Hot People: Chicago’s Strategic Subject List, LAPD LASER, and Pasco County’s Intelligence-Led Policing Program

Full report

On this page

Executive findings and research method

Chicago, Los Angeles, and Pasco County built three of the best-documented American experiments in person-based predictive or “intelligence-led” policing. Each converted historical police information into a designation about an identifiable person and then made that designation available for operational use. Beyond that common architecture, however, the programs predicted different things, used different analytical methods, and imposed radically different consequences.

Chicago’s Strategic Subject List, or SSL, and its successor Crime and Victimization Risk Model, or CVRM, were principally models of future involvement in gun violence, combining the probability of being shot with the probability of being treated as an offender. They were not clean offender-prediction models. Los Angeles’s LASER chronic-offender component was an explicitly offender-oriented, manually administered point system in which analysts ranked people based on arrests, supervision status, alleged gang membership, gun history, and police contacts. Pasco County’s original Intelligence-Led Policing program most directly claimed to identify people likely to continue or begin offending and then subjected selected people, their families, and sometimes minors to repeated visits and enhanced enforcement. Pasco later overlaid a federally funded, somewhat more conventional focused-deterrence program that excluded minors and offered services, but it retained the language of predicted reoffending and “increased accountability.”

The central finding is that the designation mattered most where it altered police behavior most concretely. Chicago’s 2013 pilot produced substantially more documented police contacts for selected people but no demonstrated reduction in their likelihood of being murdered, shot, or arrested for murder. LAPD’s chronic-offender bulletins told officers whom to notice, investigate, locate, visit, check for warrants, and subject to parole or probation compliance activity, although the Inspector General found that LAPD could rarely attribute particular stops or arrests to LASER. Pasco’s operational doctrine went further: selected people could receive repeated, suspicionless home visits, code-enforcement attention, warrant checks, intensive monitoring, and a declared “zero tolerance” posture. In Pasco, the list was not merely informational; it supplied the reason for recurring government contact.

The programs also illustrate three different forms of circularity:

• Chicago used arrests, victimization, weapons history, and recent police-recorded activity to estimate future police-recorded violence involvement. Selection then increased contact-card documentation by approximately 50 percent and overall recorded interaction by approximately 39 percent compared with matched people, potentially enriching the same class of administrative data from which future risk assessments would be constructed.
• Los Angeles awarded points for “quality police contacts,” distributed chronic-offender bulletins to patrol and specialized units, and encouraged officers to develop additional information. An encounter generated because of the bulletin could therefore become a recorded contact supporting the person’s continued priority.
• Pasco instructed deputies to conduct periodic checks partly to cultivate information about selected people and their social networks. Visits, field interviews, citations, arrests, and reports could then confirm that a person was frequently involved with police, including when the person appeared in a report as a victim or witness.

This report applies a stricter test of predictive validity than the agencies generally did. “A selected person was later arrested” is not sufficient. Arrest is partly an outcome of police attention, and a high arrest rate among heavily monitored people does not demonstrate that the model identified conduct that would otherwise have been detected at the same rate. A valid evaluation needs an appropriate denominator, an equivalent untreated comparison group, a predeclared outcome, and separation between actual offending, victimization, police contact, and arrest. Chicago’s RAND evaluation came closest to satisfying those requirements. The early LASER studies did not. No public evaluation establishes the predictive validity of Pasco’s original prolific-offender or at-risk-youth lists.

The report’s legal conclusions are correspondingly bounded. Inclusion on a list is not automatically a constitutional violation. Reputation alone ordinarily does not establish a protected liberty deprivation under Paul v. Davis; an equal-protection challenge to facially neutral variables usually requires proof of discriminatory purpose, not disparity alone; and municipal liability under 42 U.S.C. § 1983 requires a policy or custom that caused an underlying constitutional injury. But list-driven stops, searches, coercive questioning, home intrusions, retaliation, family disruption, or unequal treatment can supply the additional constitutional injury that the designation alone may lack.

Research was conducted through August 2, 2026. The evidentiary hierarchy is: official audits and directives; technical documents and public datasets; court filings and settlement instruments; grant-funded or peer-reviewed evaluations; public-records disclosures; and investigative reporting. Allegations in complaints are identified as allegations. A settlement is not treated as a judicial finding of liability unless the agreement or court order expressly says so. Facts for which coefficients, source code, denominators, or retention records were not publicly available are labeled unavailable rather than inferred.

Comparative program matrix

Chicago: Strategic Subject List and Crime and Victimization Risk Model

Origin, objective, and institutional structure. CPD began using person-level violence-risk models in 2012. Five iterations were called the Strategic Subject List; a sixth was called the Crime and Victimization Risk Model. Illinois Institute of Technology researchers received deidentified CPD data, developed the models, and calculated scores or tiers. CPD then reidentified the results and connected them to named people. RAND was retained to evaluate versions one, five, and six. Federal support totaled approximately $3.8 million across grants associated with the model-development period, and the peer-reviewed pilot evaluation identifies National Institute of Justice award 2009-IJ-CX-K114.

The stated endpoint was becoming a “party to violence” during the next 18 months. That term combined at least two analytically and morally distinct events: being shot and being treated as the offender in a shooting. This distinction is not semantic. A model that assigns a young shooting survivor a high score may be detecting repeat victimization, proximity to violence, or police exposure—not an intention to commit violence. CPD’s public descriptions nevertheless sometimes used offender-focused language, contributing to the widespread impression that the SSL was a “bad guy list.” The later technical assessment presented to the National Academies characterized the system as largely a victim-prediction model whose interventions were never adequately matched to the reasons a person was at risk.

Reconstructed scoring logic. Version one was primarily a social-network model. It counted first-degree co-arrest links to people who later became homicide victims and second-degree links through a co-arrestee, generally using a five-year window. IIT transformed those measures into a ranked risk score; CPD’s Deployment Operations Center then subjectively reviewed the output, meaning the final 426-person list was not simply the 426 mathematically highest scores.

By SSL version five, the model had moved toward an individual-attribute design. The public variables were:

Race and sex were fields in the released dataset but were not direct inputs. The score was scaled from 0 to 500. Public reverse engineering suggested that age explained an unusually large share of score variation, but CPD did not publish the complete version-five coefficient table or executable source code. Consequently, a precise person-level formula cannot be responsibly reconstructed from official materials alone.

