# Source Methodology

AI Daily Digest starts from a fixed public-source set rather than discovering
feeds dynamically at runtime.

## Source List

The script embeds 90 RSS/Atom sources from the Hacker News Popularity Contest
2025 list referenced by Andrej Karpathy. Examples include:

- Simon Willison
- Paul Graham
- Dan Abramov
- Gwern
- Krebs on Security
- Daring Fireball
- Troy Hunt
- Mitchell Hashimoto
- Steve Blank
- Eli Bendersky
- Fabien Sanglard

The full source list lives in `scripts/digest.ts`.

## Processing Steps

1. Fetch feeds concurrently with timeout controls.
2. Parse RSS and Atom entries with lightweight XML helpers.
3. Filter by publication date using the requested time window.
4. Ask Gemini to score relevance, quality, and timeliness.
5. Ask Gemini to summarize selected articles and translate titles when needed.
6. Render one Markdown digest with category groupings and statistics.

## Scoring Dimensions

| Dimension | Meaning |
| --- | --- |
| Relevance | Technical usefulness for builders, operators, and AI practitioners |
| Quality | Depth, clarity, evidence, originality |
| Timeliness | How current or newly useful the article is within the time window |

## Output Philosophy

The report is designed for a fast morning scan:

- the top section answers "what changed?"
- top reads answer "what should I open first?"
- category groupings make browsing less linear
- Mermaid and ASCII charts make the same report useful in GitHub, Obsidian, and terminal contexts