Reading the news without a news API

Drift handled the timeless well and the present not at all. You could wander Byzantine history for an hour, but if something was happening this month the app had no idea. A fair amount of curiosity starts from “what is going on with that”, so it was a real gap.

What I wanted was to pick a subject and wander the encyclopedia articles behind what has been happening in it.

A news API is the obvious route and would have taken about ten minutes. Three problems, in descending order of how much they bothered me:

• Licensing. Everything else in Drift is openly licensed, so it can be cut into cards, stored and shown, and I can say under what terms. Headlines and article text are copyrighted work under someone's terms of service, and I would be taking on obligations I am not set up to meet.
• Editorial position. Showing headlines means choosing outlets, and that is a stance. A one-person project should not acquire one by accident.
• Tone. News is written for urgency. Dropping it into a reading app aimed at the opposite does not work.

The thing I actually wanted was never headlines, it was subjects: which articles this month has been about.

Wikipedia's editors maintain a current events portal, one page per day, grouped into stable sections such as politics, science and technology, disasters and armed conflicts. Each entry is a line of prose full of links to encyclopedia articles. So the list already exists, in the corpus Drift was using anyway.

The rule is narrow. Drift reads only the link targets: which articles are pointed at, how often, and in which section. The prose around them is parsed to find the links and then discarded. No headline or editor-written sentence about an event is stored or shown.

The result is that “in the news” is not a new content source. It is the same encyclopedia, entered from a different angle: instead of picking a field, you pick a set of articles the world happened to be interested in. No new licensing, no editorial position, and every card is still an ordinary article.

One convenient detail: a month is about thirty day pages, and the API returns up to fifty pages of content per request, so a whole month costs one call.

Different from a news app, and more useful than I expected. You get the articles a month has been about, ranked by how often they came up, and then drift from there normally. It gives you background rather than updates, which is the part that is still true next week.

When a section runs out of unread articles, Drift widens outward from them instead of repeating, and says that it has.

The part worth remembering is that I nearly took on copyrighted content and an editorial stance to answer a question the openly licensed corpus already answered, mostly because the commercial API was the first thing that came to mind.

How all of this is licensed is on sources.