CVRM removed gang affiliation and narcotics arrests from the version-five set and produced risk tiers rather than a nominal 0–500 score. The model still predicted combined victim-or-offender involvement over 18 months. CPD’s Inspector General found that the department did not possess the underlying algorithm when the program was under review and depended on IIT to run or explain it. That is a major governance defect: a police department was taking operational action on outputs from a model it did not fully control, reproduce, or audit.

Universe, thresholds, and denominator problems. Later versions were not a compact “list” in an ordinary sense. Anyone with at least one arrest in the preceding four years could receive a score or tier, regardless of whether the arrest involved violence. OIG identified 399,412 SSL records in July 2018 and 313,513 CVRM records in March 2019. People known to CPD solely as shooting victims could be excluded if they had never been arrested and therefore lacked the required identification number. Thus an ostensible victimization model was built on an arrest-conditioned population and could fail to score a victim with no arrest record.

CPD publicly indicated that people scoring at least 250 drew heightened attention. An advocacy analysis of the historical release found 287,404 of 398,684 records above that level, approximately 72.1 percent. That is not necessarily the fraction subjected to a Custom Notification or enforcement action; it demonstrates instead that “selection” had multiple meanings. A person could be in the scored universe, above an attention threshold, in a district operational subset, referred for notification, or placed in another enforcement initiative. CPD never maintained a sufficiently clear treatment ledger to support a complete outcome analysis across those stages.

The 2013 pilot was far narrower. Its 426 subjects represented approximately 0.049 percent of the 873,281-person arrest-history file examined by RAND. But using the entire arrest file as the denominator is not the same as using the 17,754 people with at least one first- or second-degree link to a homicide victim, which was the more relevant high-risk comparison pool. This illustrates why agencies can make a program look precise by choosing a very broad denominator or look inclusive by choosing a narrow one.

Intervention ladder.

OIG found no consistently trained, centrally supervised intervention protocol. Officers were not uniformly told what a score meant, what they could lawfully do because of it, or how to distinguish someone predicted to be victimized from someone predicted to offend. RAND similarly found little executive follow-up and almost no practical direction in the pilot beyond taking some action or increasing contact.

Did selection change police behavior? Yes, at least in the pilot. Seventy-seven percent of SSL subjects received at least one contact card during the following year, averaging 8.6 cards; 60 percent were arrested at least once, and nearly 90 percent had some recorded interaction with CPD. Against propensity-weighted matched comparisons, list membership increased the probability of having a contact card by about 50 percent and the probability of any recorded interaction by about 39 percent. It did not significantly increase the overall probability of arrest or incapacitation.

That finding is critical. Even though the list did not demonstrably reduce violence, it changed the administrative record around selected people. More contacts can mean more observations, more field-interview documentation, more association data, more opportunities to discover warrants or contraband, and more entries that later analysts interpret as evidence of continuing risk. RAND’s mediation analysis did not find that additional contacts statistically explained the increased shooting-arrest result in the pilot, so it would be inaccurate to say the study proved a self-fulfilling arrest mechanism. It did, however, prove the first link in such a mechanism: designation increased recorded police attention.

Predictive validity and outcomes. During the pilot follow-up, Chicago recorded 405 homicides. Three victims had been on the 426-person SSL. The homicide-victim precision of the list was therefore approximately 0.7 percent, while recall was approximately 0.74 percent—three of 405 citywide homicide victims. Both numbers require caution: the model’s stated endpoint was broader than homicide victimization, and many city residents were outside its arrest-conditioned eligible population. Still, the figures show that the pilot could not plausibly prevent a large share of Chicago homicides through interventions limited to those 426 people.

For the broader combined endpoint, 29 of 426 subjects, or 6.8 percent, were recorded as either a shooting victim or a shooting arrestee. The unadjusted rate among linked but nonselected former arrestees was 3.2 percent. That difference is not an estimate of treatment benefit or pure model precision because it combines victimization and arrest and because selected and unselected people differed at baseline. After propensity weighting and regression adjustment, list placement did not significantly reduce murder victimization, shooting victimization, or murder arrest. It was associated with a 2.88-fold greater likelihood of shooting arrest, based on only nine selected subjects.

The citywide interrupted-time-series model initially appeared to show a homicide reduction, but placebo dates revealed that the decline began before SSL’s introduction. RAND therefore concluded that the program did not cause a further reduction beyond the existing trend. This is the appropriate response to a common evaluation error: a before-and-after decline is not program impact if the trend was already changing.

For later SSL and CVRM versions, public materials do not supply the full confusion matrices needed to calculate precision, recall, false-positive rate, false-negative rate, calibration error, or area under the receiver-operating curve at operational thresholds. The 0–500 score was a ranking scale, not a publicly calibrated probability—“400” did not mean a 40 percent chance of a shooting. OIG found no evaluation of versions two through four and serious shortcomings in the implementation and assessment of versions five and six. Any claim that those versions were “accurate” without threshold-specific event rates and matched baselines is unsubstantiated.

Data quality, retention, and removal. Scores were not reliably refreshed. OIG found that SSL scores remained static from August 2016 until January 2019, nearly two and a half years. The records contained stale or inconsistent information, including missing ages, incidents dated after a person’s death, and conflicting weapons data. A dynamic risk model that is not rerun is no longer a current prediction; it is a historical label that can continue influencing officers after its factual basis has changed.

There was no meaningful subject-facing procedure to correct a gang designation, explain a mistaken arrest linkage, contest identity matching, or request removal based on rehabilitation or time. Decommissioning did not erase the historical record: Chicago continues to publish a deidentified historical SSL dataset. Public archival retention is valuable for accountability, but operational retention requires different rules. The available record does not establish that all person-identifiable scores and downstream references were purged from every investigative or arrest file.

Disclosure, closure, and current status. Much of the model became public only after journalism, public-records disputes, litigation, and OIG scrutiny. CPD announced its intention to decommission the program in August 2019 and ended it November 1, 2019, after the grant period closed. As of August 2, 2026, official data sources describe SSL as historical, and no evidence establishes that CPD reinstated SSL or CVRM under those names. The lasting lesson is not that Chicago successfully predicted future shooters. It is that a victim-and-offender risk ranking migrated into enforcement systems without a stable operational theory, complete model custody, reliable updating, or a defensible process for affected people.

