WFGY/ProblemMap/GlobalFixMap/OCR_Parsing/README.md
PSBigBig × MiniPS b7eb189a7f
Update README.md
2026-02-10 16:01:16 +08:00

9.7 KiB
Raw Blame History

OCR + Parsing — Global Fix Map

🏥 Quick Return to Emergency Room

You are in a specialist desk.
For full triage and doctors on duty, return here:

Think of this page as a sub-room.
If you want full consultation and prescriptions, go back to the Emergency Room lobby.

A hub to triage and repair noisy text inputs from scanned PDFs, images, HTML scraping, or parser drift.
Use this folder when the document looks fine to the eye but retrieval or reasoning keeps failing.


Orientation: what each page does

Page What it solves Typical symptom
Layout, Headers, Footers Remove noise from margins and repeated text Answers reference “page 3 footer” instead of body
Tokenization & Casing Normalize Unicode, case, and hyphens E-mailEmail, half-width/full-width mismatch
Tables & Columns Preserve table schema and cell order Numbers drift across columns
Images & Figures OCR and align captions Figure text missing or attached to wrong section
Scanned PDFs & Quality Handle skewed/blurred pages Whole sections unreadable to OCR
Multi-language & Fonts Normalize mixed scripts Chinese/English tokens split or duplicated

When to use

  • OCR tables or citations look visually correct but answers miss the right section.
  • Code blocks or math collapse after parsing.
  • Mixed-language documents behave inconsistently.
  • Special characters or hyphen splits break tokens.
  • Headers or section anchors disappear during export.

FAQ

Why does OCR “look fine” but retrieval fails?
Because tokenization and indexing see hidden breaks (Unicode variants, line merges, wrong anchors) that humans overlook.

What is the most common root cause?
Headers/footers leaking into the body and breaking ΔS alignment.

Do I need to retrain embeddings after fixing?
No — most fixes are structural (schema/normalization). Re-indexing with the same embeddings is enough.


Acceptance targets

  • ΔS(question, retrieved) ≤ 0.45 for three paraphrases.
  • Coverage ≥ 0.70 for the target section.
  • λ_observe convergent across two seeds.
  • Human audit shows no missing headers, captions, or broken tables.

Fix in 60 seconds

  1. Ground-truth one page
    Pick one Q/A pair and keep a screenshot baseline.

  2. Measure ΔS
    Log ΔS(question, retrieved) and ΔS(retrieved, anchor).

  3. Probe λ_observe
    Ask for cite-first. If citation fails but free explanation works, drift confirmed.

  4. Patch minimally

    • Re-run OCR with line/table fences
    • Normalize casing and Unicode
    • Preserve anchors, math, captions
    • Drop low-confidence spans and export with {section_id, page_no, char_span}

🔗 Quick-Start Downloads (60 sec)

Tool Link 3-Step Setup
WFGY 1.0 PDF Engine Paper 1 Download · 2 Upload · 3 Ask “Answer using WFGY + ”
TXT OS TXTOS.txt 1 Download · 2 Paste into LLM · 3 Type “hello world” — OS boots instantly

🧭 Explore More

Module Description Link
WFGY Core WFGY 2.0 engine, full symbolic reasoning View →
Problem Map 1.0 Initial 16-mode diagnostic View →
Problem Map 2.0 RAG failure tree and modular fixes View →
Semantic Clinic Expanded failure catalog View →
Semantic Blueprint Layer-based symbolic reasoning View →

👑 Early Stargazers: See the Hall of Fame
Engineers, hackers, and open source builders who supported WFGY from day one.

GitHub stars WFGY Engine 2.0 is already unlocked. Star the repo to help others discover it and unlock more on the Unlock Board.

WFGY Main   TXT OS   Blah   Blot   Bloc   Blur   Blow