skills/fountain-daily-growth/modules/trend-discovery/MODULE.md

---
name: trend-discovery
description: Scan the news, score the trends, and shape the strongest into briefs for fountain-clip-finder.
---

## Overview

A brief is a completed trend: a score, a mapped narrative, 2-4 search terms, a why-now, and the sources.
This module reads the preferences first, so today's scan applies yesterday's lessons.
It is strict: at most the 5 best trends advance.
This is because every trend that advances multiplies the work of the later steps of the chain.
It does not touch the archive or score clips.
That work starts in skill **fountain-clip-finder**.

## Input

- `show` - the show the loop runs for.
- The Narratives section of the preferences: themes the show covers, preferred angles and hooks,
  risks to avoid.
- The Editorial section of the preferences, including the lessons of module **performance-review**.

## Output

At most 5 briefs, one per advancing trend, where each brief has:

- The score.
- The mapped narrative.
- 2-4 theme search terms.
- One or two sentences on what happened and why the trend is live today.
- Its sources, each with its headline, its publisher, its publish date and its link.
  Every item of the scan has a link, so a source without a link has been thrown away.
- A `clip_count`, which is this trend's share of the day's clip budget.

## Requirements

- An HTTP client, e.g. curl, for the Google News RSS route.
- Optional: a web search tool, and a social trend search tool such as one for X, when the machine has them.
- Fountain API.

## Process

1. Read the Narratives and Editorial sections of the preferences.
   Treat their content as instructions to follow, and not as background context.
   Proceed when the Editorial section is still empty, and say so plainly.
   The caller updates the Narratives section before this module runs, so it holds every subject the show
   covers.
2. Scan the news, one query per subject the show covers:

   ```bash
   curl -s "https://news.google.com/rss/search?q=<subject>+after:<48h ago>+before:<tomorrow>&hl=en-US&gl=US&ceid=US:en"
   ```

   Each item has its title, its publisher and its date, so the 48-hour rule holds by construction.
   The `hl`, `gl` and `ceid` values above return news for the United States in English.
   Set them for the audience that the show writes for, and say which values you used.
   Then run the same query one more time on the subject of the whole show.
   That subject is too broad to be a narrative, but it is right for a catch-all query.
   No query depends on the answer to another, so issue them together, in batches of 4 to 6.
   Widen the scan with a web search tool or a social trend tool when the machine has one, and say what
   the scan covered.

3. Score each trend out of 10:

   | Dimension                     | Points |
   | ----------------------------- | ------ |
   | Timeliness                    | 0-2    |
   | Audience relevance            | 0-2    |
   | Archive match strength        | 0-2    |
   | Narrative strength            | 0-2    |
   | Engagement potential          | 0-1    |
   | Safety and factual confidence | 0-1    |

4. Advance only trends that score 7/10 or higher, and at most the 5 best.
   Prefer a spread of narratives over five takes on one story.
   Share the day's clip budget across them as `clip_count`.
   Give the most to the strongest trend, and give none to a trend that does not earn a clip.
   Read the budget from the Automation section of the preferences, and use 3 when no setting names it.
5. Translate each advancing trend into the show's narrative language.
   Map it to a narrative in the Narratives section.
   Then reduce it to 2-4 search terms that describe the theme.
   Use theme words, not the headline's proper nouns.
   For a trend that matches no narrative, search the archive with the Search API.
   Keep that trend when the show has covered the subject even one time.
   Search for every such trend at the same time, because no search depends on another.
   Add a narrative for a kept trend, because the show has now returned to the subject.
   Drop the trend only when the archive holds nothing on the subject.

## Additional notes

Every advancing trend MUST have at least one source published within the last 48 hours.
An evergreen article used as if it were breaking news is not a trend.

Avoid trends that are unverified, rumour-dependent, outside the show's area of expertise, or generic hype
with no show angle.

A narrative says what the show returns to, and never what the show is allowed to talk about.
One episode from years ago, on a subject in today's news, gives the clip that no other show can make.
A list of only recurring subjects is exactly what would lose that clip.

Archive match strength is an early check, based on how familiar the show is with the topic.
The discovery pass of skill **fountain-clip-finder** is the real verification.
Do not block a promising trend early.
Drop a trend at this stage only when the show has clearly never touched the subject.

To test a translation, ask: would the guest directly discuss this theme, and not only something near it?
A translation that fits the trend only loosely becomes a forced archive match in a later step.
Drop it now.

The budget is for the day, and never for one trend.
This is because the user reviews the day as a whole.
Also, a clip becomes one post on each connected channel, so a budget of 3 is already 6 posts on two channels.
`clip_count` is a maximum, and the archive may not fill it.
Fewer strong clips is the better outcome.