Los Angeles: LASER, chronic-offender bulletins, and Palantir-supported workflows

Origin and declared purpose. Los Angeles Strategic Extraction and Restoration—LASER—grew from a Bureau of Justice Assistance Smart Policing Initiative. LAPD and the consulting firm Justice & Security Strategies began research and development around 2009 and deployed the program in Newton Area in 2011. It ultimately operated in 16 of LAPD’s 21 geographic Areas. The program paired place-based “LASER zones” with a person-based Chronic Offender Program. Its rhetoric was more explicitly incapacitative than Chicago’s: official materials described identifying offenders, “extracting” them from targeted zones, restoring peace, and removing the anonymity of gang members and gun offenders.

The federal-grant record identified in the LAPD Inspector General’s review included a 2009 BJA award of $499,959 and a 2014 award of $400,000. Associated expenditures through September 2018 totaled approximately $899,551. LAPD paid Justice & Security Strategies approximately $413,142 for research and technical work. Palantir applications cost LAPD roughly $1.8 million annually during the period reviewed, although the OIG stressed that the platform supported many functions beyond LASER and that the full Palantir cost should not be attributed to the chronic-offender program alone.

Scoring logic and selection. The original point sheet was straightforward:

A 2017 revision placed greater emphasis on recency and repetition, awarding five points for each violent-crime arrest in the preceding two years and five points for each gun-related incident in the same period. Areas were generally instructed to identify 12 active chronic offenders and maintain additional backups.

This was not a statistical model in the ordinary predictive sense. There was no published logistic equation, fitted probability, calibration curve, or validated risk threshold. The points encoded policy judgments: an alleged gang designation was made numerically equivalent to probation status or a qualifying gun history, and repeated police contacts incrementally raised priority. Analyst discretion then determined whom to work up and whom to keep active. The “top 12” was a resource quota, not a threshold derived from an acceptable false-positive rate. Two people with identical histories could be treated differently depending on Area, analyst practice, and who else happened to be in the local candidate pool.

“Quality police contact” was especially problematic. LASER documentation did not consistently define it. Personnel told the OIG that it generally meant a field-interview contact with a gun-related narrative or other intelligence value, but the absence of a stable definition permitted local variation. Because field interviews are initiated and documented by police, the variable measured both a person’s conduct and the intensity of police observation.

Palantir’s role. Palantir was not shown to have invented the LASER point formula. Its important role was infrastructural: it integrated records, accelerated searches, supported analyst workups, generated or populated chronic-offender bulletins, and made accumulated information readily visible. Analysts could combine arrest histories, field-interview cards, gang information, addresses, vehicles, warrants, recent contacts, associates, and descriptive information into a portable person dossier. Investigative reporting described bulletin production that could be reduced from roughly an hour of manual work to minutes using Palantir-supported workflows.

That distinction matters for vendor accountability. The principal civil-liberties risk was not necessarily a proprietary black-box prediction. It was the ability of an integration platform to make a loosely governed police heuristic operationally powerful. A five-line point sheet can be more consequential when it is connected to a searchable dossier, displayed at roll call, available in vehicles, and updated through daily field activity. The vendor shaped speed, scale, visibility, and institutional memory even if the point values originated with LAPD and JSS.

Bulletins and instructions. A chronic-offender bulletin could contain a photograph, name, aliases, birth date, address, physical description, tattoos, Social Security or identification information, gang designation, arrest history, parole or probation status, vehicles, warrants, locations frequented, and recent police contacts. The bulletin warned that the person was not necessarily wanted and that the document alone could not supply probable cause to detain. It was nevertheless explicitly designed to sensitize officers to the individual.

Commanding officers could assign patrol, gang-enforcement, parole-compliance, or other personnel to a subject. Recommended steps included:

Some notification letters referred people to organizations such as the Gang Reduction and Youth Development network, Volunteers of America, or Community Build. Those referrals support LAPD’s argument that the program evolved from pure incapacitation toward deterrence and assistance. The OIG nevertheless found that most records did not establish sustained service engagement or even successful contact with the selected person.

Implementation breakdown. At suspension in August 2018, the LASER database contained 637 people: 234 marked active and 403 inactive. Because the database stored status as a checkbox without reliable activation and deactivation dates, auditors could not reconstruct how long individuals were targeted or whether removal rules were consistently applied. One duplicate reduced the distinct active-person demographic count to 233.

The point system was not uniformly followed. Thirty-seven active subjects and 75 inactive subjects had zero recorded points. Active and inactive scores ranged from 0 to 101, and approximately 59 percent had 25 points or fewer. Among records with detailed calculations, 44 percent had zero or one violent-crime arrest; about half had no gun-related arrest; nearly 10 percent had no quality contact; and some accumulated dozens of contacts. Several Areas acknowledged relying on detective or patrol referrals rather than point totals.

These findings defeat the claim that LASER created a uniform, objective ranking. It created a common vocabulary and database around practices that remained substantially discretionary. A score could legitimate a nomination, but a nomination could also occur without a score. That asymmetry is dangerous in accountability terms: the department could claim data-driven objectivity when defending selection while relying on officer judgment when the data did not support the choice.

Outcome evidence and appropriate baselines. Early program presentations reported that 87 of 125 chronic offenders identified between August 2011 and January 2012 were arrested by June 2012, a nominal “arrest precision” of 69.6 percent. Among 88 people scoring 15 points or more, 60 were arrested, or 68.1 percent. But among the 37 people scoring below 15, 27 were arrested, or 72.9 percent—a higher rate than the higher-scored group. A later cohort brought the combined total to 189 selected people, of whom 112, or 59 percent, had been arrested at least once.

Those statistics do not validate the score. First, the outcome was arrest for any reason, not verified violent offending. Second, subjects were actively sought by police. Third, there was no comparable untreated group of people with similar histories and police exposure. Fourth, the apparent lack of monotonicity—the lower-scored group had a slightly higher arrest percentage than the group scoring 15 or more—undercuts the idea that the threshold separated meaningfully different risk levels. No recall rate can be calculated because the program did not report how many future arrestees or violent offenders were outside the list. No false-positive rate can be calculated because there is no equivalent nonselected denominator. No calibration measure is possible because the scores were not probabilities.

