Listen: https://serve.buildtall.systems/8626c6092cfe1cc17dbf82e8fbc0ee27a47a3fe7729a66217861197ab050658d.mp3
marginal revolution read out: Adlerlicht, reporting on crime in Germany.
Adlerlicht, reporting on crime in Germany is a brief announcement from Marginal REVOLUTION. The author's note says the project is bilingual, built with AI agents, and that we are going to see much more of this kind of effort. An English-language explainer is linked separately.
The first linked page is titled In both German and English. It is the main German-language site of Adlerlicht, and the author's commentary frames it as bilingual, built with AI agents, and an early example of a coming wave. The page opens with an arresting claim: thousands of police reports are published every day across dozens of German portals, public in principle but practically invisible. Adlerlicht wants to end that. It presents itself as the first live picture of Germany in real time. The landing page shows more than two hundred thirty-two thousand mapped data points, updated every fifteen minutes, and a feed of reports from the last twenty-four hours. The examples are concrete and grim: a thirty-two-year-old fleeing police in Sailauf who struck a thirteen-year-old on the pavement; a knife attack in Gera; an exploded cigarette machine in Neubrandenburg; a Cologne man who found transfers and cash withdrawals after an online date. The categories run from killings and knife attacks to weapons, sexual offences, assault, robbery, theft, arson, drugs, fraud, vandalism, traffic, missing persons, residence law, extremism, environmental offences, and public order. The site emphasizes that everything comes from police releases, not newspapers or social media. It maps cases to the street when the police name one, otherwise to the town, and it never guesses. It also makes clear that showing where cases happen also shows where it is quiet. There are pages for each offence and each town. The Berlin page, for example, lists ninety reports in November from five localities and forty-four postal codes, with Charlottenburg, Kreuzberg, and Neukölln among the most frequent. Each case links to the original police release, with no names and no faces. The project says it does not comment, it shows, and it leaves the classification to the reader.
The second linked page is the English-language explainer, titled here. The author's note points to it as exactly that, the English-language explainer. The page is credited to Alan Özalp and describes the project as civic data infrastructure. It explains that Germany's police forces publish thousands of press releases daily, an unusual degree of openness, but the reports are scattered across dozens of portals, in different formats, and nearly impossible to take in. What is public in theory stays invisible in practice. Adlerlicht collects the releases, puts each one on a map, and makes them readable by town, district, state, and offence. It is free, requires no account, and describes itself as not a news outlet and not an activist project. It exists, the page says, so people can form their own picture, ask their own questions, and decide with the facts in front of them. The English page also walks through the method in seven steps: collect police press releases every hour, filter out non-incident notices, extract structured fields using a language model, locate the scene only as precisely as the text allows, rewrite a neutral headline and short report in plain German, rank roughly eight hundred reports a day by fixed criteria, and publish each case with a link back to the original release. It is unusually candid about limits: the police do not report everything; machines make mistakes; and the record has gaps, including an outage in May twenty twenty-six. It also says plainly that no person reads individual reports before they go live, because at eight hundred a day that claim would not be honest. The project does not name suspects or victims, does not speculate about origin, motive, or guilt, and does not sell reader data.
The through line is the transformation of scattered public data into a live civic layer. The German page shows the product; the English page explains the machinery and the boundaries. The author's brief commentary frames both as a sign of what is coming when AI agents can assemble this kind of infrastructure quickly and cheaply.
