Finding the location of gunplay used to be a guessing game.
“If shots ring out in the middle of the night, you wake up, you think you hear gunshots,” Thomas Stangrecki, acting chief of the Pittsburgh Bureau of Police, told PublicSource in an August interview. In a call to 911, you might guess where the sound came from, he continued. Police would then arrive, “listen for gunshots, look for evidence … casings or whatever.”
Increasingly, though, Pittsburgh police pinpoint the locations of shots with the help of acoustic sensors provided by the California company ShotSpotter under a $1.2-million-a-year contract. In 2021, the bureau received data on 2,874 incidents in which one or more bullets were fired according to the ShotSpotter sensor system now arrayed across one-third of the city.
“ShotSpotter is really the best means of getting any kind of gauge of what actual gun violence looks like in the city in terms of shots fired,” said Heath Johnson, a bureau crime analyst.
The use of ShotSpotter’s sensors, algorithms and human reviewers to “save lives, solve cases and deter crime” — as its website claims, and as some city leaders echo — has raised questions, locally and nationally, about the fairness of computer-driven justice. Civil libertarians contend that arrests and convictions based on technology that is likely opaque to juries could tilt the scales against defendants who are, constitutionally, presumed innocent.
Those concerns have become part of a case now before the state Supreme Court in which a lawyer for an incarcerated Pittsburgh man argues that ShotSpotter data was inappropriately allowed as evidence in his trial. The Innocence Project and the American Civil Liberties Union have filed briefs in the case, arguing in their filings that ShotSpotter is part of “a larger pattern of flawed science polluting our criminal legal system.”
“You’re seeing GPS, you’re seeing facial recognition, you’re seeing cell phone location, in addition to ShotSpotter,” said Justin Romano, a Downtown-based lawyer who is challenging Angelo Weeden’s conviction for aggravated assault, following a trial that included ShotSpotter-generated evidence. “The defense is fighting a battle with at least one hand tied behind its back if it’s facing this evidence with no means of questioning it.”
ShotSpotter has countered such critiques by calling its technology “simple and transparent.”
The company’s Vice President for Analytics and Forensic Services Tom Chittum, a veteran of federal law enforcement, said there was once skepticism about the evidentiary value of digital photography.
“In time,” he told PublicSource, “these things become accepted.”
‘The way of policing in the future’
Pittsburgh began using ShotSpotter in early 2015 over a 3-square-mile area in the northeast of the city. Ricky Burgess, the city councilman for that area, pitched it as a crime deterrent, a tool for prosecutors and a boost to police officer safety. It was also billed as a sort of automatic 911 system, summoning police and medical assistance to gunshot victims even when nobody picks up the phone.
Public safety officials went to council in 2016 to recommend expansion of the system, saying it was successful in identifying shots that were not reported in 911 calls. A Bureau of Police analyst told council that the bureau used ShotSpotter data to “strategically deploy officers to areas with high gun violence.”
“This is the way of policing in the future,” Burgess said at that meeting. “Cameras and ShotSpotter and using technology for greater accuracy.”
An area by the 2100 block of Arlington Avenue in Arlington. 
An area by 2500 block of Bedford Avenue in Bedford Dwellings. 
An area by the 5400 block of Second Avenue in Hazelwood. 
An area by 7300 block of Fleury Way in Homewood South. 
An area by the 1000 block of Brushton Avenue in Homewood North. 
An area by the 1200 block of Mohler Street in Homewood North. 
An area by the 600 block of Collier Street in Homewood South.
These seven areas reported the most ShotSpotter alerts in Pittsburgh from May 2021 through May 2022. (Photos by Stephanie Strasburg/PublicSource)
ShotSpotter installs acoustic sensors in locations like the tops of buildings, streetlight poles and cellphone towers. The sensors listen for “loud, impulsive sounds,” which can include everything from gunfire to fireworks, car engines and jackhammers.
