We are releasing all 41,995 verses from the "Standard Works" (sourced from public-domain editions), each paired with an LLM-generated chapter-context block and a 3072-dimension embedding of the two combined. This dataset is an independent test of Anthropic's Contextual Retrieval technique on a domain its original evaluation did not cover: short, archaic, formulaic religious text from antiquity. The technique was introduced in Introducing Contextual Retrieval (Anthropic, 2024), which describes two components — Contextual Embeddings (prepend generated context before embedding) and Contextual BM25 (prepend it before lexical indexing). Anthropic reported that Contextual Embeddings alone reduced…