How it works

A crawler pulls news feeds and social posts continuously — wire services, newspaper front pages, Hacker News, Reddit, Bluesky, Mastodon, Lobsters. Everything lands in one table with its publish time, its score, and its comment count.

Items that describe the same event get grouped into a story by comparing their text. Stories that keep returning to the same subject get grouped again, into a narrative thread.

Claude then reads the strongest stories and extracts what a reader actually needs: a summary, a timeline of what happened when, the people involved and where they stand, and quotes. The quotes are copied verbatim from the source that published them. Nothing is paraphrased into quotation marks.

What the charts say

Every story carries a bar chart of arrivals by hour. Bar height is attention: each item counts once, plus the logarithm of its score and comments, so a wire report and a thousand-point thread do not look the same. Hover any bar for the plain item count.

Charts in a row share one ceiling, so a tall bar means more than a tall bar somewhere else. An hour with nothing in it draws nothing.

What is missing

Parsing costs money, so only the strongest stories get read. The rest still have pages — headline, activity, and the full list of who covered it — they just have no summary or timeline yet. Those pages say so.

Times are the publisher's own. Some feeds post a little into the future; those items land in the current hour rather than disappearing.