The product guarantee
pericopa is the AI-era Bible tool whose core writes no Scripture. The retrieval and attribution core uses embeddings, vector search, and mechanical NLP — which can only surface verses that already exist — and never a model that produces text presented as, or about, Scripture.
What the core uses
Embeddings, vector and semantic ranking, and mechanical NLP run in the core. They surface real verses that already exist in a held edition; they cannot fabricate Scripture. That constraint is what makes the core trustworthy — its output is bound to the corpus, never composed.
What the core never does
No generated verse, no paraphrase standing in for Scripture, no authored take on what a passage means reaches a core surface. A reader can retrieve, read, open-to-source, compare, and save having never invoked a generative model.
The opt-in tier
A generative tier may exist, but it is opt-in, never the default surface, and clearly fenced from the core. When present it is grounded in retrieval — it surfaces and explains what a real, cited source says — and it never speaks in pericopa's own voice as a ruling.
Computed, never generated
Whether a book belongs to a canon is a mechanical fact of the versification scheme — computed from the canon tables, never generated. The same locator is accepted under one scheme and rejected under another, with no model in the path.
Tobit 1:1
Protestant (kjv-prot)
not in this canon
Catholic (catholic-vulgate)
in canon
Each row below is a real decode from @pericopa/corpus — the accept / reject outcome is the canon table's, computed at build time.