The place-based evaluation was more favorable but cannot be attributed to the person list alone. In Newton, LAPD deployed approximately 7.9 additional officer-hours per day across five hotspot corridors. JSS’s interrupted-time-series work reported reductions in gun crime, homicide, robbery, and violent crime. Other official summaries highlighted a roughly 5.4-per-month decline in Part I violent crime and substantial homicide changes. But LASER simultaneously changed patrol dosage, hotspot presence, parole activity, and chronic-offender enforcement. Without a factorial design, it is impossible to determine whether any reduction came from locations, persons, general visibility, regression to the mean, or contemporaneous changes.

The LAPD OIG found that early evaluation evidence was limited to the initial Newton implementation and that no adequate subsequent studies demonstrated LASER’s effectiveness across the expanded program. It also found that attempted locates often failed, most selected people had few documented program-specific contacts, and stops or arrests generally could not be causally attributed to the designation. That does not prove LASER had no effect; it means LAPD lacked the treatment records necessary to know.

Disparities. Of 233 active chronic offenders, 53.2 percent were Latino, 30.9 percent Black, 14.6 percent white, and 1.3 percent categorized as other; 93.1 percent were male. LAPD’s more appropriate internal baseline was its 82,808 Part I violent-crime arrests from 2012 through 2018: 49 percent Latino, 34 percent Black, 12 percent white, and 5 percent other, with 79 percent male. Against that arrest baseline, the racial composition roughly tracked the population already arrested for serious violence, while men were more overrepresented.

The arrest baseline is methodologically preferable to the city population if the question is whether LASER added disparity after conditioning on violent arrest. It is not normatively neutral. If arrest and gang databases already reflect unequal enforcement, matching them reproduces rather than cures that structure. The relevant fairness questions therefore include both conditional disparity—whether similarly situated arrested people were selected at different rates—and cumulative disparity—how policing, arrest, gang labeling, field interviews, and LASER together distributed state attention. LAPD’s data permitted only a partial answer.

Feedback and self-fulfilling records. LASER presents the clearest formal feedback mechanism. Police contacts added points; points increased priority; priority encouraged more police observation; and observation generated contacts. The warning that a bulletin was not itself reasonable suspicion reduced but did not eliminate that mechanism. Officers can initiate consensual encounters, observe from public locations, check databases, or prioritize legal enforcement without conducting a formally unlawful stop. Those actions can still create an asymmetric record around a selected person.

The point system also confused prediction with institutional capacity. Being on probation or parole increased a person’s score partly because supervision supplied additional search or compliance powers. That variable therefore predicted not only conduct but police ability to intervene. A high “success” rate measured by arrest could reflect legal access rather than underlying offending.

Termination, retention, and present status. LAPD suspended chronic-offender database entries in August 2018 while OIG reviewed the program and terminated LASER in April 2019. A December 20, 2019 chief’s notice directed all personnel to cease creating, using, or disseminating chronic-offender bulletins and LASER-zone materials, remove them from use, and archive existing documents under retention policies. Archival language is not a purge requirement; historical bulletins could remain in departmental files even after operational use ended.

LAPD subsequently adopted a Data-Informed Community-Focused Policing framework emphasizing community engagement, victims, problem solving, and place-based analysis. Critics argued that it preserved parts of LASER’s philosophy under a less controversial name, while official materials presented it as a different model. The defensible conclusion through August 2, 2026 is narrow: formal LASER zones and Chronic Offender Bulletins remain discontinued; data-driven policing and integrated police intelligence remain part of LAPD practice, but the record does not establish continuation of the same top-12 chronic-offender list under the same operational rules.

Pasco County: Intelligence-Led Policing, prolific offenders, and children

Origin and philosophy. Pasco County’s program began after Sheriff Chris Nocco took office in 2011. Its manual described Intelligence-Led Policing as a strategy for focusing on problem people, places, and groups and identifying the relatively small number of people believed responsible for a disproportionate amount of crime. The adult prolific-offender list, youth-risk analysis, real-time intelligence functions, social-network analysis, school-resource-officer intelligence, and later focused-deterrence grants were related but not identical programs. Treating them as one algorithm obscures important changes.

The original ILP list was the most coercive component. The manual defined a prolific offender as a person of any age meeting a threshold based on three years of arrests and suspicions for burglary, theft, or violence. A person generally needed at least two qualifying arrests to enter the candidate pool. Candidates were then ranked by offense count and severity, alleged gang affiliation, and recency. The top 100 active people after analyst vetting were designated prolific offenders.

A later public-records filing reported that the broader pool contained approximately 1,800 people in one reporting period. Thus the active-list selection rate was about 5.6 percent of that pool, although the pool itself was conditioned on Pasco records and the figure may not apply to every year. The exact coefficients and complete historical person-level scores have not been published in a form permitting independent replication.

Reconstructed adult scoring logic. The 2018 manual and related filings indicate this general structure:

1. Determine whether the person had at least two qualifying arrests within three years.
2. Add weights for the number and seriousness of arrests and suspected offenses.
3. Account for time since the most recent arrest and spacing between events.
4. Add enhancements for alleged gang affiliation, supervision violations, failures to appear, or similar justice-system events.
5. Consider repeated appearances in incident reports, potentially including appearances as a victim, witness, suspect, or other involved person.
6. Consider social relationships and associations.
7. Rank the eligible pool by points.
8. Have intelligence personnel vet the candidates and select approximately 100 active subjects.

The exact point assigned to each offense and enhancement is not reliably available in the official public record cited here. The logic nonetheless reveals why the term “prolific offender” could exceed adjudicated conduct. Arrests counted regardless of ultimate disposition; suspicions could count without charge; social ties could increase exposure; and appearing repeatedly in police reports did not necessarily mean perpetrating crime.

Youth-focused scoring. Pasco’s at-risk-youth process was a separate mass-screening system. The Sheriff’s Office combined data from the school district’s Early Warning System, the Florida Department of Children and Families’ child-welfare database, and sheriff records. More than 30,000 middle- and high-school students were analyzed, and approximately 420 were placed in a flagged category at the time reported—a selection rate around 1.4 percent. The office did not systematically notify the children or their parents.

The youth system assessed 16 categories and classified children as “on track,” “at risk,” “off track,” or “critical.” Inputs included D or F grades, multiple absences, discipline, exposure to abuse or neglect, domestic violence, parental incarceration, child-welfare involvement, runaway reports, curfew issues, drugs or alcohol, custody disputes, police contacts, and associates. A D grade or several absences could contribute to risk.

