Changelog

Added 13 foundation language families with reviewed plain-cardinal rendering and explicit capability boundaries

Improved Spokenform Gold benchmark validation with release provenance and complete-corpus checks

Changed dependency floors to abbr2words 0.2.16 and numeralform 0.1.9 with matching CI coverage

Fixed Spanish ordinal normalization to fail closed outside the reviewed range

Added deterministic Spokenform Gold benchmark consumption with immutable release-asset and runtime-source verification

Added contextual Spanish month/day date recognition without a year

Improved German comma-separated number and short social-identifier rendering

Fixed French coordinate recognition for dot and comma decimal separators while preserving fail-closed quantity parsing

Fixed URL-like literals being treated as math and made unsupported math locales fail closed

Fixed malformed locale-mismatched currency amounts being rewritten while preserving valid currency normalization

Improved German speech normalization for technical sequences, mathematical expressions, contextual Roman forms, durations, and semantic versions

Added opt-in German URL, e-mail, and FTP literal promotion with localized punctuation while preserving default literal protection

Added reviewed German source-letter expansion for IBAN through the dependency registry

Fixed German IBAN precedence and sentence-final IPv4 recognition without weakening literal protection

Added locale-aware currency and exchange-rate speech rendering for English, German, and Spanish

Improved locale-aware numeric, currency, identifier, emergency-number, and German reference normalization

BREAKING: Changed numeric runtime dependency from upstream num2words to Numeralform 0.1.4

BREAKING: Changed the public numeric locale resolver and documented independent routing

Improved release hardening for the Numeralform 0.1.4 dependency boundary

Added registry-backed language and locale support with capability-aware numeric and sequence policies

Added explicit language-selected speech preparation with source-aligned provenance and generic speech configuration

Added source-aligned semantic normalization coverage for migrated locales and abbreviation behavior

Added immutable isolated speech profiles for domain-specific abbreviation pronunciation

Improved German contextual year and Euro minor-unit speech with identifier-boundary protection

Added typed German plural paragraph references and ranges with domain-aware precedence

Added Korean native counter forms for quantities and clock hours

Added source-aligned Thai preparation for times, decimals, identifiers, and semantic symbols

Added conservative Arabic, Hebrew, and Kazakh KokoroG2P profiles with aliases

Added public abbreviation controls and KokoroG2P profile metadata

Improved currency speech to preserve excess fractional precision instead of silently rounding or truncating

Fixed German time and date normalization to reject partial-word matches and invalid calendar values

Documented German structured normalization boundaries and exact currency precision

Added regression coverage for German boundaries, provenance, protection, and false positives

Added Swedish runtime support for locale-aware numbers, reviewed quantities, temperatures, and SEK amounts

Added Vietnamese runtime support for locale-aware numbers, reviewed quantities, temperatures, and VND amounts

Added Thai runtime support for locale-aware numbers, reviewed quantities, temperatures, and THB amounts

Added typed German plural paragraph-list and range recognition with precedence and domain controls

Added conservative contextual German year pronunciation using the shared year renderer

Changed the abbr2words dependency floor to the release containing the reviewed German lexical registry

Prevented German time and date normalization from claiming partial word or context matches

Preserved excess fractional digits in currency speech instead of silently truncating or rounding

Fixed German currency realization to label reviewed minor units explicitly

Fixed German date and time recognition to reject partial matches inside alphanumeric identifiers

Documented Swedish identifiers, numeric punctuation, ownership boundaries, and fail-closed semantics

Documented Vietnamese identifiers, numeric punctuation, ownership boundaries, and fail-closed semantics

Documented Thai identifiers, numeric punctuation and digits, calendar boundaries, ownership, and fail-closed semantics

Documented German structured normalization and source-faithful currency precision

Added Swedish regression coverage for precision, source mapping, protection, and sequence safety

Added Vietnamese regression coverage for precision, grouping, mapping, protection, and sequence safety

Added Thai regression coverage for decimal precision, Thai digits, quantities, THB, calendar protection, mapping, and sequence safety

Added German regression coverage for structured boundaries, provenance, protection, and false positives

Added Japanese, Korean, and Chinese runtime support with locale-aware numbers and conservative sequences

Changed abbreviation expansion to consume source-aligned replacement records and preserve dependency metadata

Changed the minimum abbr2words dependency to 0.2.11 and added minimum-dependency CI checks

Changed preparation and structured candidate processing into focused stages while preserving supported behavior

Improved Spokenform Gold evaluation for the v2 corpus with integrity checks and provenance-rich reports

Changed optional Lexhint compatibility to cover 0.2.x while retaining 0.1.2 support

