The Algorithmic Panopticon: A Comprehensive Analysis of Chicago’s Strategic Subject List and the Failures of Predictive Policing

Full report

On this page

Introduction

In the early 2010s, the intersection of big data analytics and municipal law enforcement birthed a new paradigm in civic governance: predictive policing. While many jurisdictions across the United States experimented with place-based algorithms designed to forecast the geographic distribution of property crimes or burglaries, the Chicago Police Department (CPD) embarked on a substantially more ambitious and highly controversial person-based approach. Funded initially by a $3.8 million grant from the United States Department of Justice’s Bureau of Justice Assistance, the CPD entered into a collaborative partnership with researchers at the Illinois Institute of Technology (IIT) to develop an algorithm capable of calculating the exact probability that a specific individual would become a "party to violence" (PTV)—meaning they would be either a victim or a perpetrator in a shooting or homicide1. This initiative manifested as the Strategic Subject List (SSL), colloquially referred to within the department and the media as the "heat list." For nearly a decade, the SSL and its eventual successor, the Crime and Victimization Risk Model (CVRM), served as the highest-profile person-based predictive policing systems in the country3. The system was ostensibly designed and marketed to the public as a public health intervention tool aimed at directing municipal social services to those at the highest statistical risk of urban gun violence. However, its operational reality within the CPD transformed it into a mechanism for heightened surveillance, targeted enforcement, and enhanced prosecution, fundamentally altering how police interacted with marginalized communities5. The narrative surrounding the Strategic Subject List is one of technological solutionism colliding with the complex, systemic realities of deep-rooted urban violence. Through successive algorithmic iterations, millions of dollars in federal funding, extensive national press coverage, and subsequent civil rights litigation, the program ultimately collapsed under the weight of its empirical inefficacy, constitutional scrutiny, and profound internal data mismanagement. In November 2019, following a scathing review by the municipal inspector general, the CPD quietly decommissioned the predictive risk models2. This report provides an exhaustive, peer-level analysis of the theoretical underpinnings, algorithmic architecture, operational deployment, empirical evaluations, and ultimate systemic failures of Chicago's premier predictive policing experiment.

Epidemiological Foundations: The Contagion Model of Gun Violence

To understand the algorithmic architecture of the Strategic Subject List, one must first examine the sociological and epidemiological research that provided its intellectual foundation. The conceptual framework for Chicago’s predictive policing model was heavily influenced by the network science research of urban sociologist Dr. Andrew Papachristos. Through extensive research into urban gun violence in cities such as Chicago and Boston, Papachristos demonstrated that homicides and non-fatal shootings do not occur randomly across geographic areas, nor are they evenly distributed across demographic profiles7. Instead, gun violence behaves akin to a blood-borne pathogen, concentrating intensely within highly insular, identifiable social networks8.

Social Distance and Co-Offending Networks

Through the rigorous analysis of co-offending networks—defined as individuals arrested together for the same offense—Papachristos found that violence spreads through social contagion8. In a seminal 2014 study of a high-crime neighborhood on Chicago's West Side, research revealed that 41% of all gun homicide victims in a community of 82,000 people belonged to a singular social network comprising merely 4% of the neighborhood's population9. The concentration of violence was even more pronounced in other jurisdictions; a parallel study of 763 individuals in Boston demonstrated that 85% of all gunshot injuries within the sample occurred within a single, highly dense social network7. An exhaustive epidemiological analysis of 138,163 individuals arrested in Chicago between January 2006 and March 2014 determined that social contagion accounted for a staggering 63.1% of the 11,123 gunshot violence episodes during that period8. The probability of gunshot victimization was found to be directly related to an individual's network distance from other gunshot victims. Each network association removed from a prior victim historically reduced the odds of gunshot victimization by 25%7. This established a critical criminological premise: indirect exposure to violence through peer networks exerts a massive independent effect on an individual's risk profile, operating above and beyond individual risk factors such as age, prior criminal activity, or residential geography7. The implication of this research for municipal policy was profound. By viewing violence as a localized epidemic transmitted through specific social interactions, interventions could theoretically be targeted at the highly concentrated nodes of a network to arrest the contagion8.