These variables predicted several different constructs at once. Academic disengagement and trauma may be associated with future justice-system contact, but they are also evidence that a child needs educational, disability, mental-health, housing, or family support. Abuse history is especially troubling as a police risk feature: it converts harm suffered by a child into information that can justify greater law-enforcement attention. Sheriff’s officials defended the program as a way to identify mentorship needs, victimization, self-harm, and resources, while the manual repeatedly discussed children as possible future offenders. Both descriptions are supported by the record.

School-resource officers and intelligence analysts could use the scores and underlying information. Reports described officers asking children about offenses, associates, or neighborhood intelligence and recording field-interview information. There was no independent validation showing that the four risk categories accurately predicted offending, victimization, suicide risk, or service need, and those outcomes should not have been conflated in one operational score.

Intervention ladder under original ILP.

The Sheriff’s Office disputed the characterization that its policy required harassment and argued that contacts were lawful, crime-preventive, and sometimes supportive. Tampa Bay Times reporting, former-deputy accounts, plaintiff testimony, and later settlement recitals documented repeated visits and intrusions that went substantially beyond a single consensual knock.

The distinction between policy and implementation became central in litigation. A federal court concluded that the ILP philosophy and written policies were not facially unconstitutional, while preserving for trial the question whether the policies were unconstitutional as applied to the plaintiffs. The eventual settlement recited that checks at plaintiffs’ homes exceeded the ordinary implied license to knock, including through nighttime visits; substantially interfered with family association; and interfered with liberty interests. At the same time, the settlement expressly stated that plaintiffs asserted liability and the sheriff denied it.

The grant-funded VPO/NPO retooling. Pasco received a BJA Strategies for Policing Innovation award in October 2018 to implement a more structured focused-deterrence model. Its design drew on Tampa Police Department methods and, notably, LAPD’s earlier chronic-offender work. The project went live December 2, 2020. It created separate Violent Prolific Offender and Narcotic Prolific Offender lists and required at least two relevant arrests in three years. Unlike original ILP, it excluded people under 18.

The later model considered criminal history, gang membership, involvement as a suspect or victim in a shooting, drug distribution or use, overdose information for the narcotics list, social-network centrality, and contacts with Pasco deputies. The program assigned each listed person to a detective, coordinated with state and federal prosecutors, contemplated Project Safe Neighborhoods referrals, and incorporated Behavioral Health Intervention Team personnel and service offers.

Notification was more formal. Letters told recipients that they had been selected for increased accountability and warned that declining assistance would make them the focal point of further law enforcement and full prosecution if criminal activity continued. A BHIT detective could offer lists of mental-health, substance-use, housing, employment, and related resources. The National Policing Institute found, however, that service capacity was limited, partner organizations often knew little about the program, some providers excluded people under supervision, and staff generally contacted selected people only once or twice.

Civil-rights organizations raised concerns that federal grants were underwriting a renamed continuation of Pasco’s earlier practices. Records identify BJA Smart Policing award FAIN 2018WYBX0004, a related block grant FAIN 2018YSBX0034, and a STOP School Violence award FAIN 2019YSBX0040 associated with youth and school-safety activities. BJA issued a stop-work order in August 2021 and commissioned an external assessment. The record supports a careful conclusion: federal funding did not create Pasco’s original ILP system, which predated the grant, but it financed a descendant program, encouraged formal scoring and prosecutor coordination, and triggered federal review after public controversy.

Evaluation of the later program. The National Policing Institute expressly limited its assessment to grant-funded VPO/NPO activity between December 2020 and August 2021. It did not evaluate the original ILP prolific-offender checks or the youth list. It found no evidence that VPO/NPO list placement increased arrests, traffic stops, or pedestrian stops; recorded contacts tended to be lower during list periods. The evaluators suggested possible deterrence but stressed that the findings were preliminary and could not establish effectiveness. The short implementation period, pandemic, stop-work order, small sample, and limited successful contact all weakened causal inference.

Black residents constituted 16.4 percent of the VPO list and 10.7 percent of the NPO list, compared with approximately 8.5 percent of Pasco County’s residential population. Those are representation ratios of roughly 1.93 and 1.26, respectively. Evaluators cautioned that the lists were small and found no evidence that race, or race interacted with list status, predicted enforcement contact among listees. That finding does not validate the original ILP or youth program and does not answer whether Black residents were more likely to enter the qualifying arrest pool.

No public confusion matrix establishes precision, recall, calibration, false-positive rate, or false-negative rate for either original adult ILP or youth scoring. For the later VPO/NPO lists, the evaluation measured enforcement contacts rather than verified reoffending and did not report sufficient untreated outcome data to calculate standard classification metrics. “Contacts went down while listed” is compatible with deterrence, incarceration, relocation, regression to the mean, reduced police activity during the pandemic, or measurement changes.

Children, disability, and racial disparity. Federal school data cited in the investigation showed that Black students and students with disabilities in Pasco were roughly twice as likely as comparison students to be suspended or referred to law enforcement. Because school discipline and law-enforcement involvement fed youth-risk assessment, preexisting disparities could be operationalized as future risk.

Reports documented selected children with autism, developmental disabilities, competency issues, or serious trauma. Disability can affect attendance, behavior, communication with officers, impulse control, and response to repeated visits. A scoring process that interprets those manifestations as criminogenic without accommodation risks discrimination under the Americans with Disabilities Act and Section 504, apart from any constitutional claim. It also risks converting schools’ failure to provide appropriate services into a police-risk signal.

FERPA generally restricts disclosure of personally identifiable education records but contains exceptions for school officials, emergencies, subpoenas, and law-enforcement units under defined circumstances. The key legal issue is not whether police may ever receive school information; it is whether broad, recurring extraction of grades, attendance, and discipline for general predictive policing fits the stated exception and the school district’s legitimate educational or safety purpose. FERPA itself has limited private enforceability, but noncompliant disclosure can expose a district to federal administrative action and support state-law, privacy, or constitutional theories. The public record documents disputes and policy revisions but not a final merits judgment that every transfer violated FERPA.

Disclosure and records litigation. Pasco operated with unusually little public notice. Parents were not systematically told that grades, abuse histories, discipline, or child-welfare records contributed to sheriff risk categories. Public understanding came primarily through the Tampa Bay Times’s “Targeted” investigation and subsequent records requests.

