This week, a Connecticut jury convicted Richard Dabate of murder in the killing of his wife, Connie Dabate, after a five-week-long trial that hinged — in part — on data from her Fitbit. Richard said a man dressed in camouflage broke into their home in 2015 and shot Connie. But Connie wore a Fitbit, and data from the device showed movement for around an hour after Richard said the break-in happened.

To make the case that the Fitbit data helped show Richard killed Connie, prosecutors called on Keith Diaz, an exercise physiologist and professor at Columbia University Medical Center. Diaz has done studies validating the accuracy of Fitbits and testified about their precision.

Diaz often testifies as an expert witness in criminal trials. He told The Verge that it’s personally gratifying to participate in trials but that it’s also a challenge. The way he usually thinks about Fitbit data as a scientist is different from the way he’s asked about it in a courtroom. “The scientific questions we’re answering are different from the criminal questions,” he says. “What I’ve strived to do in these cases is translate that.”

Fitbits are generally accurate devices, Diaz says, but they’re by no means perfect. For scientific research, that’s expected — nothing is ever absolute, and it’s rare to get an answer that comes with 100 percent certainty. Science lives with some amount of statistical error. But the law operates under different guidelines: it wants to know if something happened beyond a reasonable doubt. “This has very serious implications for if someone could go away for 25 years to life,” he says. “So is it okay to live with some gray area — and under what conditions? That’s been some of the challenge and the hurdles to clear.”

Diaz talked to The Verge about his experiences testifying in the trials and how he navigates the tensions between science and the legal system in the process.

The following has been lightly edited for clarity.

What’s the difference between what a scientist might want to know about a Fitbit and what a jury might need to know?

In science, we care about things — like if I took 100 steps but the device says I only took 92 steps. That matters in science, but how much does it matter in the criminal cases? With those, it’s often just a matter of if the device detects movement. That’s good enough for the case. It doesn’t matter if it’s 100 steps versus 92; it’s really just about if they moved. With this Connecticut case, there was a lot of back and forth in the cross-examination about the error rate. But the error rate was on the number of steps someone took — not the error rate of if they were moving. That’s a very different story.

And in science, especially in a lot of the studies I work with, you’re working with large sample sizes that could have thousands of people. Any noise gets dissipated with so many people. But, with a criminal case, you have just one person — so noise matters and can bring in some doubt about what really happened. In criminal cases, there’s a level of precision you want to have around, okay, who moved exactly at this minute. Versus in science, if there’s an error because you have such a large sample size, it can get filtered out. So it’s been challenging to figure out how you convey that these devices can’t tell you perfectly what’s happening.

How do you adjust the science to meet the needs of a criminal case?

You can leverage the things that science might see as errors. So, if you’re brushing your teeth and wearing a device on your wrist, there are cases where the device misregistered that as steps. Obviously, in science, that’s an error. But that can actually be really helpful with some of these cases because it shows that it’s a high-sensitivity device. It’s actually a strength to use these devices in this context because of their high sensitivity. When there is no motion detected, when it is not recording any steps, we should be pretty darn confident that a person is not moving.

So it’s been having to shift what we looked at in the science to say, ‘Well, what is the most important thing in the context of this criminal case?’ We can say that this person wasn’t moving anymore because the device wasn’t recording anything. So what does that mean in this context?

In a criminal trial, there are two sides, and you’re asked to testify for one side or the other. What’s that like as a scientist?

I’ve been hired by prosecutors, never by the defense. And some of the prosecutors have made efforts to bias my interpretation. It may be unintentional, but they are going to have the perspective that they believe this person is guilty and that they need someone to come in and support the case. But I don’t want to just go in and be ammunition for the prosecutor; I want to give a balanced interpretation of what the science means and how it can be applied in the situation. It could be easy for me to go in and say things that would support the case for the prosecutor. But how do you balance that with also acknowledging that there is error, and it’s not a perfect device?

It’s hard because obviously the prosecution doesn’t want to highlight the errors. But from the defense side, I’ve noticed them not asking the right questions about that. They’re not scientists, so they don’t know what to ask to highlight the limitations of the device. And I can only answer the questions that are asked of me.

Has this experience changed how you think about your research?

If this is something I go further down the road into, I would want to conduct more scientific studies. In a lot of the situations that I’ve seen with these cases, the science that exists is not really designed to answer the important questions. Like: is it possible for a person to get from a bed to a bathroom 15 feet away without the device registering? So I might look into things like that because I think it would strengthen my ability to give expert testimony. And for a jury, it would be helpful to have better answers to these kinds of questions that come up in a criminal case that don’t come up in science. But for now, it’s kind of a side gig. This is a bit outside of what my day-to-day job is.