Sources, definitions, and limits

A grade is a relative position within a peer group, never an absolute verdict. Every number below renders directly from the same configuration the engine runs on, so this description cannot drift from the code.

• Peer groups. Agencies are compared only within their own type. Municipal departments are further banded by mean reported officer count: 1-14 officers, 15-29 officers, 30-59 officers, 60+ officers. Every other type (corrections, sheriffs, prosecutors, state) forms a single statewide band. A group needs at least 8agencies to grade; smaller groups render an honest "not graded" state.
• Components. IA volume (investigations per 100 reported officers, multi-year mean; weight 40%), discipline severity (sanction-weighted major discipline per 100 officers, 2020-2025; weight 40%), and allegation severity (mean tier of IA allegations; weight 20%). Each component becomes a percentile within the peer group, where higher means more reported volume or severity.
• Grade. The weighted mean of component percentiles is re-ranked among peers; the grade is the fifth of the peer group it lands in. A: lowest fifth. F: highest fifth. At least two computable components are required.
• Confidence.Shown separately from the grade, never blended into it: years on the AG's non-reporting list, few usable reporting years, and small officer counts (noisy rates) each lower confidence, and the reasons are printed on the card.

Sanction weights: termination 4, demotion 3, suspension 2, separation while IA pending 2, other sanctions 1. Allegation tiers are an editorial choice, published in full below; values the AG reports as unknown or not provided are excluded from severity math rather than guessed.

Some officer pages show a salary or pension figure from New Jersey Treasury's YourMoney transparency data. This is a different source from the Attorney General discipline records, and it is the most identity-sensitive feature on the site, so the rules below are deliberately strict. When a match is not certain, nothing is shown.

• Which pension systems.Only the Police and Firemen's Retirement System (PFRS) and the State Police Retirement System (SPRS) are used. PFRS membership alone does not mean a person is a police officer: it also covers firefighters and many corrections titles. So the fund is a filter, not proof. Authoritative police relevance comes from the employer, resolved to an agency this site already covers. For active members the data also carries a "police officer" group that excludes firefighters up front; the retired file has no such flag, so retired municipal PFRS records are never shown (police and firefighter cannot be told apart there). SPRS members are sworn State Police.
• How a figure is linked to an officer.A pension record is attached to a named officer only at high confidence: the normalized full name matches and the pension employer resolves to the officer's exact agency (or, for SPRS, to the State Police). A timeline check rejects namesakes: a retiree whose retirement predates the officer's earliest discipline record cannot be that officer, so the match is dropped. If more than one pension record could be that officer, or the same record could be more than one officer, the match is withheld rather than guessed; some of those candidates are later resolved by the evidence-cited verification process described above, and the rest stay unpublished. Lower-confidence matches (same county but not the exact agency, or name-only) are never published.
• What the amount is. Figures are as reported by NJ Treasury as of 2026-03-31. A salary is the pension-basis figure and may not include overtime, accrued leave, stipends, or other pay. A retiree's figure is the monthly allowance. Blank fields read "not reported."
• Corrections. Because the officer-to-record link is an inference, a disputed match is removed on request through the corrections process, which is honored as a permanent suppression.

The police and teacher pay pages compare two workforces using the same NJ Treasury snapshot described above. Nothing on those pages is attached to a person, and the committed school-staff subset carries no member names at all: the columns filtered out of the raw file are the name columns, so the identity never enters this repository. Every constant quoted below is read directly from the configuration the computation runs on, so this description cannot drift from the code.