CAIR-Florida requested deidentified demographic and scoring records in April 2021. The Sheriff’s Office resisted production for nearly two years, ultimately producing records shortly before a mandamus hearing. In May 2025, Florida’s Second District Court of Appeal held that the records had been unlawfully withheld and that the requester was entitled to fees. The decision is significant independently of the predictive-policing merits: accountability for an algorithmic program is impossible when the government denies even deidentified information needed to evaluate who was selected.

Settlement and current status. Four residents’ federal suit challenged repeated checks and family targeting under the original ILP program. In December 2024, the parties settled for $105,000. The agreement states that Pasco “absolutely and unambiguously” discontinued the process of designating prolific offenders under the 2018 manual and making repeated, suspicionless visits to them or their families, and promised not to return to a substantially similar process. The court retained a role in enforcement.

It would be inaccurate to say the settlement adjudicated the entire philosophy unconstitutional. The court had found the philosophy and written policies not facially unconstitutional; the agreement resolved as-applied claims before trial; and the sheriff denied liability. It is also inaccurate to dismiss the agreement as a mere nuisance payment: the document contains express factual recitals that the plaintiffs’ home checks exceeded the implied license and interfered with family association and liberty, coupled with a binding nonrecurrence commitment.

As of August 2, 2026, no evidence shows that Pasco revived the prohibited prolific-offender designation-plus-repeated-check system. The Sheriff’s Office continues youth-diversion, school-safety, intelligence, and data-driven operations, which are not necessarily equivalent. The legally material current status is that the specific process challenged in Taylor v. Nocco was discontinued and cannot return in substantially similar form under the settlement.

Comparative validity, disparities, feedback, grants, and vendors

Prediction targets must be separated. The three systems are often grouped together as “predictive policing,” but they did not predict a common outcome:

A program cannot claim success by switching endpoints after implementation. If Chicago predicted “party to violence,” it should report victimization and alleged offending separately. If LAPD predicted chronic violent offending, an arrest for a technical supervision violation is not equivalent to a violent true positive. If Pasco’s youth tool predicted service needs, a service referral is relevant; if it predicted crime, verified offending against a comparable group is relevant.

Available performance metrics.

The Chicago false-negative figure requires particular care. Saying that the pilot “missed 99.26 percent of homicide victims” is mathematically true against the citywide denominator but can overstate model failure because SSL did not score every resident and predicted more than homicide victimization. The fair conclusion is narrower: the 426-person intervention reached too few eventual victims to produce a large citywide prevention effect, exactly as RAND concluded.

Conversely, LAPD’s 69.6 percent arrest statistic is not a 69.6 percent precision estimate for violent offending. The list was composed of people with substantial prior police records; many would have been arrested under ordinary practice; selected officers were trying to locate them; and the result included any arrest. Without similarly situated nonselected people, the number cannot answer whether LASER predicted better than baseline or caused crime reduction.

Racial disparity. Directly excluding race does not make a model race-neutral in effect. Arrests, gang designations, field interviews, weapons enforcement, school discipline, neighborhood patrol, and victimization are socially patterned. A model may reproduce those patterns even when race is absent from its input table.

Chicago’s pilot cohort was 77 percent African American and 95.8 percent male. Its weighted comparison group was similarly composed because RAND matched people on the underlying risk structure. That helps isolate treatment effect but does not answer whether the pipeline creating the candidate population was racially just.

LAPD’s active list was 30.9 percent Black and 53.2 percent Latino. Compared with violent-crime arrestees, those shares were not dramatically different; compared with the general city population, they would appear far more disparate. The arrest baseline is appropriate for measuring incremental selection disparity but not for assessing the legitimacy of the full enforcement pipeline.

Pasco’s later VPO list was 16.4 percent Black against an 8.5 percent county population share. The external evaluator appropriately cautioned that small numbers made the estimate unstable. For the original youth and adult programs, years of records resistance prevented timely independent computation of selection, disability, and outcome disparities.

Self-fulfilling data. A full self-fulfilling prophecy would require evidence of the following chain:

\[ \text{selection} \rightarrow \text{more police attention} \rightarrow \text{more recorded events} \rightarrow \text{higher future score or continued selection}. \]

Chicago empirically established the second term: selection caused more recorded contacts. It did not establish that contact mediated shooting arrests, and the long period during which scores were not refreshed interrupted immediate score feedback. The structural risk nevertheless remained because arrest and activity data were model inputs and downstream police documents could preserve the designation.

LASER embedded the chain directly in its scoring rules. A “quality contact” added a point; the bulletin encouraged officers to notice and document the subject; and the database accumulated those contacts. No published evaluation tested how many points were generated after designation or whether removal would have occurred absent list-driven contacts.

Pasco paired selection with deliberate recurring checks and intelligence cultivation. Its scoring also credited repeated incident-report appearances and associations. It therefore presented the strongest opportunity for list-induced data to justify continued targeting. The 2024 settlement’s reference to repeated, suspicionless visits confirms that, at least for the plaintiffs, police attention was not merely a passive consequence of independent calls for service.

How grants shaped local practice. Federal grants supplied money, legitimacy, consultants, evaluation language, and cross-jurisdictional diffusion.

Chicago’s NIJ and BJA support helped fund the IIT partnership, technical infrastructure, and RAND evaluation. Grant-funded experimentation encouraged six model versions, but program sustainability lagged: CPD ended with inadequate model custody, outdated scores, and no durable operational doctrine. The grant produced research capacity without ensuring institutional competence or due-process protections.

Los Angeles’s Smart Policing awards financed JSS research and Newton implementation. Grant presentations then converted early local results into a transportable model for other agencies. Pasco’s later scoring system expressly drew from LAPD and Tampa methods, illustrating how a federally sponsored “promising practice” can travel before the original jurisdiction completes a rigorous person-level evaluation.

Pasco shows federal oversight working late but materially. Its original ILP program existed independently of federal funding. The 2018 award encouraged a more formal focused-deterrence structure with service referrals and an adult-only rule. After reporting and advocacy raised concerns, BJA stopped grant activity and commissioned an independent review. That response reduced additional program exposure, but it occurred after local practices had operated for years.

Future grants should therefore condition payment on publication of the target outcome, eligible population, feature definitions, treatment protocol, civil-rights impact assessment, subject-notice rules, retention limits, and an independent evaluation design established before deployment. Grants should not treat “data driven” as a substitute for constitutional and scientific review.