Algorithmic Translation at the Illinois Institute of Technology

While Papachristos’s research identified the structural contours of urban violence, he was not directly involved in the coding or creation of the CPD’s proprietary predictive algorithm10. The task of translating these nuanced sociological insights into a functional, automated predictive policing tool was undertaken by Dr. Miles Wernick, the Motorola Professor of Electrical and Computer Engineering at the Illinois Institute of Technology, operating alongside a team of researchers1. Dr. Wernick’s professional background did not lie in criminology, but rather in advanced predictive analytics for military target recognition and the automated medical diagnoses of dementia via brain scans12. Wernick approached the creation of the Strategic Subject List much like a computer-aided diagnosis in the medical field, frequently comparing the algorithm's predictive mapping to identifying potentially dangerous anomalies in a mammogram12. The theoretical justification for the algorithm rested on the premise that, much like smoking is a statistically demonstrable risk factor for lung cancer, specific patterns in an individual's criminal history and social associations operate as quantifiable risk factors for future violent crime10. The primary novelty of the IIT approach lay in attempting to evaluate the risk of violence in an allegedly unbiased, purely quantitative manner, thereby identifying individuals on the precipice of violent involvement before a tragic event occurred10.

Shift from Network Analysis to Actuarial Risk

Between 2012 and 2019, the Chicago Police Department utilized six sequential iterations of the Party to Violence (PTV) predictive models. Versions 1 through 5 operated under the Strategic Subject List (SSL) nomenclature, producing a continuous "risk score" on a scale ranging from 0 to 500, with 500 representing the most acute risk of violent involvement1. Version 6, launched in January 2019, was rebranded as the Crime and Victimization Risk Model (CVRM), which shifted from a continuous numerical score to a tiered system (Tiers 1 through 5\) intended to estimate the exact probability of violent involvement over the subsequent 18 months2. A critical component of the model's public relations strategy was the explicit exclusion of standard demographic variables. To preempt immediate accusations of racial profiling or geographic redlining, the algorithm deliberately omitted race, gender, ethnicity, and home address1. Instead, the early iterations of the SSL heavily weighted the social network component inspired by Papachristos. Version 1.0 explicitly factored in the number of times a subject was co-arrested with an individual who was already designated as a prior Party to Violence, attempting to map the contagion pathways10. However, maintaining real-time social network mapping for hundreds of thousands of individuals proved computationally and operationally unwieldy. As the model evolved, the explicit reliance on complex co-arrest network mapping was systematically deprecated. By the time Version 5 of the SSL was deployed, the algorithm had fundamentally transformed from a dynamic network-contagion model into a traditional actuarial risk assessment tool, relying almost entirely on historical individual-level arrest attributes2.

Variable Selection and Model Inputs

The algorithms were generated using raw, de-identified data provided by the CPD’s Information Services Division (ISD). The IIT researchers processed this data to calculate the risk outputs, which the CPD then re-identified to map the scores to specific individuals using their Internal Record (IR) numbers2. Version 5 of the SSL relied on eight distinct attributes derived from an individual's criminal record.

Table 1: Variables utilized in Version 5 of the Chicago Police Department Strategic Subject List1. The eventual shift from the SSL to the CVRM (Version 6\) in 2019 sought to refine the output by removing gang affiliation and narcotics arrests—variables that had drawn intense civil liberties criticism for serving as proxies for race17. The CVRM stratified individuals into discrete risk tiers corresponding to the probabilistic likelihood of becoming a victim or an arrestee in a shooting or homicide within an 18-month window.

Table 2: Crime and Victimization Risk Model (CVRM) Risk Tiers and corresponding empirical probabilities15.

The "Dirty Data" Paradigm and Tech-Washing

The fundamental flaw in the Strategic Subject List—and a persistent vulnerability in predictive policing broadly—is the reliance on historical police data as the ground truth for training algorithms. This phenomenon is extensively documented in legal and sociological literature as the "dirty data" problem. As noted in comprehensive analyses by the New York University Law Review, police data does not objectively measure the incidence of crime within a society; rather, it measures the geographic and demographic distribution of police activity5. The Chicago Police Department possesses a well-documented, decades-long history of racially biased policing, unconstitutional investigatory stops, and the imposition of informal arrest quotas in minority neighborhoods5. Because the SSL algorithm heavily utilized variables such as prior arrests for weapons, narcotics, and violent offenses, it inherently absorbed the historical biases embedded in those discretionary arrests2.

