politikradar(1)   cluster diagnostics

Cluster diagnostics

§ 01

The map

Every active event placed at its UMAP coordinates, colored by cluster. Noise points are greyed out. Cluster centroids are labeled. Hover any dot to read the event.

§ 02

Cluster anatomy

One block per cluster, ranked by member count. Bars show which domains, types, scopes, and parties are over- or under-represented in this cluster relative to the era as a whole. The five most-central events are the ones the LLM saw when writing the label.

§ 03

How the data flows

Every event contributes to every tag it carries — an event with two domains shows up in both ribbons. Click any node to filter the flow. The list below shows the actual events in the filtered slice.

Filter Click a node to focus a path.
01 Reichweite Wo spielt es?
02 Domäne Worum geht's?
03 Ereignistyp Was für ein Vorgang?
04 Cluster Wo landet es?
Click a node above to see the events flowing through it.
§ 04

Fingerprint composition

How clusters differ across the explicit fingerprint dimensions. Each row is a cluster, each column a feature. For Domänen, Typen and Reichweite, cells show how much more or less that tag appears in this cluster vs the era as a whole — measured in percentage points. So +25 pp means the cluster is 25 percentage points more "foreign_policy" than average; dark cells mean strong over-representation, cool slate means strong under. For Parteien, the cell is the mean party-benefit score within the cluster (range roughly −1 to +1). Read horizontally for what defines a cluster; vertically for what separates clusters.

§ 05

Era summary

LLM-written narrative of the previous era, generated on refit. Empty on the first era of a project.