How we report stories

What this page covers

How a story moves from a publisher’s RSS feed to a published Play The Kitchen article, and what a human editor approves before anything goes live.

1. Ingestion

Our scout fetches a curated set of pickleball publishers on a schedule (every 30 minutes for tier-1 sources, every hour for tier-2). Each item is normalized — we compute a content hash so we can recognise the same story arriving from a different outlet.

2. Clustering

Items reporting the same news are grouped into a story cluster. A cluster is the atomic unit our editors review — “Ben Johns wins Cincinnati” is one cluster, even if six outlets cover it.

3. Classification & scoring

Each cluster is classified into one of eight types (breaking news, match result, rankings update, player movement, tournament update, business, feature, or low-value). A combined score weighs source trust, corroboration, recency, and entity importance. Clusters below the threshold are held or discarded.

4. Drafting

Above-threshold clusters get a draft article generated from the underlying source bodies. The draft must reference at least two distinct cluster items, must not exceed a paragraph-level overlap threshold with any source, and must include a list of cited sources with URLs.

5. Human approval

Nothing publishes automatically. A human editor reviews every draft against a publish gate — headline length, source count, semantic overlap, factual claims — before the article goes live.

What we never do

• Publish a story sourced only from a single tier-3 (signal-only) source.
• Re-write a publisher’s article verbatim — overlap is checked at trigram and semantic levels.
• Strip attribution. Every claim links back to where we found it.