The wire story is the beginning, not the article.
After the fact-check the newsroom keeps digging: background, comparison figures, context — with web search and your own market data.
What you can do with it
- Start a research run for a single story and read its log and result right there.
- Research whole filter results in bulk.
- Run research as an automatic stage, bounded by news value and time window.
- Cap the output length per run — 1,200 tokens by default, previously unlimited.
- Let the result feed into the draft rather than sit beside it as an appendix.
The thinking behind it
Wire copy tells you that something happened. What happened — the figure behind it, the comparison, the context — is rarely in there. If everyone runs the same wire story, no distinguishable article comes out of it.
That is the step we automate: not the writing, the looking up. What the model gets in the end is a dossier, not a wire story.
The run belongs to the story, not to a page of its own. Before, you had to switch surfaces to see what research had turned up.
Improvements
Research sits with the story
The run for a single story — starting it, reading the log, seeing the result — used to hang off a stage page. It now hangs off the story itself. That is where it belongs.
Time window and output length bounded
The automatic stages' time window dropped from 14 days to 2: after an outage, a large window worked through the entire backlog in one go. Plus a cap on output length per run, where there was none.
Chained to the stages before it
Classification, fact extraction, fact-check and research now hand the story on to each other instead of each scanning the inventory independently.
Give feedback or ask a question
Missing something, bothered by something, unsure about something? Write it down here — it goes straight to the people working on this feature.