Narrative Radar
Narrative Radar finds the themes running through a set of documents week by week, and always says how lopsided the sources behind them are. Each theme drills down to the documents and passages it comes from, and who published them. It is about documents and publishers, never about people.
Open it from Research in the main navigation, then Narrative Radar: the corpora, each corpus's themes week by week beside its bias report, and each theme's drill-down to documents, passages and source coverage.
Who can do what
| Role | Read corpora, themes and reports | Create and edit corpora, run them, name themes |
|---|---|---|
| Reader | Yes | No |
| Researcher | Yes | Yes |
| Project lead (in their project) | Yes | Yes |
| Editor | Yes | Yes |
| Owner | Yes | Yes |
Corpora
A corpus is the set of documents Narrative Radar reads. Give it a name and its permitted sources:
- a jurisdiction (a government, such as a county or a city), whose public documents it reads (for example Kitsap County's notices and staff reports), or
- your organization's own documents, from one of your connectors or all of them. A connector reads one source on a schedule.
A corpus reads only those sources: never another organization's documents. Removing a source keeps its record. A corpus you no longer need is archived.
Runs
Every Monday morning (UTC) each active corpus is run for the week; Run now runs it on demand. A run:
- collects the permitted documents of the last eight weeks and splits them into passages;
- groups similar passages into themes (a theme has at least three passages from at least two documents);
- matches each theme to last week's, so a theme keeps its name and history across weeks, with a trend (new, rising, steady, falling);
- writes the week's bias report.
Running a week again replaces nothing: each run is recorded with its own themes, and the latest completed run of a week is the one shown.
Themes
A theme is named by its top terms until a person names it. AI can propose a name and description; Accept takes the proposal (it is then marked as AI-proposed, accepted by you), or type your own. Every rename is in the audit log.
Drill-down shows, for a theme in any week, the documents and the exact passages behind it, with each document's publisher, source and date, and a source coverage panel: how many passages each publisher gave, and a sentence such as "80% of this theme comes from 2 publishers".
The bias report
Every week's themes come with the corpus's bias report, shown beside them: how the documents divide by publisher, kind of source, jurisdiction and plane (public or your own), which themes depend on one or two publishers (60% or more), and a warning when one publisher gives half the corpus or more. AI can add a short narrative over those numbers; a narrative whose figures do not match the report is withheld.
Themes and their AI names and narratives need the platform's AI tasks for Narrative Radar. Where they are not configured, themes keep their top-term names and the report has numbers but no narrative.