• The two groups.Police are active members of the Police and Firemen's Retirement System in the "police officer" group, which excludes firefighters at source, further restricted to employers that resolve to municipal or county law enforcement. Employers of type state or other, and statewide bodies with no county of their own, are excluded, because a state prison or NJ Transit is not what a community pays for its own policing. School staff are every active member of the Teachers Pension and Annuity Fund.
• TPAF is not only classroom teachers. Every active member of the Teachers Pension and Annuity Fund. The fund covers certificated school staff and the file carries no job title, so principals, supervisors and other certificated staff are counted alongside classroom teachers. School support staff (aides, secretaries, custodians, drivers) are enrolled in PERS instead and are not counted here, so this is not a school payroll figure. That is a real limit on the comparison and it is stated on every page of the vertical, not only here.
• How county is assigned.From the source's own location field, which the active file populates for every fund. No name matching, no district crosswalk, and no inference. Rows the source places in its "regional" or "statewide" buckets are excluded from county figures rather than assigned somewhere. The published file states exactly how many rows entered each figure and how many were excluded, by reason.
• What the amount is. Pensionable base salary as reported to the pension system. It excludes overtime, stipends, longevity, off-duty detail pay and accrued-leave payouts for both groups. Actual compensation is higher for both, and this data cannot say by how much for either, so it cannot say which direction the omission moves the ratio. A school contract year and a police work year are not the same length. The source carries no work-year field, so these are annual salaries compared as reported, not pay per day or per hour.
• The tenure adjustment. Police can retire after 25 years, so the active police workforce is less senior than the active school workforce: median police service years are lower in all 21 counties. Both groups are therefore re-weighted onto one reference distribution of service years (statewide-tpaf-service-years), using the bands under 5 years, 5 to 10 years, 10 to 15 years, 15 to 20 years, 20 to 25 years, 25 years or more. A band contributes only when both groups have at least 20 people in it, and the adjusted figure is withheld entirely unless the contributing bands cover at least 60% of the reference. Because police are the less senior group, this adjustment usually widens the gap rather than closing it.
• Suppression. A county or region with fewer than 30 people on one side publishes no figure for that side and no ratio. A single department needs at least 10 officers, a lower floor because a department median is a plain description of that department rather than a comparison carrying a rank.
• Regions. North, Central and South follow the state planning-region convention. The line between them is a convention, not a finding in the data, and where Ocean County belongs is genuinely disputed. County figures are shown underneath every regional figure so the grouping never hides anything.
• What a ratio is not. Municipal police payrolls and school payrolls draw on different tax levies and different state aid formulas. The comparison describes what two workforces are paid in the same place. It does not trace a dollar from one budget to another and cannot be read as doing so.

Some department pages show a staffing section sourced from the FBI's Law Enforcement Employees (LEE) census, a separate federal dataset from the New Jersey Attorney General records this site otherwise runs on. It is the one genuinely independent staffing denominator this site has, and the only source behind any per-capita figure here.

• What it counts.Full-time sworn officers and full-time civilian employees, each split by sex, as reported by the agency as of October 31 each year (per the FBI's Police Employee technical specification). Part-time staff, special officers, and crossing guards are excluded by the FBI's own definition; the sworn head of the department is included.
• Reporting is voluntary and uneven. Not every NJ agency reports every year. A missing year renders as a gap in the trend line, never as zero. A reported zero (rare, but real on some administrative rows) and a non-report are different things, and this site never conflates them.
• Only confident matches are shown.Each FBI record is keyed by its own agency identifier (ORI). ORIs join to this site's canonical departments through one committed, hand-reviewed agency-to-ORI crosswalk (see the matching section above): each entry links one ORI to one department, scoped by year where a department's history spans a rename or a municipal consolidation, and identifiers that deliberately do not join carry a written reason. Anywhere the match is ambiguous (two real municipalities sharing a name, a county-level total that could mean more than one thing), the figure is withheld rather than guessed. A department with no confident match shows an explicit "staffing data not available," never a blank that could be mistaken for zero.
• County corrections and sheriff's offices are not covered. The FBI census has no per-facility or per-office breakdown for New Jersey county corrections or sheriffs: only a single statewide Department of Corrections figure exists in it. Every county corrections and sheriff's office profile on this site will show staffing as not available; that is expected, not an error or an omission.
• Per-capita is municipal-only. Officers-per-1,000-residents is shown only for municipal police departments with a real reported population figure, and is computed by this site directly from the raw counts rather than trusting a precomputed source field.
• This is not the same "officers" as the AG's IA data.The Attorney General's IA Agency Totals file has its own "Number of Officers" field, but per the AG's own published notes, it is a unique count of officers named in that year's internal affairs investigations, not a staffing figure. The two numbers answer different questions and are never blended into one another on this site. Where a department page shows both an IA-based rate and an FBI-staffing-based rate, they are labeled separately and are not meant to be added, averaged, or reconciled against each other.
• Rates are approximate context.Any figure built from FBI staffing combines two different reporting regimes with different time bases (an October snapshot vs. a full-year AG count) and is shown only when the staffing year is recent enough to be a reasonable stand-in for the rate's own year. A higher rate is not automatically worse: it can reflect more reported misconduct, or simply a larger, more active department.
• What this data does not resolve. This site ran a deliberate, adversarial review of where the FBI and AG datasets disagree (agency coverage, reporting years, and the definitional mismatch above) before publishing any of it. Every material limitation that review found is reflected in the caveats above.

