Briefings

There is no list of sources to keep up. You describe the topic, who reads about it and what makes a story useful to them, and can add up to 10 example stories you would have wanted. Trusted outlets add to a story's score and have their feeds read on each run; blocked hosts are never read. Discovery uses free news indexes, and your own search key when you add one.

Each step (queries, triage, extract, write, check, translate) runs on a model of your own AI key, added on the AI keys page, so the model cost is yours. A key is encrypted when you save it and is never shown again. Every run has a budget in USD, 0.25 by default, shared between the steps; a call that would pass it isn't made. The owner's note after each run says what it cost.

Without sections, an issue is full stories (a paragraph or two each) followed by one-line items. With sections, the issue is a labelled skeleton you set once: up to 12 sections in reading order, each with a title, what belongs in it, a format and the most stories it holds (up to 25).

• Each story is put in the section whose purpose it fits best, or in none. A section never holds more than its most stories.
• sentences gives each story a sentence or two with its numbers; one-liners gives each one sourced line.
• A budget share is a section's percent of the extract and write steps' budget. The shares you give add up to 100 at most; with none, the budget is split evenly between the sections.
• When the budget runs short, sections with a lower priority number are written first (1 first); without priorities, the reading order decides. A section that runs out of its share is skipped and the note says so.
• A section with no story is left out of the issue, and the note lists it as empty.
• Each section is a heading of the issue. A translated version keeps the same sections, translated.

Set them on the briefing's page in the dashboard, or with update_briefing. Sending the list replaces it, and [] goes back to full stories then one-line items.

{
  "sections": [
    { "id": "lead", "title": "Lead", "format": "sentences", "maxItems": 1, "budgetShare": 30, "priority": 1,
      "purpose": "The day's most important story, with its numbers" },
    { "id": "deals", "title": "Deals", "format": "one-liners", "maxItems": 8, "budgetShare": 40, "priority": 2,
      "purpose": "Acquisitions and funding rounds, each with the amount" },
    { "id": "people", "title": "People moves", "format": "one-liners", "maxItems": 5,
      "purpose": "Executive hires and departures, with the person, role and company" }
  ]
}

Each run lists its sections with their status (written, empty, skipped for the budget, or failed), how many stories each holds and what each cost (get_briefing_run, and the run's page).

Every run makes a draft campaign, sends a test email of it to your test address, and sends you a separate note with its checks, its cost and a keep or drop link per story. Keeping and dropping teaches the briefing what you want: later runs pick more stories like the kept ones. Each link opens a page that asks before it records anything.

The pages a briefing reads are fetched by FlockStack-Briefings, which obeys robots.txt: see the FlockStack-Briefings fetcher.