Built for the half hour that decides the day

The problem we actually had

Running one news channel is a research job with a video attached. Running three is a research job you cannot finish. Every morning produced the same forty minutes: twenty tabs, three news apps, a trends page, and a decision made largely on instinct — then a full production day committed to it, right or wrong.

The expensive failures were never bad edits. They were good videos about stories that three larger channels had covered two hours earlier, or stories that turned out to rest on a single unverified source, or stories that were genuinely huge and completely irrelevant to the audience we had spent two years building.

What we tried first

Trend lists were the obvious answer and the wrong one. They rank attention across an entire country, which is a different question from “what is worth a video on this channel today”. Keyword tools were built for evergreen search, where you have weeks to accumulate signals — news gives you hours. General-purpose AI chat was good at writing and useless at knowing what had actually happened in the last day.

What was missing was narrower than any of them: read the last 24 hours, collapse the duplicates, and rank what remains against one specific channel.

What we built instead

Four layers of sources scanned in parallel and clustered so that twelve outlets reporting one event become one story with a coverage count of twelve. Five weighted signals producing a score you can argue with, because every signal is shown. Then the part that was quietly costing us just as much: packaging written while the story is fresh, not at 9pm after the edit ran long.

The whole design brief was one sentence — spend the time on the decision and the video, not on the search.

What it deliberately does not do

• It does not publish for you, or ask for permissions on your channel.
• It does not write your script. The description gives you a sourced spine; the video is yours.
• It does not decide. It ranks, shows its working, and makes disagreeing easy.
• It does not hide the AI behind credits. You bring a key and see the real cost.

Where it is going

The near-term work is about the parts of the loop that are still manual: richer scheduling, deeper competition data before you commit to a story, and better tools for teams handing packages between researcher, writer and editor. The direction has not changed since the first version — make the morning decision faster and better, and leave the judgement where it belongs.