Notes from the field

How this search works

No server, no search service, no library. The index is built in your browser when the page loads and the whole thing runs offline, including from a local file.

1. Normalise Text is lowercased, punctuation dropped, stopwords removed and words reduced to a common stem, so platforms, platform and platforming all become one term.
2. Weight by field Where a word appears matters more than how often. A hit in the title counts 3.4x, in a topic chip 2.6x, in the area and depth labels 2.2x, in the tag vocabulary 1.8x and in the summary 1x.
3. Rank with BM25 The same ranking function Lucene and Elasticsearch use. Rare words count for more than common ones, repeated hits give diminishing returns, and longer documents do not win just for being longer. score = ∑ idf(t) · f(t,d)(k₁+1) / (f(t,d) + k₁(1 - b + b · |d|/avg|d|))
4. Match as you type An unfinished word matches by prefix at a small discount, so results narrow on every keystroke instead of waiting for a complete term.
5. Forgive typos A word that matches nothing is compared against the index by character trigram similarity. Above 0.48 it is corrected, and the correction is stated rather than applied silently.
6. Expand shorthand An alias table maps how people type to how the posts are written, so k8s, iam, spof and rag reach the right articles. Expansion happens on the query only, never in the index.

Try it: redis iam identiy org design