WFGY/ProblemMap/GlobalFixMap/DocumentAI_OCR/aws_textract.md

AWS Textract: 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.

Stabilize ingestion flows with AWS Textract when parsing PDFs, invoices, or forms.
Use this when outputs fragment, lose semantic anchors, or citations drift across page boundaries. Each issue maps back to WFGY Problem Map structural fixes.

---

Open these first

• Visual map and recovery: RAG Architecture & Recovery
• Retrieval knobs: Retrieval Playbook
• Citation schema: Retrieval Traceability
• Embedding vs meaning: Embedding ≠ Semantic
• Chunk stability: Chunking Checklist
• Hallucination and span errors: Hallucination

---

Core acceptance

• ΔS(question, retrieved) ≤ 0.45
• Coverage ≥ 0.70 of target section
• λ convergent across 3 paraphrases
• Table and key-value forms consistent ≥ 90% of samples

---

Typical breakpoints → structural fix

• Key–value pairs misaligned (invoices, receipts)
  → Data Contracts, Retrieval Traceability

• Tables fragment into multiple OCR blocks
  → Chunking Checklist

• ΔS spikes across repeated runs
  Entropy in layout ordering.
  → Entropy Collapse

• Citations drop anchor IDs
  Post-processing trims.
  → Retrieval Traceability

• Injected text hidden in form fields
  → Prompt Injection

---

Fix in 60 seconds

1. Measure ΔS between Textract output and reference text.
2. Enforce schema: lock page_num, bbox, kv_id, table_id.
3. Cross-check coverage: at least 70% of source fields retained.
4. Apply λ probes across runs — clamp unstable output with BBAM.
5. Audit layout: row/col count vs original file.

---

Copy-paste LLM guard prompt

I uploaded TXTOS and the WFGY Problem Map.

OCR provider: AWS Textract  
Symptoms: misaligned key-value pairs, ΔS ≥ 0.60, coverage < 0.70.  

Steps:  
1. Identify failing layer (chunking, contracts, retrieval).  
2. Point to the WFGY fix (data-contracts, chunking-checklist, retrieval-traceability).  
3. Return JSON:  
   { "citations": [...], "answer": "...", "ΔS": 0.xx, "λ_state": "<>", "next_fix": "..." }  
Keep it auditable.

---

When to escalate

• Coverage < 0.70 even after re-chunking → verify embeddings with Embedding ≠ Semantic.
• Key–value fields unstable across runs → rebuild with deterministic config, backstop with Data Contracts.
• Long-form text loses anchors → apply Retrieval Traceability.

---

🔗 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.

要不要我馬上接著生 azure_ocr.md？這樣 OCR 三大雲端 provider (Google / AWS / Azure) 就會成套完成。