The Algorithmic Feedback Loop

This dynamic created a vicious and self-sustaining feedback loop. Historically, the CPD heavily patrolled Chicago's South and West sides, generating disproportionate arrests among Black and Latino residents for minor infractions, loitering, or via pretextual investigatory stops. This "dirty data" was fed directly into the SSL's training set, which subsequently assigned the highest risk scores to young men residing in those specific neighborhoods. The artificially elevated scores were then used by police command staff to justify further localized surveillance and escalated charging, generating fresh arrest data that further confirmed the algorithm's initial biased hypothesis5. The demographic consequences of this feedback loop were stark. While the algorithm explicitly excluded race as a programmed variable, it operated as a highly efficient proxy for it. Investigations conducted following freedom of information lawsuits found that 85% of individuals with the highest risk scores were African American, and an astonishing 56% of all Black men in Chicago aged 20 to 29 possessed an SSL score5. By relying on arrests rather than convictions, the system also penalized individuals who were swept up in mass enforcement actions but were never found guilty of any crime. It is extremely difficult, if not impossible, for machine learning systems trained on this data to detect and separate objective crime indicators from data generated by corrupt or biased policing practices18. Consequently, the predictive technology served primarily to "tech-wash" historical prejudice, laundering systemic bias through a veneer of mathematical objectivity and presenting it to field officers as empirical gospel17.

Operationalizing the Algorithm: Public Health Rhetoric vs. Punitive Reality

The most profound paradox of the Strategic Subject List was the schism between its public framing and its operational deployment. The CPD, alongside its academic and municipal partners, frequently framed the algorithm to the media as a progressive public health mechanism designed to direct limited social services to potential victims4. Yet, simultaneous departmental directives explicitly instructed officers to use the algorithmic output to escalate punitive law enforcement actions and surveillance5.

The Custom Notifications Program

The primary overt mechanism for operationalizing the SSL was the Custom Notifications program, a pillar of the CPD’s broader Violence Reduction Strategy (VRS)15. Modeled loosely on the "focused deterrence" and Group Violence Intervention (GVI) strategies pioneered in Boston, Custom Notifications involved targeted face-to-face interventions23. When an individual's SSL score breached a designated threshold (typically 250 or above on the 500-point scale), a coalition comprising police officers, social workers, and community leaders—often coordinated in partnership with the John Jay College of Criminal Justice—would conduct an unannounced home visit4. The intervention delivered a dual message: an offer of social services (such as job training, cognitive behavioral therapy, or housing assistance) combined with a severe, highly specific warning that the individual was under heightened law enforcement scrutiny. They were explicitly informed that any future violent activity by them or their associates would result in maximum federal and state sentencing, as well as aggressive asset forfeiture23. Between 2013 and 2016, the CPD reportedly conducted over 1,400 of these visits4. While the rhetoric surrounding Custom Notifications emphasized care and community support, the underlying threat of enforcement was omnipresent and codified in policy. Internal CPD directives, specifically Special Order S10-05, dictated that if an individual received a Custom Notification and subsequently engaged in any criminal behavior, the department was mandated to pursue the "highest possible charges" against them5. This transformed the notification from an offer of help into a procedural prerequisite for enhanced prosecution, leveraging an algorithmically generated score to deny leniency or diversion programs in the future.

Escalated Enforcement: T.R.A.P. and Exceptional Response Plans

The punitive nature of the SSL was further formalized through the Targeted Repeat-Offender Apprehension and Prosecution (T.R.A.P.) program. Under Special Order S10-06, T.R.A.P. established a direct pipeline between the CPD and the Cook County State’s Attorney’s Office. The program's explicit goal was to pursue enhanced prosecution to detain, convict, and incarcerate individuals identified as having a high propensity for violent crime2. An individual's SSL score served as a primary criterion for determining their eligibility for this escalated prosecutorial enforcement2. The SSL was also integrated into tactical operations during periods of heightened violence. When the CPD initiated an Exceptional Response Plan (ERP) following a spike in neighborhood shootings, tactical commanders were explicitly directed by CPD-21.616 checklists to immediately implement the T.R.A.P. program and conduct aggressive Custom Notifications within the affected area22. Furthermore, the integration of the SSL into daily policing operations created pervasive, inescapable surveillance mechanisms. SSL scores were populated into the Subject Assessment and Information Dashboard (SAID) and Caboodle, a geospatial mapping application used by officers on mobile data terminals in their cruisers2. The scores were heavily utilized in Strategic Decision and Support Centers (SDSCs)—real-time crime centers embedded in police districts equipped with predictive analytics software27. These centers allowed commanders to prioritize patrols, monitor social associations in real-time, and dictate mission assignments based on algorithmic outputs27. Consequently, the model functioned less as a triage tool for municipal social workers and more as a sophisticated digital dragnet for preemptive policing.

