How this brief is made
This is a daily news brief for a general reader. It reads a fixed set of public sources, finds the stories that matter most, and rewrites them in plain English so you can understand the day in about fifteen minutes over coffee. A new edition is published every morning.
Every three hours a collector reads the sources. Twice a day the day's stories are grouped, scored, and written into one brief. A person reviews quality, but no person writes the copy. The rest of this page explains how a story gets from a raw headline to a place in the brief.
The pipeline
Sources feed adapters, which normalize each item and store it. Near duplicates are merged, related items are clustered, and each cluster is scored for how much it matters to a general reader. At 6:00 am Eastern an editor stage selects and orders the day's stories, a writer stage drafts each article from the source excerpts, and a readability check runs before anything publishes.
How stories are scored
On a busy day there can be around two hundred candidate stories and room for only fifteen to twenty. To choose, each story is rated from 1 to 10. The rating answers a single question:
If a general reader missed this story entirely, would that be a real gap in their understanding of the week?
Not "is this interesting to someone," because almost everything is interesting to someone. The test is whether it matters to an ordinary reader with limited attention and no professional stake. The scale is used in full, from noise to the story of the day.
- 1 to 2. Noise.
- Press releases, rumors, and minor updates. You lose nothing by never seeing it.
- 3 to 4. Real but narrow.
- A genuine event that matters inside one industry and almost nowhere else.
- 5 to 6. Solid.
- A real development with consequences a general reader can recognize. Most legitimate news lands here.
- 7 to 8. Strong.
- Changes something concrete for a lot of people, or is the clearest sign yet of a shift worth tracking.
- 9 to 10. Lead.
- The story of the day. Skipping it would leave a reader genuinely uninformed.
A score goes up when a story:
- reaches a lot of people, and reaches them directly;
- is something that actually happened, such as a rule taking effect, a court ruling, or a number being published, rather than an announcement of intentions;
- is hard to undo;
- is the first of its kind, or the moment a long trend becomes undeniable;
- helps explain an ongoing situation instead of just adding one more update.
A score goes down when a story is:
- speculation, such as "sources say a company is considering" something;
- a repeat of a continuing situation with nothing new;
- marketing, such as a product launch or funding round that changes little for ordinary people;
- a prediction, forecast, or analyst target;
- coverage of other coverage, such as reaction pieces and commentary.
How many outlets covered a story does not raise its score. A single solid report of something consequential outranks eight outlets rewriting the same press release. Along with the score, each story is filed under one topic (Tech, AI, Business, Finance, US Politics, World, Regulation, Science, Cyber, or Sports) and given a confidence level, which is an honest measure of how sure the rating is given how thin or clear the sources were.
The score is advice, not a verdict. It feeds the editor stage below, which makes the final call on what runs. A high score does not guarantee a slot, and a modest one is not automatically cut.
Editorial policy
The scores are anchored to a fixed rubric with worked examples at low, middle, and high marks, so the same story would score the same way on any day. Each section runs at least two and at most four stories, with a global ranking across the day. Sections with nothing worth running collapse into a short "Briefly" list, and when several sections are empty the whole edition shrinks. On a genuinely quiet day the brief runs in a shorter format and says so.
Articles paraphrase the sources and always link back to them. Specific facts such as dates, numbers, and names come only from the source material, never invented. Quotes are used only when a real person is speaking, at most one per story, kept short, and always attributed and linked. Every story is checked against a plain-language reading standard before it publishes.
Running totals
36 editions published so far, 482 stories written. Lifetime totals from the permanent archive will appear here once archival is live.
The stack
The pipeline is Python. Storage is a columnar table format on object storage, read with an in-process analytics engine. The site is a static build with no client framework and effectively no JavaScript. Language models do exactly four jobs: scoring, editing, writing, and the audio script. Everything else is ordinary deterministic code.
Who builds it
Built and maintained by one person as a personal project.