When at least three sensors detect a sound, ShotSpotter pinpoints an 82-foot radius from which the noise originated using a multilateration algorithm — a method, similar to triangulation, that determines a sound’s location using the times it was picked up by the sensors. The algorithm also filters sounds unlikely to be gunfire.
After the algorithms run, ShotSpotter’s sensors send a recording of the sound to one of the company’s Incident Review Centers, where human reviewers listen to the audio and confirm whether it was likely gunfire. Reviewers make judgment calls based on factors like if there appeared to be multiple rounds of gunshots, multiple shooters or moving shooters.
Once a human reviewer determines the sensors have likely detected gunfire, they send the location of the gunshots to the bureau. The entire process takes roughly one minute. The bureau then sends the information to the county’s 911 dispatching center.
By the start of 2018, city council, then-Public Safety Director Wendell Hissrich and then-Mayor Bill Peduto were sufficiently impressed to invest in expanding the system to 18 out of the city’s total 55.4 square miles. The contract runs through 2025. The annual $1.2 million cost accounts for roughly 1% of the city’s 2022 police budget.
The ShotSpotter expansion reached into each region, with much of the North Side, Hazelwood, West End, Hill District and Homewood covered. Johnson said it might make sense to increase ShotSpotter’s footprint in the city’s South Hills.
The system has become ingrained in the police bureau’s everyday operations. Officers respond to more gunshot alerts from the California-based analysis center than gunshots reported in 911 calls.
With five years’ worth of hindsight, Burgess is still a strong proponent of the system. “It has been very effective and efficient,” the councilman said recently. “It has been an excellent tool.”
But when is ShotSpotter evidence?
The power — and perhaps the peril — of ShotSpotter are at the center of Romano’s case.
Weeden, now 71, of the North Side, was accused of cutting off a former romantic partner’s vehicle with his own, trying to force open her doors and then firing a gun as she pulled away.
In an interview, Romano called it “a circumstantial case. No one ever saw Angelo Weeden with a weapon, discharging a weapon.”
The prosecution wrote in its briefs that a 7-year-old daughter of Weeden’s ex saw a gun.
The trial included testimony from the ex, another passenger and from Weeden’s alibi witnesses, but no ballistics evidence. “And then there’s the ShotSpotter,” Romano said.
A report called an Investigative Lead Summary, generated by ShotSpotter, indicated that two rounds were fired near where Harbison Avenue meets Shadeland Avenue in Brighton Heights at 7:43 p.m. Dec. 15, 2018. That roughly matched the ex’s account.
Over Romano’s objections, Judge Jill Rangos of the Allegheny County Court of Common Pleas allowed the report to be put into evidence. She permitted testimony explaining it by a city detective, without requiring that a ShotSpotter expert detail the technology.
A jury found Weeden guilty of aggravated assault and related charges, and Rangos sentenced him to 10 to 20 years in prison.
Romano has appealed the conviction to the state Supreme Court. He has argued in his briefs that ShotSpotter has “inherent unreliability,” but that Weeden was not able to “confront this evidence in any meaningful way,” which violated his constitutional rights.
The ACLU of Pennsylvania and The Innocence Project, in a jointly filed friend-of-the-court brief, argued that ShotSpotter “relies on an unvetted computer algorithm and the subjective impressions of human reviewers to characterize the audio snippets.”
The Allegheny County District Attorney’s office has responded in filings defending the prosecution that the ShotSpotter evidence didn’t matter much. The jury apparently believed the ex, and not the alibi witness, rendering the report “superfluous.”
Romano, though, said in an interview that ShotSpotter’s report was viewed as corroborating the ex’s account. “I think it’s fair to say that was a key piece of evidence.”
He’s seeking a new trial.
Supreme Court decisions often set precedents, so a ruling on the admissibility of ShotSpotter evidence could affect the way such technology is used in future trials.
Chittum declined to comment on the case, but said he did not see cause for concern.
ShotSpotter can send experts to explain the system at trials, he said. While the government would have to cover costs, he said, “It’s not expensive.”