The fatal encounters section is built from the Fatal Encounters dataset, an independent, crowdsourced journalism project (fatalencounters.org). It is not a government source. It is the only third-party dataset on this site, and every page built from it labels that provenance distinctly from the Attorney General and federal data.

• What a record is, and is not. Fatal Encounters documents deaths that occurred during police encounters of any kind. The New Jersey subset covers 2000 through 2021: 549 records. About 24 percent involve vehicle pursuits; 66 records are ones the project itself codes as suicides; others are medical emergencies, drownings, and overdoses that happened while police were present or responding. This is not a list of police killings, and a record here is never a finding of fault or misconduct by any officer or agency.
• The dataset is frozen. The project suspended collection in December 2021. The absence of records after 2021 means the dataset ended, not that deaths ended. This site does not extend or update it.
• Fetch and cache.The data is fetched programmatically from the project's public Google Sheet (pnpm fe:fetch). The New Jersey subset is committed to this repository with per-row provenance (each record carries its source row), and its checksum is pinned in the source registry. Builds are deterministic: every page renders from the committed, verified snapshot, never from a live fetch.
• How records are attributed to departments.The dataset names agencies as free text. Those strings are resolved to this site's canonical agencies through a crosswalk with confidence tiers, and only high-confidence matches attribute an encounter to a department page. Ambiguous matches and out-of-county municipal matches are never auto-trusted: the source strings include out-of-state name twins, so a same-name municipality in the wrong county attributes only after a researched confirmation citing incident-specific sources (see the matching section above). Records with no confident match still get their own pages, unattributed. 525 of the 549 New Jersey records are attributed.
• Fields the site displays but never computes from.The dataset's own column titles say it plainly: "Dispositions/Exclusions INTERNAL USE, NOT FOR ANALYSIS" and "Intended use of force (Developing)." Record pages display these values inside a clearly caveated block, because hiding them would be its own distortion, but the site never computes anything from them: no metric, no filter, no ranking, no index. A build gate enforces this.
• Race is shown as reported only.The project publishes statistical race imputations with probabilities for records where race was not reported. Those imputations are excluded from every page on this site and carried only in the machine-readable mirror (/data/fe.json). The reasoning: imputation is a defensible technique in aggregate analysis, but it cannot support asserting an individual person's race on a page about their death.
• Also excluded from pages:source image URLs and the project's unfinished "Temporary" columns.
• Known quirks in the source data. The project states that vehicle-pursuit deaths are undercounted. Race reporting is spotty before 2013. Two records carry county values that are not New Jersey counties; they render as recorded, because correcting a source value would break the as-reported rule.
• Corrections and removal. The corrections process covers these pages, and it is extended for family members of the deceased, who can request correction or removal of a record page.

The county maps draw outlines published by the New Jersey Office of GIS. They are a cartographic source and carry no claim about any agency: they supply the shapes, and the 2020 Census population that the per-resident rates divide by. Every measure drawn on them comes from the same Attorney General records as the tables. The maps live on the county maps page, one static page per measure.

