About

Who makes Radar?

Radar is a personal project by Yusuf Özkan, who runs it; the site's source code is open.

ozkanyusuf.com · GitHub · Radar's source code

What Radar is

Radar is a short, numbered daily edition of tech and AI news. An edition usually has five to eight stories, and each one ends with a single line on why it matters. Then the edition closes.

It is published in Turkish and English. Both are written from the same fact card, so they say the same thing.

How a story is made

A pipeline of seven steps prepares the stories. Only two of them call a language model; the other five are plain code.

  1. Collect. Pulls every source on the source list.
  2. Deduplicate. Groups the same story from different sources.
  3. Prefilter. Ranks the groups and keeps the top 30 candidates.
  4. Score. A small model, in one batched call, rates each candidate for relevance, novelty and usefulness and flags risky topics.
  5. Write. A mid-sized model first builds a fact card from the source, then writes the Turkish and English text from that card.
  6. Verify. Code checks that every number, version and name in the text appears in the source.
  7. Route. The story is published automatically or sent to a person for review.

The check is a safety net, not proof of accuracy: it cannot tell when a real number from the source is used in a wrong claim. That is why anything doubtful goes to a person.

The echo meter

The five arcs beside each story show how many independent sources covered it: one arc lights up per source, up to five, with the exact count next to them. It separates a story that is making waves from one that a single blog is excited about.

Who writes the stories?

Every story is written by a language model from the sources linked under it.

Stories on risky subjects are published only after a person has read them. The rest are published automatically once they pass the automatic check.

A story that is wrong is corrected or taken down.

Risky stories wait for a person

Some stories never publish themselves: security incidents, accusations, funding figures, third-party benchmark claims and regulation. The same goes for stories that fail the check or have no primary source.

They wait in a review queue. A story nobody reviews within 72 hours expires; nothing is published by default.

Found a wrong story?

If a story is wrong or misleading, please open an issue with the story's link. It is the most valuable report there is.

Open an issue on GitHub

Privacy

This site sets no cookies, does not track you and loads nothing from third parties: no analytics, no ads, no outside fonts. Every page is a static HTML file and is complete without scripts.

The site's own small script only adds conveniences: the archive search (its data comes from this site), the radar's tilt and the marks on stories you have read, the way the “Why it matters” line appears, a short opening on the day's first page, and the button that pauses the looping motion. If you have asked for less motion, none of it moves.

What you type in the archive search stays in your browser and in the page's address, so it is also in your browser history. When you reload the page or click a result, that address goes to this site only; it is sent to no one else.

At most two keys are written to the browser's session storage: radar:intro, which remembers that the opening has been shown, and radar:still, which remembers that you paused the motion. Both are gone when the tab closes. Nothing else is stored.