Skip to content

Source PDFs Index — Deep Reference Layer

Last updated: 2026-04-27 Tags: sources, PDFs, deep reference, citations

📖 Full Oracle Documentation: OAC Documentation Hub

Summary

The raw/pdfs/ folder contains the authoritative Oracle documentation in PDF + searchable text. When wiki pages don't have enough depth, I (Claude) grep these text files at query time and cite the result. This is the deep-reference layer that makes the knowledge graph comprehensive.

Available Source Documents

PDF Text Size Content Maps to Wiki
getting-started-oracle-analytics-cloud 37 KB Orientation, editions, service mgmt OAC Overview & Architecture, Subscribe & Provisioning
whats-new-oracle-analytics-cloud 262 KB Release notes (current + historical) Whats New & Release Updates
building-semantic-models-oracle-analytics-cloud 1.1 MB Semantic Modeler full guide Semantic Model
smml-schema-reference-oracle-analytics-cloud 162 KB SMML JSON schema reference Semantic Model, APIs, Embedding & Integration
connecting-oracle-analytics-cloud-your-data 589 KB All data source connections Data Sources & Connections
visualizing-data-and-building-reports-oracle-analytics-cloud 2.7 MB Workbooks, dashboards, BI Publisher Workbooks & Visualizations, Classic Dashboards & Analyses, BI Publisher, Maps & Geospatial Analytics
administering-oracle-analytics-cloud-oracle-cloud-infrastructure-gen-2 570 KB Service admin (Gen 2) Administration & Service Console, Subscribe & Provisioning
configuring-oracle-analytics-cloud 1.1 MB System configuration deep dive Administration & Service Console
OAC_REST_API_Guide 217 KB OpenAPI spec for OAC REST API APIs, Embedding & Integration, OCI REST APIs & CLI for OAC

Total deep reference: ~6.8 MB of text (~1.5M tokens) — fully searchable.

How to Query Deep Content (For Claude)

When a user asks a question:

Step 1: Grep raw/pdfs/*.txt for keywords from the question
Step 2: Read the matching sections (use line offsets)
Step 3: Read the related wiki page for structure
Step 4: Synthesize answer with citations:

   - Wiki Page Name
   - (Source: <pdf-name>, near line N)

Example

Q: "How do I set up an event polling table?"

Process: 1. grep -l "event polling" raw/pdfs/*.txt → finds matches in building-semantic-models.txt and smml-schema-reference.txt 2. Read those sections (~50 lines context) 3. Read Semantic Model for high-level structure 4. Answer with full procedure + cite source PDF + line numbers

Why This Beats Pre-Synthesized Wiki

Approach Coverage Accuracy Effort Currency
Pre-synthesize 6.8 MB into wiki 100% but lossy Risk of summarization errors Days Stale on next docs update
Grep raw text at query time 100% full fidelity Direct quotes from Oracle None — automated Always current with PDF

The Karpathy LLM Wiki pattern explicitly endorses this: synthesize structure in the wiki, keep sources raw and citable.

Refreshing Sources

When Oracle releases new doc versions:

  1. Download updated PDFs from docs.oracle.com/en/cloud/paas/analytics-cloud/
  2. Replace files in raw/pdfs/
  3. Re-run pdftotext: 
    cd raw/pdfs
for f in *.pdf; do pdftotext -layout "$f" "${f%.pdf}.txt"; done
  4. Update log.md with refresh date
  5. Optionally ask Claude to re-lint the wiki against the new PDFs