Crime Data API & Crime Map API Blog
Developer insights on crime data APIs, crime map APIs, embeddable crime maps, real-time crime intelligence, and public safety data integration.
The Number the FBI Isn't Allowed to Print
About twelve thousand agencies send the FBI their police use-of-force reports, and the bureau publishes percentages only. A four-row table OMB wrote in 2018 keeps the total count locked until coverage reaches 80 percent of the nation's officers. The latest figure is 77.6 percent of an estimate that keeps getting bigger.
More Articles
The Baltimore That Didn't Happen
Malcolm Gladwell says Mayor Brandon Scott deserves the credit for Baltimore's homicide drop. Homicide fell across the US, Canada and England at the same time, so we went looking for the catch. Baltimore really did beat the trend. The best evidence comes from the city's own contractors, and the office that gets the credit is fighting its inspector general in court over its records.
The Agency Field Says HEART
San Francisco's dispatch file says the agency column holds three values. It holds four, and the fourth, an unarmed street team, has taken 56,910 calls since 2023 without writing a single police report. What happens to a crime trend when the person who would have written it down stops showing up.
Ten Pages or Nineteen Hundred
A Washington sheriff stopped sending the local paper its ten-page weekly log. The records request for the same month came back at more than 1,900 pages. In Leesburg, Florida, the police just stopped sending theirs: the city retired the software, and there's no other way to make the list. Police departments have the data; what keeps breaking is the simple spreadsheet the public can read.
The Make Column Isn't There
In Milwaukee in early 2021, Kia and Hyundai thefts went from 92 to 1,920 while every other make fell. Two brands drove much of America's car-theft spike and its decline, and almost no public crime feed records what kind of car was taken. Dallas's file shows the fingerprint anyway, in the notes.
Stitches or No Stitches
In Los Angeles County, a fistfight becomes a violent crime if the cut needed stitches, and victims often refuse the stitches. In 2014 the LAPD was found coding aggravated assaults down while the Sheriff's Department, same county, same year, was coding them up. Here is how to spot the line moving in a public feed.
One House, Two Police Departments
Cornell's own police took the Chi Phi report in November 2024. Ithaca's police found out from the news twenty-two months later. When the institution with the most to lose employs the police, runs the hearing, seals the outcome, and counts its own crimes, the Clery Act gets you a number in a PDF.
The Store That Doubled San Francisco's Shoplifting
In September 2021, one Target in San Francisco filed 154 shoplifting reports, ten times its August count, and the whole city looked like it had doubled. The store had a new reporting system. Shoplifting is the one common crime whose victim has a procedures manual, and the count moves when the manual does.
The Rules Engine Runs Every Day
In its first ten months of automatic sealing, Pennsylvania sealed more than thirty-four million court cases, in a state of thirteen million people. Michigan's version runs every day. The public criminal record now gets edited on a schedule, and every copy of it goes stale on the same schedule.
One Point Three Miles of Sidewalk
In the summer of 2009, Philadelphia put pairs of new officers on foot around its most violent corners, and violent crime fell 23% against matched beats that got nobody. The trick seems to have been how short the beats were. When the officers left, the crime came back.
The Twenty Percent Rule
El Salvador says it had eighty-two homicides last year, under a definition that leaves out police killings, prison murders, and bodies in unmarked graves. Mexico has four counts for 2025, and its two official ones sit 19.7% apart, just inside the line a 2015 protocol drew in Bogotá.
The Week Phoenix Had No Crime
Phoenix's public crime file has no records at all from September 3 to 8, 2025. The police department switched records systems on the 9th, and the seam runs through seven of the file's eight columns. September still looks like a normal month, which is the problem.
The Orphan Table
Every NIBRS record carries a one-character field at position six, and one of its values deletes an incident outright. There is no edit operation for a crime incident — correcting one field means deleting the whole thing and sending it back. What that does to a percent change, and the six things we do about it.