Romano, though, hopes the case will spur the court to set much-needed standards for the admissibility of high-tech evidence. “A thoughtful decision on this case could have broader implications in the testimonial evidence and the forensic evidence world.”
Rising arrests, more victims found
Pittsburgh police use ShotSpotter for more than just the occasional trial. The bureau claims that the system:
- Alerts the bureau and dispatchers to gunfire faster and more reliably than 911 calls
- Allows officers to safely and strategically deploy
- Helps to locate victims — including 13 over a two-year period — that might otherwise have been discovered too late
- Points officers to evidence that might otherwise have been lost.
A summary provided by the bureau suggests that ShotSpotter is becoming more and more important, as reports, arrests and firearm confiscations attributable to the sensor system in the first six months of this year outpaced 2021.
The bureau also uses ShotSpotter data to produce daily maps of shots fired.
“It may drive some proactive patrols” and investigatory attention, said Stangrecki, adding that manpower deployment hasn’t really changed much since ShotSpotter was deployed.
‘Too many mixed results’
The inspector general of Chicago, a city that uses ShotSpotter over 100 square miles, suggested that the system gives officers “generalized perceptions” of certain areas and “may be substantively changing policing behavior.” The inspector’s August 2021 report warned that ShotSpotter may cause officers to respond to incidents “with little contextual information about what they will find there — raising the specter of poorly informed decision-making by responding members.”
The inspector general’s review of Chicago’s ShotSpotter usage found that during a 17-month period, 9% of ShotSpotter alerts resulted in evidence of a gun-related crime and 2% resulted in an investigatory stop.
ShotSpotter responded that gun recoveries driven by its alerts skewed much higher in some parts of Chicago than others, “indicating that how ShotSpotter is used and hyper-local factors can impact its effectiveness.”
In Pittsburgh, according to data provided by the police bureau, 20% of ShotSpotter alerts in 2021 resulted in any type of evidence collection, 3% resulted in an arrest and 4% led police to a victim in need of assistance.
Even though the vast majority of alerts don’t result in arrests, they may result in the discovery of a shell casing or other piece of ballistics evidence that may tie into another case, Stangrecki said.
The bureau is now “accustomed to ShotSpotter and the accuracy with which it locates where the shooting occurred and the speed with which officers are dispatched,” and losing that tool would reduce efficiency, the acting chief added.
City Controller Michael Lamb told PublicSource he needs to see more data to be convinced of ShotSpotter’s worth.
“To me, I don’t know that we’ve got viable proof that it has really assisted public safety,” Lamb said. “My feeling is that sometime between now and the next time we renew the contract we need to take a deeper dive. … Right now, there are too many mixed results that we’re seeing.”
In their friend-of-the-court brief, the ACLU and the Innocence Project cast doubt on one of ShotSpotter’s core advertised claims: that it advances prosecution of gun-related crimes.
“Even the most cursory glance under the hood reveals a fundamentally subjective, untested system that is dependent on human intervention at every step of its operation,” the brief states.
PublicSource asked both the Bureau of Police and the DA’s office if they knew how many prosecutions were informed by ShotSpotter evidence. Neither had a definitive tally.
Mike Manko, spokesperson for the district attorney, called ShotSpotter “useful to law enforcement to the extent that it acts as an audial 911 system to help locate potential crimes” but “not critical information for the prosecution of those crimes and suspects.”
Stangrecki agreed that even if ShotSpotter evidence was not admissible in court, the evidence the bureau assembles as a result of the system’s alerts would help. “Hopefully,” he said, “everything together is impacting crime.”
Fooled by fireworks?
The ShotSpotter system isn’t perfect.
Pittsburgh police received 36 ShotSpotter alerts on July 4, 2021, about four times more than the daily average. More than half of the alerts came at night as fireworks boomed across the city. Police only generated seven reports of gunshot-related crimes.
