Enrichment nodes¶
Enrichment nodes refine or look up the entities identified by an extractor or assign metadata to the article text itself.
| Node | What it adds |
|---|---|
| Article metadata | Tags for the article as a whole, such as format, topic, subject, or other configured metadata categories |
| Geocode | Turns a place name into map coordinates and a normalized address, so locations can be plotted and searched geographically |
Article metadata¶
The Article Metadata node classifies the full article, rather than one extracted entity. Use it when you want each processed item to carry tags that can later power search, filtering, dashboards, or downstream analysis.
You can choose from preset tag categories or define project-specific metadata types. The selectable categories include:
| Category | What it describes | Taxonomy |
|---|---|---|
| Critical information need | Civic information needs the story helps answer | Critical Information Need |
| Custom | Project-specific metadata defined by your team | Custom |
| Format | How the story is structured, such as news story, profile, obituary, or opinion | Format |
| Scope | The story's geographic, civic, or editorial scope | Scope |
| Subject | The concrete thing the story is primarily about, such as a legal case, public meeting, traffic crash, or development project | Subject |
| Timeframe | The time horizon the story concerns, such as breaking, recent, ongoing, or historical coverage | Timeframe |
| Topic | Broad coverage areas, such as local government, public safety, sports, or education | Topic |
| User need | The audience need the story serves, such as explaining, updating, guiding, or contextualizing | User Need |
Article metadata becomes available to Backfield API, where applications can use it for filtering and analysis.
The node records the selected value, a rationale, and confidence information that editors can inspect on the processed item's Meta tab. Reviewers can correct, add, or remove tags without changing the original model output.
Choose categories that serve a real downstream question. A small, consistently applied taxonomy is usually more useful than many overlapping tags.
Geocoding places¶
The Geocode node consumes the locations produced by Place Extract on the same branch. It uses the extracted name, type, address components, jurisdiction, and contextual hints to look for a defensible map result. Agate rejects flows where Geocode appears without Place Extract upstream on that branch; a Place Extract on a sibling branch does not satisfy the requirement.
A location can finish with:
- accepted point, line, or area geography;
- a match to known Stylebook geography;
- identity information but no safe geometry;
- a review-required result explaining why the lookup was uncertain.
Backfield does not force every place onto a map. Keeping “no geography” is safer than attaching a story to the wrong city, address, district, or venue. Editors can inspect and adjust results on the processed item's Places tab.
Geocoding uses the services configured under Integrations. Missing credentials, ambiguous names, conflicting jurisdictions, or incomplete addresses can all affect the result without invalidating the extracted mention itself.
Enrichment and provenance¶
Enrichment adds information to an existing article or entity result; it does not replace the source evidence. The original mention remains available so an editor can judge whether the added tag or geography is supported.