The Busiest Minute in Los Angeles
The most common time of occurrence in LA's million-record crime file is 12:00 PM, with 35,198 incidents. The minute before it holds 234. Nothing happens at noon — noon is what a records system looks like when nobody knows what time it was and the form requires an answer.
Twenty-Three of Eighty-Three
Philadelphia auditors pulled 125 incident reports, found 83 that department policy required officers to record, and found 23 recordings. The compliance rate is not the interesting part. The interesting part is that computing it took two records that disagree, and almost nobody has both.
The Highest-Crime ZIP Code in Cincinnati Has Almost Nobody Living In It
Our crime-by-ZIP MCP server named 45267 the worst ZIP code in Cincinnati. It is a hospital, and essentially nobody lives there. Why zero-population ZIP codes break per-capita rates, why smoothing fixes noise but not bias, why inferring population from crime is circular as a denominator and excellent as a detector, and the five-step cascade we would ship instead.
Eight Hundred Thousand People Stopped Showing Up
Manhattan lost about 800,000 daytime people between 2019 and 2021 and roughly 50,000 residents. Only one of those numbers has ever appeared under a crime rate. A crime rate is a fraction, and this industry audits the top of it and inherits the bottom — ambient versus residential population, the four ways researchers have tried to count who is actually present, and what changes when the denominator does.
Four Hundred Thursdays
The SpotCrime blog has posted 1,251 times since June 2008, including about four hundred weekly link roundups since December 2016. Nobody sets out to build a longitudinal dataset by posting links on a Thursday. But run it forward eighteen years and you get the one record no crime dataset contains: which agencies were publishing, and when they stopped.
Three Burglaries Is Not a Trend
A block with three burglaries can show zero next year or seven, and both are ordinary. Gelman and Price proved in 1999 that raw-rate maps mostly map which places are small; Andy Wheeler did the Poisson arithmetic for crime counts in 2014. Why percent change is the wrong statistic at block level, why the worst block always improves without anyone touching it, and the seven fixes.
The Enforcement Confound: Why Part of Every Crime Feed Measures Policing, Not Crime
Between 2019 and 2024 the adult drug arrest rate fell roughly 50% while adult violent arrests fell 16%. Nobody thinks drug activity halved — that gap measures enforcement supply, not crime. A developer-focused guide to initiation channels: the three doors into a records system, an enforcement-elasticity ordering by offense (murder near zero, drug and weapons near total), the staffing numbers behind the 2019–2024 contraction and the 2025–26 hiring rebound (PERF: +17% hires in 2025), marijuana legalization as a clean structural break, the SRS-to-NIBRS drug-offense channel trap, and six practices for keeping patrol intensity out of a block-level safety score.
Campus Crime Data Is a Separate Country: The Clery Act, the Daily Crime Log, and Why It Doesn't Merge With City Feeds
Roughly 6,000 Title IV institutions publish federally mandated crime counts every October 1 — collected under different definitions, over different geography, by different reporters, on a different clock than any municipal feed. A developer-focused guide to Clery: the four geography categories and the nested-column trap, why Campus Security Authorities make Clery a partially disjoint set from police data, the 2.2 → 7.5 sex offense series that measures the reporting funnel rather than victimization, the 2020–21 occupancy anomaly, the $14M Liberty fine and the incentive structure behind the undercount, the hazing column arriving October 1 2026 with no history, and why the two-business-day daily crime log is the only artifact here that can support a real product.
One in Five Calls Now Ends Without an Officer: Drone-as-First-Responder and the Calls-for-Service Data Layer
More than 1,000 US agencies now hold FAA waivers to fly a drone to a 911 call — the FAA issued more in the ten months after April 2025 than in the previous seven years combined. Chula Vista has logged 25,000+ missions and 4,629 calls with no patrol response at all. A data-first look at what happens to the crime record when a call is resolved from 200 feet: the CAD-vs-RMS split, the three mechanisms (disposition substitution, verification, reallocation) and why they don’t all point the same way, a bounding exercise on the size of the artifact, why this explains none of the national decline, and what developers should do about structural breaks in a block-level time series.