Collateral Interagency Surveillance

The reach of the SSL extended far beyond the patrol officers of the CPD. Despite public assurances that the data was closely guarded, the CPD shared SSL scores and CVRM tiers with a vast array of external entities. The Cook County Sheriff's Office utilized the data to feed into its Sheriff's Anti-Violence Effort (SAVE) program, which mandated cognitive behavioral therapy for incarcerated 18-to-24-year-olds from violent neighborhoods29. More alarmingly, the CPD pushed SSL data into the Law Enforcement Agencies Data System (LEADS), granting unfettered access to over 500 external agencies, including federal immigration authorities, neighboring suburban departments (such as Cicero and Elgin), and various state prosecutors20. This intergovernmental sharing occurred without formal oversight, usage guidelines, or audit trails, meaning an individual's algorithmic risk profile followed them across jurisdictional boundaries, potentially influencing bail hearings, traffic stops, and immigration proceedings far beyond Chicago's city limits2.

Empirical Evaluations and Efficacy

Despite the influx of millions of dollars in federal funding and the rapid, widespread implementation of the program across the city's police districts, independent empirical evaluations of the Strategic Subject List yielded highly critical conclusions regarding its efficacy in reducing violence.

The RAND Corporation Study

In 2016, the RAND Corporation published a comprehensive quasi-experimental evaluation of the initial 2013 SSL pilot program (Version 1.0)32. Researchers Jessica Saunders, Priscillia Hunt, and John S. Hollywood utilized highly rigorous methodologies to isolate the program's effects. They employed Autoregressive Integrated Moving Average (ARIMA) models to assess the intervention's impact on city-level homicide trends, controlling for seasonality and historical patterns. To evaluate individual-level impacts, they utilized propensity score matching, comparing the outcomes of the 426 individuals on the initial SSL against a mathematically matched control group of individuals with similar criminal histories who were not on the list32. The findings of the RAND evaluation were unequivocal: the SSL failed to achieve its primary objective. Individuals placed on the SSL were neither more nor less likely to become the victim of a homicide or a non-fatal shooting than the matched comparison group32. Furthermore, the introduction of the SSL had no statistically significant impact on the aggregate city-level homicide rate32. However, the RAND study uncovered a highly concerning, unintended secondary effect: individuals on the SSL were significantly more likely to be arrested for a shooting than the control group32.

The Conflation of Threat vs. Risk

The RAND researchers diagnosed the core conceptual failure of the program: the algorithm completely failed to differentiate between "high-risk" individuals (those likely to be victimized due to their proximity to violence or their social environment) and "high-threat" individuals (those actively perpetrating violence against others)32. Because the model collapsed these distinct, albeit occasionally overlapping, profiles into a single numerical risk score, field commanders were left with a list of names but no actionable intelligence regarding the specific nature of the risk4. Absent specific, tailored guidance on preventive treatments, the empirical evidence suggests that police officers simply utilized the SSL as a directory of leads to close active shooting cases32. Rather than preventing crime before it occurred, the algorithm functioned as a suspect generation engine, fundamentally corrupting the program's stated preventive public health aims. Furthermore, the RAND study highlighted the statistical near-impossibility of predicting rare events, a common pitfall in predictive analytics known as the base rate fallacy. While individuals placed on the SSL were mathematically calculated to be 233 times more likely to be involved in violence than the average Chicago resident, their absolute rate of homicide victimization was still only 0.7% over a 12-month period32. This exceptionally low base rate meant that even if the police could perfectly intervene and save every single life on the list, the intervention would yield minimal absolute reductions in the citywide homicide rate32. To achieve even a 5 percent drop in the city's homicide rate, the researchers noted that enormous leaps in both prediction accuracy and intervention effectiveness would be mathematically necessary33.

Transparency Battles and Freedom of Information

The veil of operational secrecy surrounding the SSL was ultimately pierced not by internal transparency initiatives, but by aggressive freedom of information requests and civil litigation. Lawsuits filed by the Chicago Sun-Times (led by journalists Mick Dumke and Frank Main) and digital rights organizations such as Lucy Parsons Labs forced the CPD into protracted legal battles4. Ultimately, the courts compelled the CPD to release a de-identified version of the SSL database encompassing arrest records from August 2012 to July 20164. The release of this data allowed independent researchers to audit the system. The resulting analyses fundamentally dismantled the CPD's meticulously constructed narrative that the SSL was a highly targeted, strategic list of active violent offenders.

