Motivation

• Remove placeholder and mock logic in the phoneme/grammar pipeline to make linguistic analysis production-grade and deterministic.
• Improve syllable, stress, and rhyme accuracy by switching to grapheme-aware tokenization and explicit silent-e handling.
• Ensure generated sentences obey grammatical constraints by making verb transitivity and adverbs lexicon-driven rather than injected as mocks.
• Source the training/poetic corpus from real assets when available to avoid relying solely on embedded sample text.

Description

• Implemented a grapheme-aware phonemizer in src/UniversalLinguisticEngine.js by adding vowelTeams, consonantDigraphs, _phonemizePart, _isVowel, and silent-e handling to replace sequential regex-rewrite placeholders.
• Added deterministic lexicon selection and transitivity-aware VP branching via _selectEntry and updated getFlatGrammar to use Adv entries from the lexicon instead of a hardcoded list.
• Replaced the placeholder corpus loader in src/App.jsx with a Vite glob-based loader using import.meta.glob('/lyrics/*.txt', { as: 'raw' }), including file load error handling, deduplication, and fallback to the embedded corpus.
• Documented the implementation rationale and expanded test coverage instructions by updating LOGIC-MAP.md, WE-CHOSE.md, and TESTING.md with logic chains and a non-mock test matrix.

Testing

• No automated test suites were executed as part of this change; the edits were implemented and committed without running CI tests.
• TESTING.md was updated with concrete, non-mock test cases for analyze() (e.g., analyze('cake'), analyze('rain'), analyze('shadow')) and generation edge cases for transitive vs intransitive verbs.
• The corpus loader test cases validate file detection, deduplication, skipping empty lines, and fallback behavior to the embedded corpus when /lyrics/*.txt resolves to nothing.
• Manual validation through the UI controls is recommended to exercise sentence generation and corpus loading scenarios described in TESTING.md.

Codex Task