03

Focuses

Topic areas that automatically classify your articles using on-device AI.

What focuses are

A focus is a topic area you care about. When you create a focus, Editions uses a zero-shot classification model running entirely on your server to score how well each article matches that topic.

No training data is needed. No data leaves your machine. Just describe what you're interested in.

TechnologyClimate & EnergyScienceMachine Learning ResearchIndie WebEuropean Policy

Plain-language topic descriptions

Classification result
Technology
94%
Climate & Energy
12%
Science
67%
European Policy
3%
Article: "The quiet revolution in reader design" — Ars Technica

Each article is scored against every focus

How classification works

When an article is analysed, the classifier evaluates it against all your focuses and assigns a confidence score (0.0 to 1.0) for each one. Articles above a focus's minimum confidence threshold appear in that focus's feed.

The classifier uses a natural language inference model (bart-large-mnli), so the focus name acts as a natural language description. Descriptive names work better:

"AI" Broad — matches airline industry, artificial intelligence, anything
"Artificial Intelligence" Better — more specific signal
"Machine Learning Research" Best — narrow, well-defined topic

Articles can belong to multiple focuses simultaneously. A climate technology article might appear in both "Climate" and "Technology".

Creating a focus

01 Navigate to Focuses in the sidebar
02 Click the + button
03 Enter a name — be descriptive (see tips above)
04 Optionally add a description for extra classifier context
05 Set the minimum confidence threshold — default (0%) works to start
06 Optionally restrict to specific sources
07 Click Create

Source associations

When linking sources to a focus, each source has a mode:

Match

Default mode. Uses the classifier and applies the confidence threshold.

Always

Every article from this source gets confidence 1.0, skipping the classifier. Useful when a source is perfectly on-topic.

Each source also has a weight within the focus, which influences ranking.

Voting in focus feeds

Votes mean different things depending on context. In a focus feed, a vote is about relevance to that topic — not general article quality.

Votes propagate through semantic similarity — a few votes on climate policy articles will shift the ranking of hundreds of similar pieces you haven't even seen yet.

Global feed
Is this a good article?
Technology focus
Does this belong here?
Relevance

Global votes rate quality — focus votes rate relevance

The focus feed

Each focus has its own article feed, sorted by:

Top Ranked by confidence, votes, and recency
Recent Chronological by publish date

Filter by time window and read status, just like the main feed.

Tips

Start with 3-5 broad focuses

You can always get more specific later.

Use the description field

"Articles about renewable energy, electric vehicles, carbon capture, and climate policy" gives the classifier much more to work with than just "Climate".

Don't worry about overlap

Articles in multiple focuses is a feature, not a bug. A tech policy article should appear in both "Technology" and "Politics".

Re-analyse after adding focuses

Existing articles won't be classified against new focuses until they're re-analysed.