“The problem was how to take multiple vantage points and put them together into a coherent analysis,” he said. “Maidan was the same problem set, but on steroids.”
The quantity of raw footage amassed by Nefertari was overwhelming, running into the thousands of hours. The problem was how to fuse it into a whole that would persuade the judges and stand up to cross-examination. Prosecutors would have to do more than cherry-pick a few convincing moments. Their theory of the case would have to be strong enough to survive every alternative scenario, from the mysterious rooftop snipers to the possibility that protesters were killed by friendly fire. Samuels worked out the basics of a collaboration with Dykan and Yatsenko over a picnic table. He planned to copy their data onto his laptop but wound up having to buy an external hard drive when he saw how much they had.
Multiple cameras recorded the deaths of three protesters whose families Dykan and Yatsenko were representing, who were selected for the video presentation. One of the dead was Ihor Dmytriv, a 30-year-old lawyer who arrived at the protest on Feb. 19. He was crouching behind a makeshift shield when a bullet pierced it and entered his chest. He fell to the ground, rolling on his back and clutching his knee. A tall construction worker in camouflage fatigues rushed forward to help Dmytriv. He was Andriy Dyhdalovych, a 40-year-old longtime protester who had marched with Ukrainian veterans of the Afghanistan campaign. He was shot in the shoulder and died that day in the street. The third victim was Yuriy Parashchuk, who was 47. He was shot about 15 minutes later on the same street, in the front of the head. None of the three men were armed.
The Maidan reconstruction is a product of its time, an age when high-quality video can be recorded from any street corner or citizen’s hand, and when gigabytes of data can easily circulate among experts in Pittsburgh, Brooklyn and Kiev. The archive the lawyers handed over was huge — a folder of more than 400 videos with different naming conventions and file types. “This was as robust a data set as we’ve ever had the opportunity to work with,” Samuels says. Nefertari had spent months going through the footage on her computer, trying to synchronize the videos and wrangle them together. The Center for Human Rights Science in Pittsburgh subsequently tried to automate this process, using an A.I. algorithm that could quickly analyze each file’s audio component and propose possible matches. “Machine learning is not a magic bullet,” says Jay D. Aronson, the center’s director. “It’s just a tool. You still need a lot of human judgment.” Once the videos were assembled into a database, SITU narrowed them down to a smaller number — fewer than 20 — that were relevant to the cases. Then, working with collaborators on the ground in Ukraine, they built a virtual model of Instytutska Street. The first version, created using existing site surveys, wasn’t sufficiently detailed, so Nefertari organized surveyors with laser scanners who could capture details at what Samuels calls the “sub-centimeter level.” The scan was so fine that it documented paving-stone patterns and individual leaves of foliage. The surveyors stood in the streets of Kiev with tall white poles to pinpoint exactly where each victim fell. The granularity was necessary to get a fix on not only the victims and the supposed shooters but also on the people holding the cameras. The 70-gigabyte master layout, known as a “point cloud,” was stitched together from 40 individual scans of the street and its environs.
When a bullet breaks the sound barrier, it produces a small sonic boom that registers as an audible crack. For people positioned downrange, the crack arrives a fraction of a second before the thump or blast of the weapon actually firing. In the Maidan case, SITU enlisted a ballistics expert to measure the time that elapsed between the cracks and the thumps. The time difference yielded a maximum and minimum distance between the shooter and the camera, which SITU rendered as a doughnut-shaped “area of interest.”