A Jev search proposal with a name for every number

A Jev user describes an alternative to opaque embeddings: ask questions about each document and use the numerical answers as a vector, or list of scores. Their email example scores customer status, urgency, billing relevance and whether a reply is needed, then uses cosine similarity to match those scores against a search query’s vector. They argue that naming every dimension makes matches understandable; adding a question creates another dimension, while changing query scores adjusts priorities. On September 25, 2026, the user said they were trying the approach in Cora to make search fast.