v1.0.0

Changelog

Other

• 13c7b93: Absorb skein tooling & process; add state-of-project review (H0) (@budhash)
• 02d2f0a: Achieve 90.1% accuracy with फ→f rule and expanded overrides (#37) (@budhash)
• 4d39d87: Add --keep-medial-schwa flag for conservative schwa retention (@budhash)
• 7bb87a8: Add --long-vowels flag and scalable options system (#17) (@budhash)
• 329d963: Add --long-vowels support to new engine (#25) (@budhash)
• e14c2e1: Add --simple-nasals flag for simplified nasal endings (@budhash)
• 89790a8: Add CLI batch processing and test validation modes (@budhash)
• e3a08cb: Add Conditional field to rules for option-dependent status display (@budhash)
• 00ef9c7: Add English loanwords to ignore list and datasets documentation (#31) (@budhash)
• b199eff: Add declarative rule system for schwa deletion (Phase 2) (#23) (@budhash)
• bf9a682: Add layered architecture spec, tasks, and proof of concept (#26) (@budhash)
• dac6797: Add multi-pass romanization engine (Phase 1) (#22) (@budhash)
• 3675e9f: Add phonetic rules and fix nukta consonants for 85.6% accuracy (@budhash)
• 0355fa4: Add romanization engine architecture proposal (#20) (@budhash)
• 7b78ea3: Add rule override system with CLI flags and unit tests (#32) (@budhash)
• 5fb4771: Add schwa deletion for consonant-ending CCCC words (#19) (@budhash)
• 5a7af17: Add schwa.delete.before-cc rule for compound word boundaries (@budhash)
• 5b99af3: Clarify scheme design in architecture proposal (#21) (@budhash)
• 3e31d50: Docs sweep: make every non-README doc truthful and current (@budhash)
• 00b378a: Documentation architecture: RESEARCH.md + DESIGN.md; panel-reviewed docs (@budhash)
• c9fa774: Final hardening: Cf handling completed, CLI/API tests added, coverage measured (@budhash)
• d3ab012: Fix schwa deletion patterns and add UnitModifier type (#24) (@budhash)
• 998419f: Fix ज्ञ conjunct and ीय adjective suffix patterns (#18) (@budhash)
• 3ffad93: H1: honest multi-reference evaluation + CER; gate on pure accuracy (@budhash)
• f73abbb: H2: vowel-length rules rejected on evidence; add lang/hindi + brahmic tests (@budhash)
• 5331bc5: H3: high-confidence lexicon layer (opt-in, honest coverage) (@budhash)
• 04618de: H3: learned schwa classifier distilled to embedded Go (opt-in) (@budhash)
• ba697cd: Migrate to layered core architecture with --debug support (#27) (@budhash)
• db18ea2: P3: override-mining tool (negative result) + char-LM re-ranker (best system yet) (@budhash)
• c98e199: Phase 2: Add nasalization rules and legacy cleanup (@budhash)
• b2142d6: Phase 2: Fix remaining patterns and restructure benchmarks (87.3%) (#30) (@budhash)
• b7b6caf: Pre-release fixes from the final three-panel repo review (@budhash)
• ac66404: README truth pass for v1.0.0 (release not yet tagged) (@budhash)
• 84eb873: Repo reorg: remove dead weight, consolidate tooling and historical docs (@budhash)
• 3342e5c: T-0007: add Aksharantar human-annotated test set with per-slice benchmark (@budhash)
• 8014abf: T-0020: frequency-ranked lexicon expansion (29.4% → 71.1% token coverage) (@budhash)
• d75ccfb: T-0021: public-domain lyrics gold seed + line-level harness; punctuation fix (@budhash)
• d825e4e: T-0022: COMI-LINGUA colloquial benchmark (how people actually type Hindi) (@budhash)
• 089d3ad: T-0023: segment words on all whitespace, preserving it verbatim (@budhash)
• e83dc5e: Track A: rule hardening — source-char matching + extract shared Brahmic rules (@budhash)
• 065d877: Track C: frequency-weighted real-world evaluation (quantifies lexicon value) (@budhash)
• 8f25f65: Update docs: accuracy 85.5%, new CLI options (@budhash)