An advice column grounded in Jesus—written by AI, not by Jesus, clergy, or a professional.
Methodology

How the letters are grounded

Every letter cites Scripture, and no citation comes from the model’s memory. The verified corpus is the source—the AI is only allowed to point at it.

1. A lawful, exact corpus

The launch translation is the public-domain, 66-book World English Bible (WEBP). The authoritative VPL archive is downloaded from eBible.org, hashed, parsed without textual normalization, and validated for book, chapter, verse, duplicate, and empty-text integrity.

2. Evidence before interpretation

The system normalizes the question, runs safety and intent checks, retrieves lexical and semantic candidates, expands context, and gives Gospel passages preference when the user asks specifically about Jesus’ teaching.

3. Opaque citation IDs

The model receives a limited evidence package and may return only stored citation IDs. The server rejects unknown IDs, resolves the reference, and hydrates the exact quotation from the corpus. A model-authored quotation is never trusted for display.

4. The letter and the Scripture stay distinguishable

The advice is the letter; the passages sit beneath it as receipts you can open, read in full, and check in context. Commentary is never presented as quoted Scripture, and disputed topics are named as disputed rather than flattened into a consensus that does not exist.

5. Important limits

Speaker metadata and passage headings are editorial aids, not unquestionable facts. Retrieval can miss relevant passages; models can produce poor synthesis; and no automated method settles theology. Reports and reviewed evaluation samples are part of the release gate.