The SpotCrime MCP Server: Crime Intelligence as a Native Agent Tool
Why crime data belongs in the agent loop as a native MCP tool, not a REST endpoint to parse. Updated: the crime-data MCP is now live via CrimeScore (mcp.crimescore.info) — one tool, published_score(region_code), returning a national crime percentile and rank for any U.S. ZIP or county, powered by SpotCrime data.
What Is NIBRS? The Crime Data Standard That Changed Everything
NIBRS replaced the old summary tally system in January 2021. All 50 states are now certified and 82% of the US population is covered by NIBRS-reporting agencies. Here's what the SRS → NIBRS transition actually means — the hierarchy rule, the data gap, the 18% still missing, and what it means for anyone building on US crime data.
Cleared Is Not Solved: Clearance Rates and the Disposition Gap in Crime Data
Roughly 61% of US murders were cleared in 2024. Motor vehicle theft clears at about 12%. But “cleared” is a police records term, not a justice outcome — it merges arrest clearances with exceptional clearances, divides by a mismatched denominator, and is self-reported by the agency it evaluates. A developer-focused guide to the disposition gap: what clearance measures, why it isn’t a solve rate, why incident feeds go silent after the report is filed, and what a pipeline should do about it.
Criminogenic Entropy: A Structural Theory of the Crime Decline
Almost every explanation of the crime decline names a cause. A new pre-print by Scott Mourtgos and Ian Adams does something different — it argues crime tracks “criminogenic entropy,” a slow, generational quantity with memory that peaked in the 1980s–90s and has fallen since. A data-first, skeptical walkthrough: what the framework explains that single-cause theories can’t (breadth, inertia, the 2020–21 spike as an exogenous shock), why the data through May 2026 are at least consistent, where the theory is vulnerable (factor selection, falsifiability, correlation), and what a broad structural downtrend means for anyone building on crime data.
Crime Has a Season: Why Month-Over-Month Comparisons Mislead
A dashboard that reports “assaults up 34% since April” on a July morning has told you about the calendar, not about crime. Violent crime runs higher in summer everywhere, every year — while robbery leans toward winter, larceny toward the holidays, and each offense keeps its own clock. A developer-focused guide to crime seasonality: the summer violence peak, the offense-by-offense differences, the daily and hourly cycles, seasonal adjustment (year-over-year, STL, X-13ARIMA-SEATS), the midnight-placeholder artifact, and why fitting a seasonal curve to a single block is a coin flip.
The Dark Figure of Crime: Why Reported Incidents Aren't Crime
Only about 40% of violent victimizations and 30% of property victimizations are ever reported to police — and reporting rates vary by a factor of three across offense types. A developer-focused guide to the dark figure: the NCVS vs. UCR split, why reported-incident counts aren't comparable across categories, why reporting drift can masquerade as a trend, and why murder (down 18.7% in the Real-Time Crime Index) is the anchor every serious crime analysis leans on.
The LAPD Restored Block-Level Data and Went Biweekly. The Volume Drop NIBRS Doesn't Explain Is Still There.
The block-level locations are back and LAPD now publishes biweekly — but plot the daily incident count and a separate defect appears. Reported incidents fell ~49% at the March 2024 handoff (637 → 322/day) and remain ~26% below the 2023 baseline even after NIBRS-only reporting stabilized. NIBRS carried 0.77 offenses/day before March 2024, so this was one feed switched off before the other switched on — a handoff gap, not a format change. Location precision was fixed; incident volume is still open and unexplained.
The Geography of the Crime Decline: What State-Level Data Reveals That the National Number Hides
US crime hit a record low in 2024 — but the safest and least-safe states differ by more than 7 to 1. A data-first walk down from the national number: the state spread (Alaska 724 vs. Maine 100 per 100k), why violent and property maps don't overlap, the Hawaii surge, the NIBRS reporting asterisk on every state comparison, and why states are still the wrong unit for almost every decision anyone makes with this data.
