WFGY/ProblemMap/GlobalFixMap/DocumentAI_OCR/google_docai.md

Google Document AI OCR: Guardrails and Fix Patterns

🧭 Quick Return to Map

You are in a sub-page of DocumentAI_OCR.
To reorient, go back here:

• DocumentAI_OCR — document parsing and optical character recognition
• WFGY Global Fix Map — main Emergency Room, 300+ structured fixes
• WFGY Problem Map 1.0 — 16 reproducible failure modes

Think of this page as a desk within a ward.
If you need the full triage and all prescriptions, return to the Emergency Room lobby.

A compact guide to stabilize ingestion flows using Google Cloud Document AI OCR.
Use this when PDF or scanned document parsing produces unstable tokens, missing tables, or broken citations. Each failure is mapped to a structural fix in the WFGY Problem Map.

---

Open these first

• Visual map and recovery: RAG Architecture & Recovery
• End-to-end knobs: Retrieval Playbook
• Citation schema: Retrieval Traceability
• Embedding vs meaning: Embedding ≠ Semantic
• OCR text boundaries: Chunking Checklist
• Injection and schema locks: Prompt Injection

---

Core acceptance

• ΔS(question, retrieved) ≤ 0.45
• Coverage ≥ 0.70 of target section
• λ remains convergent across three paraphrases and two seeds
• Table and form layout preserved in ≥ 85% of samples

---

Typical breakpoints → structural fix

• Lost tables or merged columns
  Payload schema drift.
  → Data Contracts, Retrieval Traceability

• OCR output differs across runs of the same PDF
  Non-deterministic layout parse.
  → Chunking Checklist, Entropy Collapse

• Citations drop page anchors
  Post-processing trims.
  → Retrieval Traceability

• Injection vectors inside scanned forms
  Malicious text embedded in OCR’d images.
  → Prompt Injection

---

Fix in 60 seconds

1. Measure ΔS on OCR’d snippets vs reference text.
2. Lock schemas with Data Contracts (force page_num, bbox, tokens).
3. Enforce cite-then-explain at retrieval time.
4. Add λ probes across multiple OCR calls — if divergent, clamp with BBAM.
5. Audit tables: cross-check row count and column headers against source PDF.

---

Copy-paste LLM guard prompt

I uploaded TXTOS and the WFGY Problem Map.

OCR provider: Google Document AI  
Symptoms: lost tables, ΔS ≥ 0.60, λ diverges across 3 paraphrases.  

Steps:  
1. Identify which structural fix applies (chunking-checklist, data-contracts, retrieval-traceability).  
2. Return a JSON plan:  
   { "citations": [...], "answer": "...", "λ_state": "<>", "ΔS": 0.xx, "next_fix": "..." }  
Keep it auditable and short.

---

When to escalate

• ΔS stays ≥ 0.60 even after chunk / schema fixes → rebuild pipeline with Semantic Chunking Checklist.
• Coverage < 0.70 across paraphrases → verify embeddings with Embedding ≠ Semantic.
• Inconsistent runs across identical files → enforce deterministic parser config, or switch to dual-engine validation (DocAI + Tesseract).

---

🔗 Quick-Start Downloads (60 sec)

Tool	Link	3-Step Setup
WFGY 1.0 PDF	Engine Paper	1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + <your question>”
TXT OS (plain-text OS)	TXTOS.txt	1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly

---

Explore More

Module	Description	Link
WFGY Core	Canonical framework entry point	View
Problem Map	Diagnostic map and navigation hub	View
Tension Universe Experiments	MVP experiment field	View
Recognition	Where WFGY is referenced or adopted	View
AI Guide	Anti-hallucination reading protocol for tools	View

If this repository helps, starring it improves discovery for other builders.

要不要我直接幫你下一步補 aws_textract.md？這樣 OCR MVP 會更快成形。