The Illusion of a "Strategic" List

The released dataset revealed a staggering 398,684 individuals possessing an SSL score3. This did not represent a targeted subset of dangerous offenders; rather, it represented the entire universe of individuals who had been arrested or fingerprinted in Chicago over a four-year period3. The CPD had previously maintained in press briefings that an SSL score of 250 or above was the threshold that placed an individual on the department's "radar" for heightened scrutiny4. However, rigorous data analysis by the technology and justice advocacy group Upturn revealed that 287,404 individuals—more than two-thirds of the entire database—possessed a score over 2503. For a municipal police department comprising roughly 12,000 officers, the concept of strategically monitoring nearly 300,000 "high-risk" individuals was operationally absurd3. More damningly, the data exposed catastrophic flaws in the model's inclusion criteria. Upturn's analysis demonstrated that more than one-third of the individuals on the list (133,474 people) had never been arrested for any crime3. Within this cohort, 127,513 individuals had never been arrested and had never been a victim of a shooting, yet approximately 90,000 of them were assigned a score above the 250 high-risk threshold3. Furthermore, 1,551 individuals who had never been arrested or shot were inexplicably flagged as gang-affiliated, generating high scores3. The presence of hundreds of thousands of un-arrested, non-victimized individuals on a predictive policing hit list remained entirely unexplained by the CPD, suggesting profound algorithmic overreach, catastrophic database contamination, or the silent inclusion of individuals solely because they had been fingerprinted for employment or background checks3.

Age Dominance and the "Age-Out" Theory

By reverse-engineering the algorithmic weights used to generate the scores, independent analysts discovered that the complex machine-learning model was overwhelmingly dominated by a single, highly simplistic variable: age. Age accounted for approximately 89% of the variance in an individual's SSL score3. The model systematically and aggressively penalized youth, virtually guaranteeing that any young person entering the criminal justice system for even minor infractions would be flagged as high-risk. Consequently, the algorithm did little more than digitally reproduce the long-established criminological "age-out-of-crime" theory, a basic sociological tenet which posits that individuals naturally desist from criminal behavior as they enter their 30s3. The millions of taxpayer dollars spent on advanced machine learning ultimately yielded a system that simply sorted younger Chicagoans to the top of a risk matrix, offering no novel strategic value while imposing heavy surveillance burdens on the city's youth3.

Systemic Collapse: The Office of Inspector General Advisory

The fatal blow to the predictive risk program was delivered not solely by external critics, but by severe internal auditing. In January 2020, the City of Chicago Office of Inspector General (OIG) released a scathing, comprehensive advisory report detailing the programmatic, administrative, and ethical failures of the SSL and its successor, the CVRM2.

Data Decay and Algorithmic Negligence

The OIG determined that the risk scores and tiers actively used by officers were fundamentally unreliable due to severe data decay. The audit revealed that between August 2016 and January 2019, the CPD failed to update the SSL scores entirely. For nearly two and a half years, police officers and external agencies relied on static, obsolete risk assessments that did not account for individuals' aging, their desistance from crime, or even their deaths2. The OIG identified numerous individuals whose profiles indicated ongoing criminal activity occurring after their documented dates of death2. Furthermore, the model scored individuals based merely on the occurrence of an arrest over a four-year period, completely ignoring the judicial disposition of those arrests2. Individuals who were arrested but never convicted—or who had their charges immediately dropped due to lack of evidence—retained permanent high risk scores. Because SSL scores were used to influence harsher interventions under programs like Custom Notifications and T.R.A.P., the CPD was effectively penalizing citizens for crimes they were never legally convicted of committing, a blatant violation of core due process principles2.

The "Gang Database" Conflation and Operational Ignorance

The OIG found that the CPD failed to provide its sworn personnel with basic training on how the algorithm functioned. The only training materials available to command staff in 2019 had been drafted years prior for Version 3 of the SSL, which utilized entirely different variables than the active Version 52. This knowledge vacuum led to dangerous operational assumptions on the street. Because older versions of the SSL heavily targeted gang networks, field officers erroneously assumed the list was synonymous with a gang database. One district commander testified to the OIG that they believed 95% of the individuals on the SSL were gang members, and a District Intelligence Officer claimed everyone with a score was a confirmed gang member2. In reality, only 16.3% of the individuals on the active Version 5 SSL were confirmed gang members2. This systemic misunderstanding virtually guaranteed that officers would interact with hundreds of thousands of individuals on the list with unwarranted prejudice and suspicion, treating non-affiliated citizens as violent gang operatives. This conflation was exacerbated by the CPD's notoriously flawed internal gang databases. A separate OIG review of the CPD's gang information systems found that over 15,000 individuals designated as gang members had no specific gang membership listed20. Furthermore, auditors found that officers had entered highly unprofessional and biased occupations for individuals on gang arrest cards, including "SCUM BAG," "BUM," "CRIMINAL," and "LOOSER \[sic\]"20. Relying on this highly subjective, often derogatory data to feed a predictive algorithm completely compromised the integrity of the predictive outputs.