• The outlines are not legal boundaries.The source says so itself: the layer "is not a survey document and must not be used as such," and its polygons "do not represent legal boundaries." They are also generalized for display. The full-detail source carries about 201,000 points; the maps here draw about 3,000, which at the size they render is finer than a screen can resolve but is not a survey line.
• Rates come first, counts second. A map shaded by raw investigation totals is a picture of where people live. Essex reports 2,953 investigations and Salem 66, and that gap mostly tracks population. Raw counts are still available, on their own pages, labeled as counts.
• Per-resident rates mix vintages. The population is from the 2020 Census. The internal affairs figures are more recent. A rate that divides one by the other is still the most comparable measure available across 21 counties of very different size, but the two numbers were not counted in the same year.
• Nothing here divides by department staffing, because that number does not exist in this data.The Attorney General's officer column counts distinct officers named in internal affairs investigations, not officers employed. It totals 8,532 statewide, against a real New Jersey sworn strength several times that. So no measure on these maps is called a rate per officer without saying which officers it means.
• Shading is relative, and the boundaries are printed. Counties are sorted by rank into equal-sized groups rather than shaded on a straight scale from lowest to highest, because one outlier otherwise flattens the other twenty into two indistinguishable shades. The value at each group boundary is printed above the legend so any shade can be read back to a number. It is a description of how counties compare to each other, not a grade.
• State agencies are not on the map. The State Police, the Department of Corrections, Transit Police, and other state-level agencies operate across county lines, and the Attorney General reports them outside the county system. They appear in no county on any of these maps, and every map says so. They are in the rankings.
• The maps work without JavaScript. Every map is drawn on the server and shipped as part of the page, with a hidden table listing every county and its value for screen readers and for anything that cannot render an image. On the county map pages, a pan-and-zoom layer loads over the top once you scroll to it; if it fails or you have scripts disabled, the drawn map is what remains, and it is complete. No map on this site requests anything from a third party.

Military-equipment sections are built from the Defense Logistics Agency's "LESO Public Information" quarterly file, the federal accounting of property transferred to law enforcement under the 1033 program. This site reads the NEW JERSEY sheet: 2026-06-30 snapshot, 1,946 line items. This is federal data, a different provenance from both the Attorney General records and the third-party Fatal Encounters dataset, and it is labeled as such wherever it appears. Receiving equipment through the program is not misconduct. Statewide totals, per-county rollups, and top categories are on the 1033 program overview.

• Transfers, not inventory.The file records line-item transfers over time, not what an agency holds today. This site says an agency "received" or "acquired" an item, never that it "has" it. DEMIL code A means non-controlled property the agency may dispose of after one year; codes C, D, F, and Q mean controlled property. The file is a quarterly accounting snapshot, not a complete historical log: by the program's own publication history, records before FY2017 are incomplete.
• What the dollar figures mean.Acquisition value is the original cost to the Department of Defense per unit of issue, not current or market value. The site computes each line's value as quantity times that per-unit cost; that multiplication is this site's assumption, stated here.
• Quantities aggregate only within a unit of issue. The file counts in units like Each, Kit, and Box. Adding a kit to an each produces a meaningless number, so this site never publishes a cross-unit item count. Dollar values are summable, and totals are stated in dollars.
• Fetch and cache.The quarterly workbook is fetched programmatically (pnpm leso:fetch), the New Jersey sheet is committed as a subset with its checksum pinned in the source registry, and builds render deterministically from that verified snapshot. The refresh runbook is quarterly, matching the file's publication schedule.
• How recipients are matched to departments.Recipient names in the file are ALL-CAPS free text with no county column. They are resolved deterministically, including parsing embedded county tags ("WASHINGTON TWP PD GLOUCESTER CO"). Only high-confidence matches power department pages: 179 of 185 recipient names. The remaining names are shared by two or more real municipalities and no public evidence singles out the recipient, so they are withheld by design (see the matching section above); their transfers still count in statewide totals rather than being dropped silently.

Tactical or mundane: the equipment classification

The first four digits of an item's National Stock Number are its Federal Supply Class (FSC). Every FSC present in the New Jersey data is explicitly curated in this site's configuration as tactical or mundane, and every tactical assignment carries a written rationale. The table below renders directly from that configuration, so this documentation cannot drift from the computation. Two judgment calls to note: military trucks and Humvees (FSC 2320) are classed tactical but displayed as their own category, distinct from weapons and armored vehicles; ambulances and utility vehicles (2310), gas masks (4240), radios, computers, and clothing are mundane. A two-digit fallback exists only as a tripwire: if a future quarterly refresh introduces an FSC that has not been curated, the build fails rather than bucketing it silently.