Documented Lexhint 0.2.x schema-8 dataset requirements and the runtime dataset upgrade path

Added an explicit CI compatibility gate for the supported Lexhint version families

Added opt-in conservative orthographic fallback for residual sequences

Added diagnostics for fallback decisions and structured policy traces

Changed contextual countdowns to retain a generic spaced segment boundary

Changed countdown ownership to temporal and added fail-closed domain allowlisting

Changed sequence and code segment rendering to use one generic boundary

Changed benchmark oracle coverage to isolate policy expansion results

Documented recognition evidence, domain ownership, fallback, and punctuation responsibilities

Added multilingual mapping coverage, recognizer extraction, and release validation evidence

Added Async TN benchmark suite with deterministic evaluation, data management, and HTML reporting

Added candidate oracle for measuring structured-candidate selection headroom across benchmarks

Added extended currency symbol recognition for Spanish with labeled amounts and feminine agreement

Added hyphenated date recognition for Spanish and Italian (e.g., 15-julio-2026, 3-marzo-25)

Added Spanish time period recognition for bare hour-period patterns (e.g., 3 pm)

Added Italian dotted-time ambiguity guard to prevent false time recognition on decimals and units

Added failure reporting module with risk tiers, ownership mapping, and diagnostic aggregates

Improved Spanish ordinal rendering with feminine compound forms, quince special case, and duodécimo for twelve

Improved Spanish fraction rendering with natural forms for halves (media, medio) and feminine agreement

Improved English fraction rendering to use fourths instead of quarters

Improved French and Italian slash-fraction rendering with locale-specific natural forms (moitié, metà, centième)

Improved sports and score recognition with broader multilingual context keywords and higher value ranges

Expanded software version recognition to include macOS and WordPress

Improved ISBN rendering for Spanish, Italian, and French with locale-specific separator words

Refactored structured module to separate candidate collection from conflict resolution for reuse

Added sentence-start capitalization for Spanish numeric output at input boundaries

Improved Spanish phone number rendering with cardinal group mode and comma-separated groups

Improved number rendering with digitwise fraction mode for plain numbers and explicit leading-decimal sign handling

Improved Italian ordinal rendering with feminine elision for ottava after la

Added context-conditional number rendering for French identifier codes and Spanish page quantities

Improved Spanish Mexican time rendering with de la noche suffix for evening hours and auto-detected clock periods

Expanded version context recognition to support is/ist/est/es copula patterns and software-version label

Fixed plural initialisms such as IDs, PCs, and ICs being misclassified as chemical formulas

Fixed French and Italian two-digit year date rendering to use two-digit expansion

Fixed Spanish currency grouping to accept comma-grouped decimals with three-digit tails

Fixed coordinate sequence priority and phone sequence priority to use correct precedence levels

Restricted bare-dot ordinal detection to German to prevent false ordinals in other locales

Fixed English date context detection to exclude pieces and diluted as non-date right-context

Added explicit coverage, documentation, artifact, wheel, and downstream release gates

Improved Polynorm and Proteno benchmark evaluation with numeric lexeme and sequence coverage

Added Russian runtime support for locale-aware numbers and reviewed measurement quantities

Fixed direct prepare() keyword validation to match PreparationConfig

Removed stale package manifest metadata and added distribution-content checks

Fixed plural initialism shapes such as IDs being misclassified as chemical formulas

Fixed explicit English leading-decimal signs and digit-wise precision for readings such as .3 and .02

Documented the abbr2words 0.2.9 release boundary and configuration policy modes

Documented Russian identifiers, morphology boundaries, dependency ownership, and caller-managed currency

Added explicit coverage, documentation, artifact, wheel, and downstream release gates

Improved maintainability with shared date validation and dead-code cleanup

Added Russian regression coverage for grammar boundaries, protections, mapping, and fail-closed domains

Added conservative acronym mode for context-aware initialism handling

Added contextual long-number mode for intelligent verbalization of 4+ digit numbers

Added registered acronym mode for controlling registered initialism expansion

Added bibliographic citation and reference sequence recognizers

Added decade rendering with natural English ordinal decade forms

Added duration recognition for HH:MM:SS timestamp formats

Added score recognition for sports results with contextual plausibility checks

Added chained score recognition for multi-set or multi-game results

Added software version recognition for named software packages

Added spaced ISBN label recognition for OCR or speech-recognized text

Added German Roman numeral title rendering for monarchs and popes

Added parenthesized initialism and ticker symbol recognition

Added superscript and Greek letter symbol rendering

Added additional Unicode fraction characters for fifths and sixths

Added compact license plate and vehicle code recognition

Added Google TN benchmark for English text normalization evaluation

Added benchmark ownership table with safety, extended, and quarantine gates

Added benchmark comparison identity validation and fresh report enforcement

Improved year detection with expanded temporal keyword context and bibliographic patterns

