How a story gets here

The pipeline

1. Monitor. Official RSS, Atom and JSON feeds published by the vendors themselves, plus GitHub release feeds. The full list is public.
2. Deduplicate. By canonical URL, then by headline similarity, then — only for genuinely ambiguous pairs — by asking a model. Items that duplicate something already published are dropped.
3. Triage. Is there a concrete change here? Most feed items fail this and are rejected with the reason recorded.
4. Cluster. Items describing the same development are grouped into one story.
5. Research. We fetch the specific page the feed linked to and extract its readable text, then pull out claims — each with the verbatim sentence from that page which supports it.
6. Verify. Every supporting sentence is checked back against the text we actually fetched. A claim whose evidence is not really there is marked unverified and does not count toward publication.
7. Write. The article is drafted from the verified fact set — not from the source's prose, which the drafting step never sees.
8. Review and score. Fact-consistency and quality checks, then a confidence score. See the AI disclosure for the exact thresholds.
9. Publish, hold or reject.

What our fetcher does

Our user agent is:

Ali3NDevBot/1.0 (+https://www.ali3nstudios.com/dev/methodology)

• It reads the feed URLs listed on our sources page, on the schedule set per source (typically hourly or slower).
• It uses conditional requests, so an unchanged feed costs a single 304.
• When researching a story, it fetches the one URL that the feed entry itself pointed at — nothing else.

What it does not do

• It is not a crawler. It does not follow links out of a page, does not walk a sitemap, and does not discover URLs. It fetches feeds it was told about, and the specific articles those feeds published.
• It does not fetch anything behind a paywall or a login.
• It does not archive or republish page content. Extracted text is used to verify facts and is never served to readers.
• It does not hammer anything. One request per feed per polling interval.

If you publish one of the feeds we read and would rather we did not, or would like a different rate, say so and we will remove or adjust it.

Uncertainty

When a source does not say something, the article says the source does not say it. We would rather publish a shorter article with a gap in it than a complete-looking one with a guess in the gap.