How vendors and institutional partners shaped practice. IIT exercised substantive model control in Chicago; CPD supplied the police data and operational response. The risk was dependency on a university partner whose algorithm was not fully transferred to the government using it. RAND’s independent work exposed implementation failure but did not govern daily use.

JSS helped design and evaluate LASER, while Palantir supplied integration and workflow infrastructure. The same consultant’s involvement in design and early evaluation raises an independence concern even without evidence of bad faith: evaluators invested in implementation can be more likely to use favorable before-and-after measures. Palantir’s influence was different—it enabled dossiers and rapid dissemination at scale.

Pasco’s later program involved BJA, the National Policing Institute, Tampa-derived scoring concepts, behavioral-health personnel, prosecutors, and prospective social-service partners. The external assessment found many intended community and service partnerships were weak or underdeveloped. A named partnership is not an intervention; effectiveness depends on whether selected people can voluntarily access sufficiently funded services without coercion.

Constitutional accountability, oversight chronology, policy, and final judgment

Procedural justice. All three systems failed core procedural-justice tests to varying degrees. People generally did not receive a clear statement of the designation, evidence used, predicted outcome, duration, consequences, or method of correction. Notification letters, where used, were not equivalent to due process because they did not provide an independent decision-maker or meaningful ability to contest an erroneous gang label, arrest linkage, identity match, school record, or association.

Procedural due process under the Fourteenth Amendment requires more than unfairness; it requires a deprivation of a protected liberty or property interest without adequate process. A secret score that produces no tangible consequence may fail that threshold under Paul v. Davis. But stigma combined with repeated home intrusions, altered supervision, denial of a benefit, heightened prosecution, or substantial interference with family liberty can present the required “plus.”

Fourth Amendment. A score or bulletin cannot itself create individualized reasonable suspicion. LAPD’s own bulletin warned officers that it was not a sole basis for detention or probable cause. Officers could lawfully observe public conduct, check a valid warrant, or conduct a consensual conversation, but a stop required reasonable suspicion tied to present or completed wrongdoing, and a search required consent, a warrant, a valid supervision condition, or another recognized exception.

The home receives greater protection. Florida v. Jardines recognizes a limited social license to approach by the customary route, knock, wait briefly, and leave absent an invitation to remain. Pasco’s settlement expressly recited that checks at the plaintiffs’ residences exceeded that license, including through nighttime visits. Repetition, persistence after refusal, entry onto noncustomary curtilage, coercive show of authority, or use of the visit as a pretext to inspect property can transform a permissible knock into a search or seizure.

First Amendment and association. Predictive policing can burden speech or association when police target people because they criticize the program, live with a listed relative, associate with alleged gang members, or decline to cooperate. Gang association can be relevant to a properly particularized investigation, but generalized social ties are protected unless connected to unlawful conduct under the applicable legal standard. Pasco’s settlement recited substantial interference with the plaintiffs’ right of intimate family association. That is not a holding that every social-network analysis violates the First Amendment; it is evidence that repeated list-driven family contacts crossed a constitutional line in the settled cases.

Retaliation requires proof that protected expression was a substantial or motivating factor in adverse government action, subject to doctrine governing probable cause and official motive. Complaints that officers intensified attention after residents objected would therefore require incident-level evidence. The public record supports serious investigation of retaliation but not a blanket finding that every Pasco check was retaliatory.

Equal protection. Racially disparate results alone ordinarily do not establish an Equal Protection Clause violation where the rules are facially neutral; Washington v. Davis requires discriminatory purpose. Statistical disparities remain relevant evidence, especially when combined with departures from policy, selective enforcement, historical background, or statements by decision-makers.

The strongest equal-protection cases would compare similarly situated candidates: for example, whether Black and white people with the same qualifying score were equally likely to be promoted to the active list, contacted, searched, arrested, retained, or offered services. None of the three agencies published the longitudinal person-level treatment data necessary for a complete analysis. The lack of data does not prove discrimination; it impedes the government’s ability to disprove it.

Federal grant recipients also had obligations under Title VI and other antidiscrimination laws. Administrative disparate-impact review can be broader than a constitutional damages action, though available remedies and private enforcement differ by statute and regulation. BJA therefore had reason to examine Pasco’s list composition and implementation even without proof of purposeful racial discrimination.

Substantive due process. Substantive due process is a difficult theory because courts generally analyze a specific search, seizure, or speech burden under the more particular constitutional amendment. Arbitrary designation alone rarely meets the “shocks the conscience” standard. A sustained campaign against a family unsupported by current suspicion may nevertheless implicate family and bodily liberty interests, particularly where no more specific doctrine fully captures the conduct. Pasco’s settlement references liberty interference but does not establish a general substantive-due-process rule for predictive lists.

Minors, schools, trauma, and disability. Person-based policing of children demands stricter safeguards than adult focused deterrence. Children have less control over residence, family associations, school attendance, transportation, and records created by adults. An abuse history is not culpability. Poor grades may indicate disability, language barriers, homelessness, caregiving duties, or inadequate instruction. Absences can reflect illness or family instability. Converting those conditions into police risk without individualized review and meaningful service provision conflicts with the rehabilitative purposes agencies invoked.

A future program using school information should presumptively exclude grades, disability records, child-welfare narratives, counseling information, and adverse childhood experiences from enforcement scoring. A narrowly tailored safety intervention should be controlled by education or public-health professionals, use the minimum necessary information, and prohibit transfer to criminal-intelligence databases absent an emergency or individualized legal basis.

Municipal liability. Under Monell, a municipality is not liable merely because it employs an officer who commits a constitutional tort. The injury must be caused by an official policy, longstanding custom, decision of an authorized final policymaker, or deliberate failure to train under demanding standards.

These programs make the policy element easier to investigate because they were formal departmental initiatives with manuals, databases, grant applications, command directives, and training materials. But plaintiffs still must prove an underlying constitutional violation and causation. Chicago’s and LAPD’s formal lists could support Monell allegations if an officer’s unlawful action followed an official instruction. Pasco’s sheriff, as the elected policymaker for many law-enforcement functions, presented an especially direct policy pathway, although the settlement ended the principal suit without a merits determination of liability.

Litigation and oversight chronology.

The five strongest official justifications and the five strongest criticisms.

