How the Repleader news pages are made
The news pages group stories from publishers' public news listings, rank them by how many outlets carry them and by the court and agency activity of the companies they name, and link every story to the publisher. Articles are not copied.
What a story card shows
- The publisher's headline, quoted exactly as written and linked to the original article. We do not rewrite headlines. A headline that breaks our house rules or falls outside our scope is not shown; another outlet's headline for the same story may be shown instead.
- At most two sentences of ours, written by a language model from two inputs only: the headline and the description the publisher put in the page's own meta tags (or in its news listing). Code then checks that every name, number and claim in the two sentences appears in those inputs and that allegations read as allegations. A summary that fails is discarded and the card shows the headline alone.
- The other outlets carrying the same story, the names we matched on Repleader pages, and the counts from our federal docket tables for those names, each with its method tag.
A card keeps the headline, the link, the publisher, the date and the publisher's own short summary. An article body is not stored or shown. A summary longer than a short paragraph is treated as article text and dropped.
Sources
Publishers' public RSS and Atom news listings, read only where the site's robots.txt allows it. Government releases (courts and agencies) are public records, and agency enforcement releases are the same ones the enforcement pages read.
Publishers: Administrative Office of the U.S. Courts, U.S. Court of Appeals for the Federal Circuit, U.S. Equal Employment Opportunity Commission, U.S. Food and Drug Administration, Consumer Financial Protection Bureau, ABA Journal, Legaltech News, The National Law Journal, The Recorder, New York Law Journal, Texas Lawyer, Above the Law, SCOTUSblog, Lawfare, Just Security, Courthouse News Service, IPWatchdog, How Appealing, Original Jurisdiction, Artificial Lawyer, LawSites, Legal IT Insider, LexisNexis, Clio, Intapp, Gibson Dunn, LawNext podcast, SEC EDGAR, SEC EDGAR.
Grouping
Two stories are the same story when they were published within three days of each other and their headlines share enough words: half of their distinct words on their own, or a quarter when both also name the same company, AI company or federal district by our exact match. Names are never merged by similarity. The card links first to the primary source when one is in the group: a court or agency page, then the company's or law firm's own release, then a press wire, then the earliest press coverage.
Matching names
A company name in a headline links to a Repleader page only on an exact match. AI companies match one of the listed spellings on the AI lawsuits tracker. Companies, funds and insurers match the registry rule on the companies method page, with two rules added for headlines: a single bare word never matches, and a bare name without a legal form matches only a company whose page is published. Federal districts match the full name, such as "Southern District of New York". The counts beside a name are read from the same tables as its page.
Ranking
Each group gets a score between zero and one: 0.35 modeled times the outlets carrying it (scaled so that ten outlets count fully), plus 0.25 modeled times the federal lawsuits naming its companies as defendants filed in the last ninety days (on a log scale), plus 0.2 modeled when an agency release or an agency action against a named company in the last year is involved, plus 0.2 modeled times its freshness, which halves after twelve hours. Groups older than a week drop to a tenth. The score orders the cards; it is not shown as a number and it says nothing about the merits of any case.
Stories on the front page
The front page at /news/ tells each day as stories. A story is one event across every source that carries it: the agency's release, the docket it names, the press coverage and our own pages for the company, judge, matter, MDL and state precedent involved. Code puts a story together from exact joins only, with no language model:
- an agency release joins the press group that carries that release, or a press headline within three days that carries a respondent's name of two distinctive words;
- a release joins the docket its case number and court name;
- the same court and case number seen in our lawsuit alerts, our case summaries and the AI lawsuits tracker are one docket;
- new suits first seen on the same day that name the same company are one story about that company, and each suit counts toward its most named company only;
- a suit joins its group of related suits, and a group joins its MDL.
Our headlines are written by rules from the records (for example "SEC files a civil complaint in E.D. New York"). A story with no record of ours behind it keeps the publisher's own headline, in quotes. A person is never named in our headline; a suit brought by an individual plaintiff is counted and never listed by number.
A story's source count is its distinct outlets and agencies, plus one for the court record when it names a docket. Stories rank by sources, then dockets, then outlets, then the newest event. The day's five top stories take the largest agency action, wave of new suits against one company, press story and new MDL or group of suits first, then the next largest. Each section below names its window. The day's page is kept at /news/ until the next day, then at its own dated address; a past day is never rewritten.
Editorial rules
- Allegations stay allegations. A lawsuit is an allegation, not a finding.
- Stories about individuals in criminal, family or immigration matters are not listed.
- Corrections: when a publisher changes a headline, the card follows. To report an error, use the "Report an error" form on any page or the corrections page.
- Takedowns: a publisher or a person named can ask us to remove a card through the takedown page; a removed story stays off every page of the site.
- Subscription sites are marked "subscription". We link to them and never copy around a paywall.