantime, forensic techniques proliferated, and popular culture portrayed them as exciting new tools for delivering justice. Detectives used “bite marks” to match serial killer Ted Bundy to one victim at his 1979 trial, helping to popularize such analysis despite its questionable accuracy. 

But in the early aughts, a few controversial cases began undermining faith in these tools. 

Cameron Todd Willingham was executed in 2004 for setting a fire that killed his three young daughters. “I am an innocent man,” he said from the gurney. Local investigators had ruled the fire an arson, but national fire experts questioned their work as based more in folk wisdom than science. One compared it to “witch-hunting.” Soon after, lawmakers created the Texas Forensic Science Commission, a group of scientists and lawyers who gained a national reputation for helping courts evaluate the validity of various disciplines, and by 2009, the National Academy of Sciences was questioning a number of them too.   

At Roberson’s 2003 trial, the shaken baby syndrome diagnosis was presented by pediatrician Janet Squires, who pointed to swelling and hemorrhages in Nikki’s brain. A pathologist named Jill Urban surmised that the girl had sustained multiple head injuries. (Neither Squires nor Urban agreed to be interviewed for this story.) 

Roberson’s lawyers didn’t contest this basic narrative. He had previously admitted to grabbing Nikki while trying to rouse her from unconsciousness, and other witnesses said they’d seen him threaten, shake, and spank her on other occasions, but there was no direct evidence tying him to her injuries.

An emergency room nurse also speculated on the witness stand that Nikki had been sexually assaulted. Testing did not corroborate her testimony, and the nurse was not an expert on the subject, but Roberson’s lawyers argue these claims further contributed to the jury’s decision to sentence him to death.

In a recent letter to the Marshall Project, Roberson described being stunned by the medical testimony. “I was already suffering [the] pain of losing my daughter Nikki . . . like a sudden, violent impact,” he wrote, adding that he’s praying for release “so I can start living again and so that I can take back what was stolen from me.”

He was nearly executed in 2016, but the Texas Court of Criminal Appeals issued a stay, and told an Anderson County judge to reevaluate the science in the case. New medical experts rejected the earlier diagnoses, linking Nikki’s death instead to her chronic infections, as well as a form of apnea that would cause her to collapse and turn blue. She’d been prescribed codeine, which can cause breathing difficulties in children. 

According to the new experts, it was common in 2003 to see the toddler’s injuries as evidence of shaken baby syndrome, but this diagnosis had spread among doctors without much research to back it up, particularly when applied to children beyond infancy. Newer studies had demonstrated that short falls, oxygen deprivation, and other causes could produce similar outcomes.

At the same time, there was a growing body of evidence that police frequently lack training on autism spectrum disorder, and misinterpret behavior in ways that cast suspicion on the wrong person. “What’s maddening in [Roberson’s] case is how they use the autism against him, as evidence of his guilt,” Sween said. (The Anderson County district attorney’s office did not respond to a request for an interview.)

Lawyers from the Texas attorney general’s office have asked the U.S. Supreme Court to reject the case. They argue that none of the witnesses against Roberson knowingly offered false testimony, and that an ongoing debate among scientists hardly justifies questioning a conviction after it has been reviewed by state courts. They note that the medical examiner largely relied on blunt force injuries, rather than a theory of shaking, when she ruled the death a homicide. 

If the Supreme Court declines to hear Roberson’s case, prosecutors could seek an execution date, and his fate would ultimately rest with Texas Governor Greg Abbott, who could commute the sentence, but seldom does so. Celebrities may start exerting pressure; author John Grisham is already at Roberson’s side.

This will hardly be the last case of its kind. Where the Innocence Project once used DNA to show how innocent suspects went to prison in cases of murder and sexual assault, now the organization is taking cases—such as those of other Texans Melissa Lucio and Rosa Jimenez—in which deaths of children were misrepresented as murders with faulty medical testimony. “It’s no exaggeration to say hundreds if not thousands of parents and caregivers are in prison in cases where no crime occurred,” said Vanessa Potkin, director of special litigation for the organization. She argues the Supreme Court now has an opportunity to “send a message that when there is a growing medical or scientific consensus that discredits a criminal conviction, it can’t stand.”