The strongest official case is not that an algorithm can tell police who will commit a crime. It is that serious violence is concentrated and that carefully governed, voluntary, service-centered outreach to people at demonstrably high risk of victimization may be beneficial. The strongest criticism is not that all risk assessment is mathematically impossible. It is that these jurisdictions attached police attention to poorly calibrated, procedurally opaque classifications without proving that the classifications improved outcomes over ordinary, less intrusive practice.

Counterfactual: what likely would have happened without the lists? Police would not have ignored the same people. Detectives, gang units, probation officers, CompStat meetings, warrant squads, field officers, and community organizations already concentrated on people with recent shootings, serious arrests, supervision status, active warrants, or known conflicts. Much of the information in the models duplicated ordinary police knowledge.

Chicago’s RAND design provides the best evidence. Matched high-risk people not on the SSL experienced similar murder and shooting outcomes, while those on the list experienced more police contacts. The likely counterfactual is therefore that most selected people would still have had substantial police exposure, some would have been arrested, and the citywide homicide trend would have continued without SSL; the list principally changed the intensity and documentation of contact rather than violence.

For LAPD, hotspot patrol in Newton may have reduced crime regardless of chronic-offender bulletins. Officers would still have monitored parolees, served warrants, investigated shootings, and focused on known gun offenders. LASER’s distinctive contribution was to standardize and disseminate dossiers and maintain a quota of named subjects. Without the list, attention would likely have been less systematically attached to the same 12 people and fewer field contacts would have been interpreted through a preexisting chronic-offender frame.

For Pasco, the counterfactual difference is greater. Deputies would still have responded to crimes, warrants, probation violations, school incidents, and calls for service. But repeated visits to a family solely because a household member was on the prolific-offender list, and minor code-enforcement scrutiny organized around that designation, probably would not have occurred at the same frequency. The settlement’s prohibition on repeated, suspicionless checks confirms that the designation itself was an operational cause, not merely a label attached after independent encounters.

Model policy for any future person-based violence-prevention program.

A lawful and scientifically defensible policy should begin with a presumption against person-based enforcement prediction. An agency seeking an exception should satisfy all of the following requirements before deployment:

1. Single, explicit target. Define one outcome—such as nonfatal shooting victimization within six months—and do not combine victimization, offending, arrest, self-harm, and service need in one score.
2. Nonenforcement default. A model output may trigger voluntary, civilian-led outreach or resource prioritization. It may not independently justify a stop, search, surveillance operation, warrant request, home visit, prosecution enhancement, school discipline, or supervision action.
3. Evidence restrictions. Exclude arrests without disposition unless separately justified; exclude protected speech, family association, school grades, disability information, child-welfare histories, immigration status, and appearances as victims or witnesses from offender-risk scoring.
4. No contact feedback. Police contacts produced by the intervention may not be recycled into the score. Training data should distinguish officer-initiated records from externally reported events and should be audited for differential enforcement.
5. Published model card. Disclose the eligible population, endpoint, prediction horizon, variables, coefficients or executable code, missing-data rules, identity-resolution procedures, update interval, vendor role, and known limitations.
6. Threshold governance. Select thresholds using predeclared costs of false positives and false negatives. Publish precision, recall, selection rate, calibration, and group-specific error rates at the actual operational threshold.
7. Counterfactual evaluation. Use randomization where ethically permissible or a prospective matched design with similarly situated nonselected people. Measure verified victimization and offending separately from arrest and police contact.
8. Notice and contestability. Before any individualized consequence, provide notice, the principal evidence, duration, expected intervention, correction procedure, counsel or advocate access where appropriate, and review by an independent body.
9. Strict duration and deletion. Expire designations after a short period unless revalidated from new nonintervention-generated information. Delete operational flags and downstream copies; preserve only deidentified audit records under controlled access.
10. Minors prohibition. Do not place minors on police offender-prediction lists. Youth safety assessment should be controlled by education, health, and community professionals and firewalled from criminal intelligence.
11. Independent civil-rights audit. Require disparity analysis by race, ethnicity, sex, age, disability, geography, and supervision status, using both general-population and pipeline-conditioned baselines.
12. Vendor and partner accountability. Contracts must guarantee government access to source code, audit logs, model versions, data lineage, retention controls, and publication rights. A contractor that designs a system should not be its sole evaluator.
13. Contact logging. Every action associated with the designation must be logged with legal basis, officer, purpose, outcome, and service result. Unlogged use should be prohibited and disciplinable.
14. Community authorization and sunset. Require public hearings, legislative approval, a short sunset date, and reauthorization only after an independent evaluation demonstrates net benefit.
15. Individual remedy. Provide correction, deletion, injunctive relief, and compensation where an erroneous or unlawfully used designation causes harm.

Final judgment. Pasco County most closely approximated policing people for crimes not yet committed.

Chicago’s system was the most statistically elaborate but often predicted victimization or general violence involvement, and its field response was inconsistent. It increased contact and may have influenced investigations, yet the available evidence does not show a uniform campaign of suspicionless enforcement against everyone with a high score. LAPD’s LASER came closer: it selected named “chronic offenders,” circulated dossiers, instructed officers to locate and monitor them, and used contacts as scoring evidence. Its explicit “extraction” language and top-12 quota made it offender-focused even when the data did not support a uniform threshold.

Pasco crossed the furthest operational line. Its system assigned a future-oriented offender identity, included minors, used arrests and suspicions rather than convictions alone, considered associates and appearances in reports, and treated designation as a reason for recurring visits and enhanced scrutiny. Children’s grades, absences, trauma, and abuse histories were converted into sheriff intelligence without routine notice. For the plaintiffs who settled, the resulting checks exceeded the ordinary license to knock, substantially interfered with family association, and interfered with liberty interests.

That judgment does not mean Pasco literally punished thought or that every listed person was innocent of prior conduct. Nor does it convert the settlement into a judicial declaration that the entire ILP philosophy was facially unconstitutional. It means that, among the three cases, Pasco most clearly used a prediction of future misconduct as the independent operational reason to impose repeated government attention before any new crime supplied individualized suspicion. Chicago most clearly demonstrated the statistical futility of confusing a ranked risk signal with a prevention strategy. Los Angeles most clearly demonstrated how ordinary point rules become powerful through integrated data and officer workflow. Pasco demonstrated the endpoint when prediction, surveillance, family pressure, and minor enforcement are allowed to fuse: the government begins not merely investigating what a person has done, but organizing its relationship with that person around what officials believe the person might do next.