How Crime Lands on the Map: The Geocoding Problem Every Crime Data API Has to Solve
Turning a police record into a point on a map is the least-discussed and most error-prone step in any crime data pipeline. A developer-focused field guide to the five geocoding methods, the hundred-block convention, the match rate that actually matters, and the spatial error budget — why a coordinate is only as meaningful as the method that produced it, and what to demand of any API that claims to put crime at an address.
ShotSpotter Fired. Police Responded. No Crime Recorded. What Happens to That Data?
When acoustic gunshot detection fires and police find nothing, the event vanishes from the crime record. That gap — between detection events and recorded incidents — has specific, measurable consequences for anyone consuming crime data professionally. A developer-focused look at the MacArthur Chicago audit, the three data artifacts ShotSpotter deployments create, and what to do about it.
After 17 Months of Shifting Excuses, LAPD Published Block-Level Crime Data the Week After the Primary
LAPD gave four documented justifications for why it couldn't produce block-level crime data. On June 9, 2026 — one week after the LA primary — it published exactly that. The blackout quantified: zero records published in all of 2025 in real time, 150,431 cases backfilled around the primary. LA's published decline of 35–42% vs. Chicago's –10% and NYC's –1.7%. The shooting-victim undercount. And the $50.9M Motorola maintenance contract on the same Police Commission agenda.
Why the Same Crime Has Three Different Names: Incident Taxonomy and the Normalization Problem
A man forces a window and takes a laptop. Local CAD codes it one way, the FBI's Summary Reporting System another, and NIBRS a third — and that is before anyone tries to put it on a map. A developer-focused walkthrough of crime classification: the SRS hierarchy rule that undercounts by design, the 2021 NIBRS discontinuity, and why normalization — not collection — is the hard part of any crime data API.
If Cancer Had Dropped Like This, We'd Know Why. Why Doesn't Anyone Study the Crime Decline?
Cancer mortality is down 34% since 1991 and we spend roughly $7.2 billion a year studying why. Violent crime is down ~60% and we spend roughly $30 million. The two declines saved a comparable number of lives. The research budgets differ by a factor of 240. Four reasons for the gap, and what a serious crime-decline research program would actually look like.
Violent Crime Is Down ~60% Since 1991. That's Roughly 10,000 Americans Not Killed Each Year — And We Still Don't Know Why.
The violent crime rate in the United States is roughly 60% below its 1991 peak. Translated into bodies, that's about 10,000 to 11,000 fewer Americans killed every year — the population of a small town, saved annually, for a generation. Almost nobody is celebrating, and almost nobody can fully explain why. The long-arc framing, the lives-saved math, and what it means for anyone using crime data operationally.
The Business Case for Crime Data in Commercial Real Estate
Commercial real estate is a geography business — every site-selection model, underwriting memo, and premises-liability defense resolves to a point on a map. Yet most run on ZIP-code or county crime averages, a unit of aggregation that can be off by an order of magnitude across a single trade area. A data-first look at where address-level crime data sharpens a CRE decision, where it double-counts risk already priced into rents, and the freshness and calibration traps in between.
AI in Law Enforcement: What Crime Data APIs Can Actually Support
Case triage, deployment, report drafting, and alerting are all moving to AI. But a descriptive crime data API records what was reported, not what occurred or what will. A grounded look at where AI genuinely helps, where it gets dangerous, and the calibration problem underneath it all — deterministic code owns the numbers, the model owns the language, a human owns the decision.
How SpotScore™ Is Calculated: A Methodological Walkthrough of Block-Level Safety Ratings
A safety score is a compressed claim about a place: incident inputs, severity weighting, temporal decay, geographic aggregation, normalization, and the calibration problem every safety score confronts. Updated: neighborhood scoring shipped via CrimeScore (crimescore.info) as a ZIP-level national crime percentile powered by SpotCrime data — this essay's block-level design is kept as historical context.
