lllin000_PaperForge/docs/superpowers/specs/README-ocr.md
2026-06-09 12:55:05 +08:00

1.3 KiB

OCR Design Index

Use this file as the quick index for current OCR design docs.

Current OCR architecture source of truth

  • 2026-06-08-ocr-anchor-first-structured-parsing-design.md
    • Current clean target architecture for OCR parsing.
    • Covers anchors, zones, reference-first tail parsing, family partition, and figure/table validation.

Current OCR implementation plan

  • ../plans/2026-06-08-ocr-anchor-first-structured-parsing-plan.md
    • Task-by-task implementation sequence for the anchor-first redesign.
  • 2026-06-01-ocr-redo-single-source-design.md

    • Separate approved design for OCR redo and canonical single-source fulltext behavior.
  • ../plans/2026-06-01-ocr-redo-single-source.md

    • Implementation plan for the single-source redo workflow.

Notes

  • This worktree keeps only the current OCR v2 architecture/design baseline plus the still-relevant single-source redo docs.
  • Older OCR reading-order, structured-pipeline phase plans, score-tuning plans, and convergence-only intermediate OCR drafts were removed after they were superseded by the anchor-first design.
  • 2026-05-31-ocr-skip-and-tags-preserve-design.md and its paired plan remain because they are mixed OCR/plugin frontmatter behavior work, not replaced by the anchor-first parsing redesign.