{
  "id": 19641839,
  "title": "How I Fold Thirty Reports of the Same Event Into One Story",
  "url": "https://urgent.news/2026/11/12/how-i-fold-thirty-reports-of-the-same-event-into-one-story",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-11-12T07:30:15.000Z",
  "source": {
    "name": "HackerNoon",
    "slug": "hackernoon",
    "url": "https://hackernoon.com/how-i-fold-thirty-reports-of-the-same-event-into-one-story?source=rss"
  },
  "original_language": "en",
  "account": "When a major event occurs, such as the sinking of a ferry or a central bank changing interest rates, multiple news outlets typically report on the incident within a short period. Typically, an RSS reader would display each outlet's coverage separately, resulting in a cluttered feed with thirty separate cards. To address this issue, Hushread was developed as an RSS and news reader without relying on algorithms. Instead, it aims to present each event as a single story, complete with a timeline detailing who reported what and when.\n\nThe process begins by gathering all feeds in a queue, ordered based on their last update. Feeds from various sources, including RSS/Atom feeds, sites without feeds, subreddits, YouTube channels, and public Facebook pages, are all placed in the same queue. Each article is then processed to ensure it is in a readable format, with metadata and dates extracted using Mozilla Readability for the body and Defuddle for metadata. If an article lacks a date, it remains undated.\n\nNext, new articles are linked to an earlier one, rather than being clustered together. Every minute, new articles are bundled into a single call, and each article links to no more than one earlier article from the preceding 12 hours. This method creates chains of related articles, which are subsequently treated as events. The chaining approach is more efficient than re-clustering, as it remains accurate even as new articles continue to stream in. Each new article makes a single decision regarding linking, without requiring adjustments to previously linked articles.\n\nTo prevent incorrect linking, a rule was implemented to ensure that no single bridging article fuses two separate events. An example occurred in October 2026 when an article titled \"Netanyahu invites the pilot\" was mistakenly grouped with a separate story about a hijacking. The two stories were only 0.66 alike, while actual fragments of the hijacking story had a similarity range from 0.73 to 0.86. The solution involved adding a rule: an article can only merge two stories if the smaller one consists of eight or fewer articles, or if the centers of the two stories have a similarity score of 0.72 or higher. In all other cases, the article joins one story while leaving the other untouched.\n\nThe number of publishers, rather than the total number of articles, is counted to avoid duplication caused by an outlet syndicating the same story multiple times. A publisher is identified by its normalized outlet name combined with its domain, and the weight of a story is determined by the number of distinct publishers covering it. For a story to trend, it must be covered by at least five different publishers. This ensures that trending stories are genuinely significant and covered by multiple sources.\n\nA written story is only published after achieving one of two criteria: a burst of coverage involving eight or more publishers, with at least four new reports within an hour, or breadth, which requires coverage by fifteen or more publishers over several hours, with a busiest hour involving at least three reports. Stories are written by an AI model using the content from the outlets that covered the event, and the byline clearly indicates that no person authored the piece. Each claim is attributed to the specific outlet that made it, and timelines are linked to the original reports. The Hushread Stories page clearly states that an AI model wrote the story.\n\nOnce coverage ceases, a story can be updated with a dated addition if at least two new articles have been published, with updates occurring at most every six hours. Finally, once the coverage period concludes—either after 48 hours of inactivity or seven days since the initial report—the story is linked to the next relevant story if their centers are 0.78 alike or more and the similarity remains above 0.70. This process allows for a coherent narrative, where events are connected into threads, and long stories are broken down into chapters, each with its own summary.",
  "summary": "How an RSS reader folds thirty reports of one event into a single story: embedding links, a merge guard, publisher counts and corroboration bars.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}