Decommissioning the Program

Faced with the expiration of federal grant funding, the departure of the IIT research team (whose grants had expired), and the mounting, undeniable weight of civil rights criticism, the CPD possessed no internal capability to manage or maintain the complex algorithms2. Consequently, on November 1, 2019, the CPD formally decommissioned the PTV risk model program2. The OIG recommended a total administrative purge of the system, requiring the CPD to revise all directives referencing the SSL or CVRM, completely remove risk scores from all databases, inform command staff of the discontinuation, and immediately revoke external access2. The abrupt termination of a program that had consumed millions of dollars and dictated citywide deployment strategies for years served as a stark admission of systemic failure. Similar reckonings occurred in other major jurisdictions; notably, the Los Angeles Police Department shut down its comparable L.A. Strategic Extraction and Restoration (LASER) Program in 2019 after its own Inspector General found a complete lack of reliable data to justify the program37.

Constitutional and Legal Deficiencies

Beyond empirical inefficacy and administrative negligence, the deployment of the SSL raised profound constitutional questions. Legal scholars and civil rights advocates highlighted multiple areas where the predictive risk models likely ran afoul of established American jurisprudence. First, the use of an algorithmic score to justify law enforcement stops or investigations tests the strict boundaries of the Fourth Amendment. A high SSL score, derived largely from historical and non-particularized data (such as age, network associations, or merely being a victim of a crime), cannot constitutionally constitute the "reasonable, articulable suspicion" required for a Terry stop, nor the "probable cause" required for an arrest39. When officers utilize the list to target individuals preemptively, they risk conducting unconstitutional searches and seizures based on statistical probability rather than individualized suspicion. Second, the system deeply implicated the Due Process Clause of the Fourteenth Amendment. By placing individuals on a "heat list" shared widely among law enforcement and municipal agencies—and associating that list with heightened criminality, gang affiliation, and escalated prosecution—the state deprived individuals of protected liberty interests. These interests include information privacy and freedom from state-sanctioned stigma. The CPD enacted this deprivation without providing any notice or opportunity to be heard20. Individuals could not view their scores, understand the variables contributing to their presence on the list, or appeal their inclusion to correct erroneous police data20. Finally, the undeniable racial disparities produced by the algorithm—whereby 56% of young Black men in the city were assigned scores—demonstrated a clear racially disparate impact17. Under frameworks like the Illinois Civil Rights Act, state or municipal policies that disproportionately burden protected demographic classes without a compelling, narrowly tailored justification are legally vulnerable40. By relying on "dirty data" generated by historically biased policing practices, the SSL codified that bias into a mathematical mandate, violating the core tenets of civil rights protections.

Conclusion

The Chicago Police Department’s decade-long experiment with the Strategic Subject List and the Crime and Victimization Risk Model stands as a definitive cautionary tale in the annals of modern law enforcement technology. Born from the well-intentioned sociological observation that violence acts as a social contagion, the translation of that nuanced theory into an automated actuarial algorithm stripped away vital socioeconomic context, replacing it with a sterile, highly punitive risk score. The failures of the SSL were manifold and systemic. Empirically, it completely failed to predict or prevent victimization, serving instead to increase arrests and perpetuate the very cycles of incarceration it ostensibly sought to break32. Technologically, it was built on a foundation of "dirty data," laundering historical police biases through machine learning to justify the continued over-policing of Chicago’s Black and Latino communities5. Administratively, it was managed with severe negligence, utilizing static data, offering no training to end-users, and operating without basic privacy safeguards or procedural justice mechanisms2. Predictive policing technologies cannot circumvent the requisite hard work of community building, nor can they cure the systemic inequalities that breed urban violence. The deployment of complex statistical models atop flawed, historically biased data environments does not yield objective truth; it yields automated prejudice. As law enforcement agencies worldwide continue to explore artificial intelligence and data analytics, the collapse of Chicago’s predictive risk models serves as a definitive mandate: algorithms must be subjected to rigorous, independent empirical validation, strict access controls, and unwavering constitutional oversight before they are ever utilized to govern the lives and liberties of the public.

