0003-citation-grammar

ADR-0003: Per-class citation grammar

Status: Accepted Date: 2026-05-19

Context

User’s #1 fear about drift: citations written inconsistently across notes (“this way vs that way”). Without a locked grammar, AI improvises and dedup/queries break.

Decision

Per-source-type citation grammar, not a unified scheme. Five classes locked at start: Divine Principle, CSG, Bible, SMM speeches (official English / tparents.org), Transcripts. Grammar table lives in CONTEXT.md. Adding a new class requires its own ADR.

SMM speeches are English-only for now; if Korean originals are added later, citations gain a language tag (SMM-EN vs SMM-KO) to disambiguate translation variance.

Alternatives considered

• Unified citation form (e.g. always [[file#section|section]]): rejected — forces Bible’s universally-known John 3:16 into something unreadable.
• Free-form citations: rejected — this IS the drift the user named.

Consequences

• (+) Citations stay deterministic, AI can be linted against the grammar
• (+) Each class reads naturally in its native form
• (−) Adding new corpus types is a small ceremony, not zero-cost