Arrest Is Not Conviction: The Documented Harms of Publishing Names and Home Addresses
Richard Jewell. Brian Banks. The Mugshots.com convictions. The Maine, Washington, and South Carolina vigilante killings of people located through public sex offender registries. A look at what the documented case record actually shows about publishing arrest data, mugshots, and name-and-address registry entries — and the line between publishing incidents and publishing people.
Crime Is Falling. So Why Did the LAPD Go Dark? The NIBRS Excuse Doesn't Survive the Math.
US crime is at multi-decade lows. The LAPD stopped publishing address-level incident data anyway, citing the federal NIBRS transition. The math runs the opposite direction — NIBRS produces more publishable detail per incident, not less. The public-panic justification has no documented precedent. And the accountability chain runs through Mayor Karen Bass.
Flock Safety Watches Everyone. Your Police Blotter Is Missing.
Flock Safety cameras log every vehicle in 5,000+ communities. Police can query 83,000 cameras nationwide with one click — no warrant required. Yet the same departments won't publish a daily incident spreadsheet, keep their own usage policies secret, and Flock's CEO calls citizens who map public cameras 'terroristic.' A case study in one-way surveillance.
Why SpotCrime's Crime Data API Is Different: Direct From Every Department, Independent of Every Platform
Most crime data APIs are aggregators of aggregators. SpotCrime contacted every police department it covers and operates under a direct arrangement with each — independent of every real estate, insurance, or location platform. A walkthrough of why the pipeline matters as much as the data.
AI Agents in Executive Protection: What Happens When the Intelligence Analyst Is a Language Model
Executive protection programs have always run on intelligence. Over the past eighteen months, a meaningful share of that workload has shifted to language-model agents querying crime data APIs. A look at what works, what does not, and the calibration problem nobody has solved.
What Crime Data APIs Suppress, and Why: A Practical Guide to Privacy, CJIS, and the Limits of Public Records
Crime data is, in most US jurisdictions, public record. That does not mean every field in a police report is publishable. A practical guide to the three legal layers (CJIS, state public records law, subject-matter privacy statutes), how the major US municipal feeds actually suppress data, and the secondary suppression developers should apply at the display layer.
Crime Hotspot Mapping for Developers: Kernel Density, Hex Grids, and What Makes a Map Meaningful
A pin map is not a hotspot map. Kernel density estimation, hex grid aggregation, and bin selection all produce dramatically different pictures from the same incident data — and most published crime maps are wrong in ways developers can avoid.
How to Evaluate a Crime Data API: Coverage, Freshness, and What Actually Matters
Most crime data APIs look the same from the documentation. Coverage claims are vague, freshness claims are optimistic, and taxonomy differences are buried in footnotes. A seven-dimension framework for evaluating what you're actually buying — before you build production infrastructure on it.
Real-Time Crime Index vs. FBI UCR: Two National Crime Datasets, Two Different Jobs
The RTCI updates every 45 days. The FBI UCR publishes annually — and the 2021 NIBRS transition knocked out coverage for 37% of the US population. Both are described as national crime datasets. They are not the same thing, they do not measure the same events, and they fail in different ways. A practical breakdown for developers and analysts.
Speed vs. Accuracy: What the CrimeRadar False Alert Reveals About AI in Public Safety
On April 13, 2026, CrimeRadar sent an active-shooter alert to parents in Mount Vernon, Missouri based on a misheard police radio call. No shooting occurred. The incident is a case study in what happens when AI alert systems optimize for speed over accuracy — and why the cost of false positives is not uniform across incident types.
HUD Just Cleared Real Estate Agents to Share Crime Data. Here's What That Actually Means.
HUD just cleared real estate agents to share neighborhood crime and school data. But look at the timeline of who removed it and why: Redfin was under active federal fair housing lawsuit. Zillow had just lost $881M on Zillow Offers. NAR was issuing apologies for its segregationist past. The story behind the data ban is more complicated than a principled stand — and now it's over.
