Hot Spots

What the map tells you before the street does. A short documentary and this field note, on the quiet arithmetic of where crime clusters — and the discipline of reading it without condemning anyone.

The kernel

Every reported offense becomes a point on the map. Each point is turned into a soft hill of probability — a kernel density estimate — and where those hills overlap, the pattern emerges. This is crime mapping: cartographic mapping in service of criminology, producing the crime map itself.

Repeat victimization

The most reliable predictor in the field isn't personality or motive — it's recency and place. A location that has been struck once is statistically far more likely to be struck again, and soon. So the highest-value question an analyst can ask on a Tuesday morning isn't "who," it's "where again."

The discipline

Predictive policing is an application of artificial intelligence to direct deployment. That's the whole definition, and the whole warning lives inside it: deployment, not verdict. The lattice forecasts patterns. It never condemns people.

A kernel assigns probability to a square of the grid, not to the soul living in it. The officer on the corner still has to think — that's the job the map can't hand them.

The arithmetic, in three numbers

Kernel bandwidth

Pick it too small and the map is noise; too large and it's a heat haze over the whole precinct. Calibration is a judgment, not a default.

The repeat-victim ratio

In most cities, a small share of addresses accounts for a disproportionate share of calls. Learn that share and you've found where tomorrow's calls live.

Deployment lag

A hot spot is a snapshot in time. Saturate it and it cools; walk away and it re-forms within weeks. The map is a heartbeat, not a headstone.

Grounded claims: · Predictive policing — Wikidata · Crime mapping — Wikidata · Companion data twin: hot-spots.json