kokorog2p migration boundary

spokenform can serve a downstream adapter for one selected language run. Call prepare_for_kokorog2p() (or pass PreparationConfig.for_kokorog2p()), preserve caller override spans with ProtectedSpan, use PreparedText.source_edits and the public span helpers to remap boundaries, and inspect warnings before passing the result to a G2P tokenizer. The adapter projection is available through PreparedText.to_adapter_dict().

The English, German, French, Spanish, Italian, Portuguese, and Czech downstream migration contracts are tested at three boundaries: spoken text, source replacement/offset provenance, and a downstream-style token/phoneme fixture. spokenform does not own tokenization, lexicon lookup, phonemization, quote/dash typography, or model punctuation. A downstream adapter should run a dual comparison before removing its legacy normalizer.

Lexical abbreviations and numeric symbol recognition come from abbr2words; spokenform supplies locale-owned semantic realization and exact source replacements. Do not delete a downstream normalizer until both paths have been compared for prepared text, source replacements, extended token positions, phonemes, warnings, and protected overrides. A real downstream gate is provided in tests/test_real_kokorog2p_integration.py and is run in CI with released kokorog2p[de,fr] packages; SpanishG2P and ItalianG2P are part of the base package and are exercised by the same gate.

Ownership audit

Language	Suitable for spokenform	Keep downstream	Status
cs	reviewed dates, ordinary numbers, quantities, temperatures, currencies, canonical units	G2P/lexicon behavior; colon times	parity-gated; `STRUCTURED_AND_PLAIN`; time caller-managed
en	dates, validated clock times, currencies, reviewed quantities, safe ordinary written numbers, contextual single-dot release labels	phoneme-sensitive years, suffix ordinals, Roman numerals, phone/ID and arbitrary multi-dot sequences, numeric suffixes, G2P decisions	active adapter / parity-gated; downstream-only categories remain explicit
es	reviewed dates, ordinary numbers, currencies, units, temperatures	G2P/tokenizer typography; time expressions	parity-gated; `STRUCTURED_AND_PLAIN`; time caller-managed
fr	dates, times, numbers, ordinals, currencies, temperatures, units, exact maps	G2P/tokenizer typography, lexicon, phonemes	parity-gated; `STRUCTURED_AND_PLAIN`
it	reviewed dates, ordinary numbers, currencies, units, temperatures	G2P/tokenizer typography; colon-time ownership	parity-gated; `STRUCTURED_AND_PLAIN`; colon times caller-managed
pt	reviewed dates, ordinary numbers, currencies, units, temperatures	G2P/tokenizer typography; colon-time expressions	parity-gated; `STRUCTURED_AND_PLAIN`; time caller-managed

English’s safe plain-number pass intentionally handles only ordinary short or grouped cardinals and exact decimal digits. A separate reviewed structured rule handles contextual single-dot release labels such as bot 2.0 and renders the fractional zero as oh; it does not change ordinary decimal wording. Quantity matches take precedence over that contextual rule. Four-digit and longer ungrouped digit strings remain raw so years, identifiers, and sequence-like values reach kokorog2p’s NumberConverter and related heuristics. A structured candidate with unsupported fractional currency precision also remains unchanged.

English is active on the kokorog2p spokenform adapter for reviewed structured semantics, contextual single-dot release labels, and safe ordinary-number categories. Phoneme-sensitive years, suffix ordinals, Roman numerals, phone/ID and arbitrary multi-dot sequences, numeric suffixes, and G2P decisions remain downstream in kokorog2p; the adapter does not claim those categories.

This audit intentionally does not port language detection, markup parsing, mixed language orchestration, lexicon lookup, phoneme suffix rules, token IDs, or model specific quote/dash behavior into spokenform. French, Spanish, Italian, and Portuguese are ready for downstream handoff only with the released abbr2words>=0.2.4 prerequisite and their real parity gates; package publication remains the release workflow boundary. Spanish, Italian, Portuguese, and Czech time ownership is intentionally deferred until reviewed time corpora exist. English semantic number categories are available in the direct spokenform API, while years, ordinals, Roman numerals, phone/ID and dotted sequences, numeric suffixes, and G2P decisions remain downstream-owned.

Preferred adapter surface

Downstream integrations should depend on the stable high-level surface: PreparationConfig.for_kokorog2p(language), prepare_for_kokorog2p(), PreparedText.source_replacements, PreparedText.offset_map, and NumberPolicy. Low-level mapping and stage helpers remain exported for advanced use but are not required for a normal kokorog2p adapter.