Did COVID Accidentally Reduce Violence? Six Theories the Experts Won't Touch
US homicides are on track for a 125-year low. The decline accelerated exactly when COVID hit. We examine six unorthodox theories — from viral natural selection to vaccine behavioral effects and digital sedation — with the data and plausibility assessment behind each.
How Public Crime Reporting Could Have Exposed a Decade of Fabricated Robberies
Five Louisiana law enforcement officials were federally charged with filing fake armed robbery reports for nearly a decade to enable immigration fraud. The scheme went undetected until federal immigration officers noticed irregular paperwork. It likely did not need to take that long.
How Family Safety Apps Can Add Crime Layer Intelligence
Life360 has 66 million users. Apple’s Find My connects hundreds of millions. Every one of these platforms knows where your family is. Almost none can tell you what’s happening there. Here’s how crime layer intelligence — real-time incidents, neighborhood safety scores, and shooting feeds — closes that gap.
Predictive Policing vs. Descriptive Crime Data: The Distinction Every Developer Should Understand
Predictive policing algorithms carry legal, ethical, and reputational risk that most developers don’t sign up for. Descriptive crime data — what actually happened, where, and when — doesn’t. Here’s why the foundation your product chooses determines its accuracy, defensibility, and long-term viability.
Which US Cities Are Seeing Crime Rise vs. Fall in 2026
The national narrative says crime is falling — and in aggregate, that’s true. But 6,501 shooting incidents across 12 cities in 60 days tells a far more fractured story. We break down the city-level data, the Seattle anomaly, the Chicago paradox, and what it all means for platforms built on crime data.
Why Has Violence Fallen Off a Cliff? The Data, the Mystery, and What Could Reverse It
US homicides fell 15% in 2024 and an estimated 20% in 2025 — possibly the lowest murder rate since 1900. Almost nobody is explaining why. We examine six competing theories, the policing paradox, and what a looming economic shock could mean for the future.
Crime Data and Property Insurance: The Risk Signal Insurers Are Finally Taking Seriously
Property insurers have priced crime risk on blunt ZIP-code averages for decades. Address-level crime data is changing that — moving from research project to underwriting infrastructure. Here’s how it works, what it means for homeowners, and where the developer opportunity lies.
How Real Estate Platforms Are Using Crime Data to Win Buyer Trust
Neighborhood safety is now a top-3 home buying criterion — and buyers expect platforms to show it. Here’s how PropTech teams are integrating address-level crime data APIs, what buyers actually want, and why build-vs-buy decisions favor third-party data infrastructure.
Gun Violence Data and the Rise of Hyperlocal Shooting Trackers
America’s gun violence data infrastructure is broken. FBI crime reports lag 18 months. City-level aggregates obscure the block-by-block reality. Here’s how hyperlocal shooting trackers are filling the gap — and what 6,463 incidents across 12 cities in 60 days actually reveals.
LAPD vs. The Public: SpotCrime Sues for Crime Data Transparency
LAPD stopped releasing detailed crime location data in early 2025. SpotCrime, LAist, and RAND all filed public records requests and received nothing. Now SpotCrime has filed a lawsuit alleging an unlawful pattern of delaying tactics — and the public’s right to know is at stake.
America's Historic Crime Drop — And Why Local Data Has Never Mattered More
Murder in America has fallen at the fastest rate ever recorded over the last three years — reaching roughly 14,000 in 2025, a level not seen in decades. Here's what the data actually shows, why public perception lags so far behind, and why granular local crime data has never been more critical.
How Crime Data APIs Work — And Why AI Agents Will Be Their Biggest Users
A deep dive into how crime data APIs ingest, normalize, and serve real-time incident data — and why autonomous AI agents are poised to become their most valuable consumers.
Ready to build with crime data?
Access real-time crime intelligence via REST API. Designed for developers and AI agents.
Request API Access