Release 0.6.0

What's Changed

Version bump: minor
Previous version: 0.5.2
New version: 0.6.0

56 commits since v0.5.2

🚀 Features

• feat(office): route image-dominant docx/pptx through the PDF image strategy (885a427) by @luisleo526
• feat(ocr): meaningful Markdown for image-heavy pages (page_markdown synthesis) + eval harness (6d8ec76) by @luisleo526
• feat(ocr): stronger table HTML guidance (full grid + colspan/rowspan) (931888c) by @luisleo526
• feat(ocr): nested OCRPage schema (figures, hierarchy, knowledge graph) (27fc1a0) by @luisleo526
• feat(ocr): richer structured OCRPage schema (raw indexes + interpretation anchors) (56c70c1) by @luisleo526
• feat(pdf): two-signal document route (coverage + text density), drop LLM classifier (de27455) by @luisleo526
• feat(pdf): document-level OCR strategy route by mean image-occupancy (928f1eb) by @luisleo526
• feat(ocr): enable + fix OpenAI inline-PDF context part (spike-verified) (863ad58) by @luisleo526
• feat(ocr): self-routing image job-router (Phase 1) (3f8778c) by @luisleo526
• feat(ocr): neighbor-page PDF context to improve image OCR (279075b) by @luisleo526
• feat(ocr): page-level OCR for image-dominant PDFs + skip decorative images (533e2b8) by @luisleo526
• feat(ocr): structured table extraction with merged cells (HTML colspan/rowspan) (560ff81) by @luisleo526
• feat(ocr): optional pre-OCR image downscale to cut Vision token cost (05661f2) by @luisleo526
• feat(chunker): optional token-aware chunking (tiktoken) (a816136) by @luisleo526
• feat(loader): opt-in cross-document batch parallelism + progress callbacks (d8bc161) by @luisleo526
• feat(formats): .eml email ingestion via stdlib email parser (7bf5d62) by @luisleo526
• feat(ocr): OCR facade, public exports, CLI flags, loader plumbing (8f64d20) by @luisleo526
• feat(ocr): structured output in OpenAI, Vertex/Gemini, Tesseract + cache v4 (324afd7) by @luisleo526
• feat(ocr): structured-output foundation (schema + extended base contract) (70abac3) by @luisleo526

🐛 Bug Fixes

• fix(ocr): normalize ragged OCR tables to a rectangular grid (0b34f5c) by @luisleo526
• fix(pdf): route designed PDFs with an illegible text layer to image OCR (0c7d80a) by @luisleo526
• fix(ocr): whitespace-tolerant firewall verbatim checks (9090324) by @luisleo526
• fix(export): emit OCR text clean (drop code-fence/ocr_result wrapper) (c756d3a) by @luisleo526
• fix(pdf): image-occupancy strategy route + clean OCR export (2389caf) by @luisleo526
• fix(ocr): dense structured OCR truncation + batch failure isolation (5529b65) by @luisleo526
• fix(pdf): never dump raw base64 on OCR batch failure (RAG-safety) (cfe4436) by @luisleo526
• fix(ocr): recover empty structured results via free-form fallback (4ab1be2) by @luisleo526
• fix(legacy): preserve metadata.extra identity with is-None guard (9264490) by @luisleo526
• fix(release+api): repair release metadata, type public API, OCR resilience (7775cc7) by @luisleo526

♻️ Refactoring

• refactor: dead-code sweep + orphaned-import cleanup (Phase 6) (22ef728) by @luisleo526
• refactor(core): share the SimpleContent model; drop dead debug probes (06fc364) by @luisleo526
• refactor(ocr): hoist shared empty-structured detection + recovery merge to BaseOCR (b5d9268) by @luisleo526
• refactor(utils): extract shared LibreOffice converter (b3c55e4) by @luisleo526
• refactor(core): extract shared OCR strategy decision to core/strategy.py (9fa25a3) by @luisleo526
• refactor: remove dead UnifiedProcessor stack + duplicate LibreOffice converter (bf280ff) by @luisleo526

📚 Documentation

• docs(readme): rewrite as a pitch with markitdown/Docling comparison (833271f) by @luisleo526
• docs: fix neighbor-context provider claim in ocr_policy.rst (f749dac) by @luisleo526
• docs: enrich + resync documentation with the current codebase (76954ee) by @luisleo526
• docs: changelog for document-level OCR strategy route (438ee65) by @luisleo526
• docs: full hand-written API reference (Sphinx, strict build clean) (67ab589) by @luisleo526
• docs(readme): document Batch-5 features (.eml, batch parallelism, token chunking, image downscale) (74e1546) by @luisleo526
• docs(changelog): record audit fixes, dead-code removal, CI, and new features (4f268fe) by @luisleo526
• docs(ocr): document structured-output redesign (facade, schema, tasks) (1a84f6c) by @luisleo526

🔧 Chores

• test(eval): average the meaningfulness judge over 3 samples for stability (6078c6f) by @luisleo526
• test(eval): register office image fixture + sharpen fragmentation judge (fac80ed) by @luisleo526
• test: characterization suite for pipelines, legacy, structured OCR, edge cases (08ca726) by @luisleo526
• ci: enforce coverage + wire lint/type/security gates (be10907) by @luisleo526

📝 Other Changes

• Merge remote-tracking branch 'origin/main' into improve/audit-fixes (8cd438a) by @luisleo526
• change(ocr): default OpenAI model gpt-4.1 -> gpt-5.4-mini (1039943) by @luisleo526
• security(ocr): sanitize LLM-produced Table.html (XSS/HTML-injection sink) (011ce55) by @luisleo526
• docs+ci+tests: document OCR cache/chunking/extras, CI hygiene, fixtures (cf00303) by @luisleo526

Full Changelog

v0.5.2...v0.6.0