My AI news pipeline was silently throwing away 4 good articles a day (13 days in)

Most AI newsletters paste RSS summaries. I built a pipeline that reads each day's top articles in full, notices when several outlets are covering the same event (that's a significance signal), and writes one synthesis of what the day adds up to. Any jargon term becomes a clickable plain-English explainer, written the same day.

Where it stands after 13 days: 126 articles analyzed, 58 glossary terms (each one grew out of a daily briefing rather than being written separately), and the whole thing runs unattended every morning, in English and Korean. It's a static site (Python + Jinja2) on Vercel with the daily reading and writing running on Claude, so a day of content costs me almost nothing.

Hardest bug so far: same-story detection merged on a single shared headline keyword. It decided that a listicle about dinner parties and a story about Google's AI search were "the same event", and it was silently discarding 4 good articles a day. Excluding company and product names as merge evidence fixed it.

Two things I'd love honest feedback on. Does the "so what" analysis actually add value over a headline list, or do you skim past it? And when you land on the page, is it clear within 10 seconds how this differs from other AI digests?

https://www.dailyaithread.com/en/
(free, no signup to read)