Improved formula recognition with balanced parenthesis context checking

Improved product code recognition with stronger label evidence requirements

Improved Roman numeral recognition with expanded context keywords for articles, acts, and scenes

Changed structured semantic precedence from regex iteration order to centralized precedence module

Fixed fraction rendering to use natural word forms for fifths and sixths instead of Unicode fallback

Fixed formula recognition to reject fragments from unmatched parenthesized spans

Fixed slash fraction rendering to skip incomplete numerator or denominator matches

Fixed identifier separator handling to accept empty marker values

Fixed version literal detection to skip bare v-prefixed versions without contextual evidence

Fixed biology recognition to exclude temporal prepositions as false species markers

Fixed code recognition to classify vehicle-shaped codes with appropriate digit policies

Added Proteno diagnostic aggregates by rule, phase, ownership, and ambiguity family

Added PolyNorm ownership table documentation for comparison compatibility and quarantine policy

Added Proteno benchmark suite with WER evaluation and data utilities

Added speech WER threshold configuration for benchmark evaluation

Added symbol handling parameters for configurable acronym and symbol processing

Added improved English date rendering with source-order preservation and decade support

Added Binder example with interactive Jupyter notebook

Improved number detection in sequences and structured text normalization

Added contextual English single-dot release-label pronunciation while preserving ordinary decimal precision and multi-dot protection

Fixed pre-commit formatting and linting issues

Added opt-in URL, e-mail, version, and contextual Roman rendering with caller protection

Added locale-aware rendering for sequences, fractions, numeric values, and identifiers

Documented ownership boundaries, literal policy, benchmark diagnostics, and dependency requirements

Added precedence, no-false-claim, locale, source-mapping, specialist, and benchmark-equivalence diagnostics

Fixed German extended quantity realization for square, cubic, hectare, and speed units with source-aligned mappings

Documented active English kokorog2p semantic ownership and downstream G2P boundaries

Added a real run-level German kokorog2p gate for extended quantity parity and protected-span behavior

Added parity-gated Spanish semantic normalization for reviewed numbers, units, temperatures, currencies, and numeric dates

Added Spanish parity and real kokorog2p es/la integration coverage with protected coordinate checks

Added parity-gated Italian semantic normalization for reviewed numbers, units, temperatures, currencies, and numeric dates

Added Italian parity and real kokorog2p integration coverage with protected coordinate checks

Added English locale with structured text normalization for dates, times, numbers, ordinals, quantities, temperatures, and currencies

Added Portuguese locale with dates, times, numbers, ordinals, quantities, temperatures, and currencies

Added Czech locale with dates, times, numbers, ordinals, quantities, temperatures, and currencies

Promoted Spanish kokorog2p preparation to structured and plain number ownership while keeping time expressions caller-managed

Promoted Italian kokorog2p preparation to structured and plain number ownership while keeping colon times caller-managed

Documented Spanish migration ownership, the released abbr2words prerequisite, and downstream release readiness

Documented Italian migration ownership, the released abbr2words prerequisite, and downstream release readiness

Expanded API with structured normalization pipeline, text protection improvements, and comprehensive packaging and dependency tests

Added French locale with dates, times, numbers, ordinals, quantities, temperatures, and currencies

Added French kokorog2p parity corpus and released downstream integration gate

Refactored structured text normalization to delegate locale-specific grammar to per-locale modules

Promoted French kokorog2p number ownership to structured and plain after parity coverage

Updated limitations documentation with French language support details

Added source replacement composition and abbr2words conversion helpers for offset mapping

Added German locale module with quantity grammar for units, currencies, and magnitudes

Added German structured text normalization for dates, times, currencies, temperatures, quantities, ordinals, and labels

Added kokorog2p integration with prepare_for_kokorog2p() adapter projection API

Added locale registry with stub modules for cs, en, es, fr, it, and pt

Fixed pytest configuration for proper test execution

Fixed tomli dependency configuration for package installation

Added kokorog2p migration guide with language ownership audit and integration boundaries

Added core spoken form normalization with number verbalization, token annotations, and CLI

Added offset mapping, text protection, and spaCy model integration

Added structured text normalization for dates, times, currencies, and measurements with German locale support

Added configurable whitespace normalization policies and benchmark suite

Removed language detection and SSML features to simplify the API

Fixed tomli dependency configuration for package installation

Fixed pytest configuration for proper test execution

Added API reference, architecture guide, CLI docs, installation guide, and usage examples

Added CI workflow updates and pre-commit configuration

Added changelog documentation updates