Officers are encouraged to respond to every ShotSpotter alert as soon as possible. On high-volume days like July 4, officers use a triage strategy — they prioritize responding to ShotSpotter alerts that correspond with 911 calls and typically investigate the remaining alerts at some point before the end of their shift, whether that’s minutes or hours after the alert was issued, according to Johnson.
This policy means that on days like Jan. 1 — when officers received 185 alerts, nearly three-quarters of which coincided with New Year’s fireworks — police could be investigating dozens and dozens of false alarms. On Jan. 1, police only reported eight gunshot-related crimes.
Officers are instructed to write an investigative report about ShotSpotter alerts only when they recover evidence, identify a victim, find damage to city-owned property, discover another officer is involved or are instructed to by their supervisor. In 2021, 21% of ShotSpotter alerts resulted in police penning a report.
Johnson said that prior to the city installing ShotSpotter, report and arrest rates for gunfire-related crimes were far lower, though he did not have specifics. “I can assure you that your average 911 caller is far less accurate in terms of identifying whether or not it was truly gunfire,” Johnson said.
ShotSpotter itself also takes measures to improve its accuracy during time periods when it anticipates an influx of explosive noises that could be wrongly categorized as gunfire. Email exchanges between ShotSpotter and the bureau show that the company updated its detection algorithms before the Fourth of July to better parse out fireworks.
Stangrecki said gunfire may have occurred at the scene of ShotSpotter alerts, regardless of whether officers deemed the scene notable enough to write a report.
“If they don’t locate any evidence, it doesn’t mean it didn’t occur — it’s just that they didn’t find any evidence,” he said. “Could be late at night, maybe there’s one shot and it’s in a wooded area.”
In 2021, about 20% of ShotSpotter alerts resulted in police collecting evidence, typically ballistic remnants like shell casings. Johnson said that although finding evidence may not yield immediate benefits, it can pay off over time.
“If we can find that the same firearm has been at these different instances, it helps our detectives build investigative leads, put together cases in important ways,” Johnson said.
Placing sensors, mapping shots
Pittsburgh police determined where to station ShotSpotter sensors based on concentrations of 911 calls for gunshots, non-fatal shootings and homicides. In the past year, the sensors have continued to identify neighborhoods like Homewood, Hazelwood and Arlington as hotspots for gunfire.
The ShotSpotter system covers one-third of Pittsburgh, placed based on public safety data. Here’s where it reported gunfire from May 2021 through May 2022. Hover over a point to see the number of alerts during that period. Note: shots in areas not covered by ShotSpotter are not reflected.
Johnson said that ShotSpotter’s precision in pinpointing where gunfire occurs helps with efficiently deploying officers to investigate scenes, especially as police manpower shrinks.
“If you just have a lot more officers aimlessly driving around trying to find where a potential site where a scene might have occurred, it’s just more opportunities for confusion, mistrust, perhaps, stops that generally would have been avoided,” Johnson said.
Even with two-thirds of the city outside of ShotSpotter’s range, Stangrecki doesn’t feel that officers are missing many reports of gunfire in the out-of-range zones.
“In a perfect world, if we had a lot of money, we probably could cover the whole entire city,” Stangrecki said. “I don’t think it makes sense at this point to try to cover the entire city if some areas are not seeing any or a large amount of shots fired.”
Johnson, the bureau’s crime analyst, called any claim that ShotSpotter prevents gun violence “tenuous.” But if it even gets one shooter off the streets, he added, “hopefully there’ll be fewer instances of gun violence perpetrated by that individual.”
Those concerned about the development of “a surveillance state,” he added, should probably focus elsewhere. “The one thing I would want to be surveilled is gun violence. And it does do that.”
Charlie Wolfson is PublicSource’s local government reporter and a Report for America corps member. He can be reached at charlie@publicsource.org or on Twitter @chwolfson.
Amelia Winger is PublicSource’s health reporter with a focus on mental health. She can be reached at amelia@publicsource.org.
Rich Lord is PublicSource’s managing editor. He can be reached at rich@publicsource.org or on Twitter @richelord.
This story was fact-checked by Matt Maielli.
