From Research to a Story Ready to Run
A newsroom’s constraint is not ideas. It is that finding, checking and shaping are done by the same people who write, on a clock that never stops.
We run an engine of this kind in production. See the unattended newsroom.
Continuous Coverage With a Human Gate
An engine reads live signals in your subject area and decides what is worth covering now. It writes against your own style specification, corroborates its claims against sources it actually fetches, and files everything into a private queue. An editor decides what runs. The engine holds no credential that can publish, so its worst day fills a queue rather than embarrasses the masthead.
The Research Desk
Reporters lose hours to documents: a budget, an annual report, a court order, a tender, a transcript. The desk turns those into a structured corpus, then surfaces the publishable angle: what changed against last year, which figure contradicts the press release, who is named and in what context, and which passage is worth quoting.
Reports, filings, orders, transcripts, whatever arrives, including material that is only a scan.
Entities, figures, dates and claims typed and addressed, so a number can be pointed at rather than retyped.
Against last year, against the stated position, against what the same body said elsewhere. Contradiction is the story.
A brief with the angle, the supporting passages, and a link to the exact page behind each one.
The desk hands over a brief. A reporter given an angle and the passages behind it still writes the piece. The byline is still theirs.
Where the Human Sits
It can write to one queue. It holds no credential for the public site, so this is a limit of what it can do rather than a policy it is asked to follow.
A separate, audited step publishes it. Tests and content checks run there, so a broken batch stops before readers see it.
The decision is recorded, which over time becomes the clearest description anyone has of what the publication will and will not run.
Those recorded decisions are worth more than they look. A house style document says what an editor believes. A few hundred approvals and rejections say what an editor actually does, and that is the thing worth feeding back into how the engine writes.
Analytics That Answer Editorial Questions
Standard dashboards report traffic. Editors need to know which subjects earn attention that lasts, which formats work for which desks, where coverage is thin against demonstrated demand, and what a piece is worth beyond its first day. That requires your content to be typed against a real taxonomy rather than tagged loosely, so a coverage figure is a fact instead of an artefact of inconsistent labels.
Discourse, Measured Rather Than Sensed
Newsrooms and researchers both want to say what a public conversation is doing. Sentiment scores are too crude to be worth printing. What can be defended is structure: which claims are circulating, who introduced them, how they mutate as they spread, which communities carry which framing, and when a subject crosses from one community into another.
The unit of analysis is a claim rather than a post, so the same assertion in ten thousand different wordings counts once. Every figure resolves back to the posts behind it, because a study that cannot show its underlying material cannot be published.
One number for a nation's feeling. Impossible to defend, impossible to act on, and wrong in ways nobody can audit.
Which assertions exist, where each first appeared, how the wording drifts, which groups carry which version, and where they overlap. Every count opens the posts beneath it.
Built for a study that has to survive review. If a finding cannot be traced to material a reader can inspect, it is an opinion with a chart attached.
Publishing Operations
For book and comic publishers the same discipline produces finished books rather than articles. A research corpus, sourced scripts, art held consistent across a title, and output a press and an editor can both use. That work is described in full in the graphic stories case study.