Works cited

1. City of Chicago \- Strategic Subject List \- Historical \- Catalog \- Data.gov, https://catalog.data.gov/dataset/strategic-subject-list-historical

2. CITY OF CHICAGO OFFICE OF INSPECTOR GENERAL ADVISORY CONCERNING THE CHICAGO POLICE DEPARTMENT'S PREDICTIVE RISK MODELS, https://igchicago.org/wp-content/uploads/2020/01/OIG-Advisory-Concerning-CPDs-Predictive-Risk-Models-.pdf

3. How strategic is Chicago's Strategic Subjects List? | Upturn, https://www.upturn.org/work/how-strategic-is-chicagos-strategic-subjects-list/

4. How strategic is Chicago's “Strategic Subjects List”? Upturn investigates. | by Brianna Posadas | Equal Future | Medium, https://medium.com/equal-future/how-strategic-is-chicagos-strategic-subjects-list-upturn-investigates-9e5b4b235a7c

5. DIRTY DATA, BAD PREDICTIONS: HOW CIVIL RIGHTS VIOLATIONS IMPACT POLICE DATA, PREDICTIVE POLICING SYSTEMS, AND JUSTICE \- NYU Law Review, https://www.nyulawreview.org/wp-content/uploads/2019/04/NYULawReview-94-Richardson\_etal-FIN.pdf

6. OIG Releases Advisory on the Chicago Police Department's Predictive Risk Models, https://igchicago.org/2020/01/23/oig-releases-advisory-on-the-chicago-police-departments-predictive-risk-models/

7. Social Networks and the Risk of Gunshot Injury | HOPLOFOBIA.INFO, https://www.hoplofobia.info/wp-content/uploads/2014/05/Social-Networks-and-the-Risk-of-Gunshot-Injury.pdf

8. Modeling Contagion Through Social Networks to Explain and Predict Gunshot Violence in Chicago, 2006 to 2014 \- \- Ben Green, https://www.benzevgreen.com/wp-content/uploads/2019/02/17-jamaim.pdf

9. Predictive Policing in Chicago \- Success or Failure? \- Smith Hanley Associates, https://www.smithhanley.com/2017/02/14/predictive-policing-in-chicago/

10. Chicago's Strategic Subject List, a.k.a. “Heat List” · Predictive Policing \- Upturn, https://teamupturn.gitbooks.io/predictive-policing/content/systems/chicago.html

11. Professor Miles Wernick's Predictive Policing Algorithm Mentioned in Chicago Tribune, https://www.iit.edu/news/professor-miles-wernicks-predictive-policing-algorithm-mentioned-chicago-tribune

12. Miles Wernick, ACE Motorola Chair Professor, and his team help CPD predict crime., https://www.iit.edu/news/miles-wernick-ace-motorola-chair-professor-and-his-team-help-cpd-predict-crime

13. Crime and Victimization Risk Model (CVRM)\* | Chicago Police Department, https://www.chicagopolice.org/wp-content/uploads/FACT-SHEET-Crime-and-Victimization-Risk-Model-1.pdf

14. The Exclusionary Rule in the Age of Blue Data \- Scholarship@Vanderbilt Law, https://scholarship.law.vanderbilt.edu/cgi/viewcontent.cgi?article=1826\&context=vlr

15. Violence Reduction Strategy (VRS) \- Chicago Police Department, https://www.chicagopolice.org/violence-reduction-strategy-vrs/

16. Going Inside The Chicago Police Department's 'Strategic Subject List' \- CBS News, https://www.cbsnews.com/chicago/news/going-inside-the-chicago-police-departments-strategic-subject-list/

17. GARBAGE IN, GOSPEL OUT \- NACDL, https://www.nacdl.org/getattachment/eb6a04b2-4887-4a46-a708-dbdaade82125/garbage-in-gospel-out-how-data-driven-policing-technologies-entrench-historic-racism-and-tech-wash-bias-in-the-criminal-legal-system-09142021.pdf

18. DIRTY DATA, BAD PREDICTIONS: HOW CIVIL RIGHTS VIOLATIONS IMPACT POLICE DATA, PREDICTIVE POLICING SYSTEMS, AND JUSTICE \- New York County Defender Services, https://www.nycds.org/wp-content/uploads/2019/03/3/Dirty%20Data%20Bad%20Predictions.pdf

