Skip to content

ADR-v2-075 — Phonetic Spelling Mode

Status: Accepted (2026-07-02) · Wave J Context links: [[adr-v2-048-entity-itn]] (normalizes prose vs character-exact), [[adr-v2-055-emoji-symbol-by-voice]], [[adr-011]]

Context

Wave J research (#2) — a deliberate character-exact mode where NATO words map to letters ("capital alpha, bravo, double lima" → Abll), with digits and symbols — the one case free dictation always mangles: passwords, codes, IDs, filenames. Distinct from Entity ITN / Emoji-by-Voice (those normalize prose); this is a mode that spells. A large motor/low-vision accessibility win with a trivial pure core. Anchor: Picovoice on-device NATO-spelling engine (user-value, not novelty).

Decision

Add an opt-in Phonetic Spelling Mode: [spelling] enabled=false. The pure core spell_parse(text) maps NATO words → letters with capital/cap/upper (and lowercase) modifiers and double/triple repeaters, plus spoken digits and symbols (dash, dot, underscore, at, hash, slash, space, …). Dependency-free. An optional constrained/biased decode for the spelling burst is deferred. OFF by default.

Consequences

  • Character-exact capture for the one thing prose dictation can't do; pure dict + modifier grammar.
  • Distinct from ITN/Emoji (spells vs normalizes).
  • Privacy (ADR-011): pure string mapping, fully offline.
  • Caveat: covers the NATO set + common symbols → exotic glyphs need the deferred biased decode; off by default.