Jake Lawrence · STS / Public Data / Data Journalism · Field Data theme
The United States does not reliably count the people its police kill. The authoritative national numbers are built by a newspaper, activists, and a volunteer, not by the state, and they disagree because they measure different things. A sourced reconciliation of five independent counters, with the federal count that could not publish a total.
The public-data arm of the research program, and the field-data companion to the Counting the Killings essay series. A content-as-code reconciliation of the five independent efforts that count people killed by US police: the Washington Post's Fatal Force (fatal shootings only), Campaign Zero's Mapping Police Violence (all killings), the crowd-sourced Fatal Encounters (all deaths in police contact), the FBI's voluntary National Use-of-Force Data Collection (the widest definition, and the emptiest, sealed until it cleared its 80 percent officer-coverage threshold in 2024), and the Guardian's The Counted. It publishes their definitional scope, per-dataset annual totals aggregated directly from each project's primary dataset, the split federal coverage series, and a record-level overlap as a versioned, hashed CSV and JSON release whose counts re-derive live in the browser and in CI. The signature reading is the decade-long definition gap: Mapping Police Violence builds on the Post and adds the non-shooting killings its rule excludes, about 110 a year. The record-level layer finds the civilian trackers converged (over 98 percent agreement), shows why a two-source capture-recapture is invalid on these nested lists, and cites the GBD study's finding that US vital statistics missed 55.5 percent of police-violence deaths. Derived aggregates only, never a re-hosted roster of the people who died. A systems-and-design reading, refreshed quarterly by pull request.
Two field-data studies of what a state cannot see about a life-and-death function it runs. The Legibility Gap measures what Ukraine's wartime registry cannot read about its own damaged homes; Counting the Killings measures what the American state cannot count about the people its police kill, a count so incomplete that a newspaper had to build the authoritative one.
Classification as Infrastructure names the residual category in the abstract. Counting the Killings watches five projects each draw the category boundary in a different place, so that a fatal shooting, a killing, and a death in police contact become three different populations, and the space between their totals is exactly the classification work made visible.
What the State Keeps reads the federal state as a classified inventory evaluated from above. Counting the Killings reads one thing the state does not keep at all: a publishable count of the people its police kill, left to a newspaper, a nonprofit, and a volunteer to assemble instead.
Seeing Like an AI Company runs Scott's legibility on a corporate proposal, asking what a frame renders visible. Counting the Killings runs the same lens on a public-safety count, and finds the state's own instrument, the FBI collection, defined the widest and saw the least, unable to publish until it cleared its own coverage threshold in 2024.
Two public-data studies in the same content-as-code, hashed-open-data mold, one civic and one national. Pingree Grove Studies publishes what one village's records can make legible; Counting the Killings publishes what five competing efforts can and cannot make legible about police killings nationwide, aggregates only, every figure re-derivable.
Back to the release table · See it on the network · Machine-readable catalog
I design and ship AI tools, full-stack apps, and data pipelines — end to end, to production. Tell me the problem in a sentence; I'll give you an honest read on fit within a day.
Work with me →