19. DIRTY DATA, BAD PREDICTIONS: HOW CIVIL RIGHTS VIOLATIONS IMPACT POLICE DATA, PREDICTIVE POLICING SYSTEMS, AND JUSTICE \- NYU Law Review, https://www.nyulawreview.org/wp-content/uploads/2019/04/NYULawReview-94-Richardson-Schultz-Crawford.pdf

20. Review of the Chicago Police Department's “Gang Database”, https://igchicago.org/wp-content/uploads/2019/04/OIG-CPD-Gang-Database-Review.pdf

21. The Lessons of Chicago's Disastrous “Crime Prediction” Experiment, https://filtermag.org/chicago-crime-prediction/

22. EXCEPTIONAL RESPONSE PLAN (ERP) CHECKLIST \- Chicago, https://directives.chicagopolice.org/forms/CPD-21.616.pdf

23. Custom Notifications: Individualized Communication in the Group Violence Intervention \- National Network for Safe Communities, https://nnscommunities.org/wp-content/uploads/2017/10/GVI\_Custom\_Notifications\_Guide.pdf

24. based violence reduction strategies \- NICJR, https://nicjr.org/files/galleries/Effective\_Community\_Based\_VR\_Strategies\_Final\_2021.pdf

25. Early Findings on Chicago's Gang Database, https://soc.uic.edu/wp-content/uploads/sites/197/2018/07/Tracked-Targeted-0217-r.pdf

26. Predictive Policing in Practice: A Case Study of Chicago's Strategic Subject List \- Scholarship @ Claremont, https://scholarship.claremont.edu/cgi/viewcontent.cgi?article=5092\&context=cmc\_theses

27. UChicago Crime Lab: Friend or Foe? \- Chicago Maroon, https://chicagomaroon.com/32821/grey-city/uchicago-crime-lab-friend-foe/

28. Police Surveillance in Chicago \- redshiftzero, https://redshiftzero.com/policesurveillance/tactics/predictive-policing.html

29. Sheriff's Anti-Violence Effort – SAVE | Cook County Sheriff's Office, https://cookcountysheriffil.gov/departments/cook-county-department-of-corrections/programs-and-services/save/

30. Sheriff Tom Dart \- Cook County Sheriff's Office |, https://cookcountysheriffil.gov/sheriff-tom-dart/

31. Stopping Crime with Digital Advantage | Illinois Institute of Technology, https://www.iit.edu/news/stopping-crime-digital-advantage

32. Predictions put into practice: a quasi-experimental evaluation of Chicago's predictive policing pilot \- NACDL, https://www.nacdl.org/getattachment/9d276b57-0d3f-477a-90fb-5a00c003edff/rand-ssl-study.pdf

33. Pitfalls of Predictive Policing \- RAND Corporation, https://www.rand.org/pubs/commentary/2016/10/pitfalls-of-predictive-policing.html

34. Governmental Automated Decision-Making And Human Rights: Reconciling Law And Intelligent Systems \[1 ed.\] 3031481240, 9783031481246, 3031481259, 9783031481253, 9783031481277 \- DOKUMEN.PUB, https://dokumen.pub/governmental-automated-decision-making-and-human-rights-reconciling-law-and-intelligent-systems-1nbsped-3031481240-9783031481246-3031481259-9783031481253-9783031481277.html

35. Urban Violence: Security, Imaginary, Atmosphere 179363730X, 9781793637307 \- DOKUMEN.PUB, https://dokumen.pub/urban-violence-security-imaginary-atmosphere-179363730x-9781793637307.html

36. Datasets for journalists \- GitHub Pages, https://voanews.github.io/datasets-for-journalists/

37. UNITED STATES OF AMERICA \- Oxford Institute of Technology and Justice, https://www.techandjustice.bsg.ox.ac.uk/hubfs/US-AI%20Justice%20Atlas-pdf.pdf?hsLang=en

38. Surveil and Predict: A Human Rights Analysis of Algorithmic Policing in Canada \- The Citizen Lab, https://citizenlab.ca/wp-content/uploads/2020/09/To-Surveil-and-Predict.pdf

39. Your criminal FICO score \- Naval Postgraduate School, https://calhoun.nps.edu/server/api/core/bitstreams/86d42ba6-e9ca-49af-80f2-196ba4756f6b/content

40. Constraining Big Brother: The Legal Deficiencies Surrounding Chicago's Use of the Strategic Subject List, https://legal-forum.uchicago.edu/print-archive/constraining-big-brother-legal-deficiencies-surrounding-chicagos-use-strategic