Office PDF Table Query

Office PDF Table Query

Use this skill when the PDF is not just reading material, but a table source that must become queryable data.

Use when

• The PDF contains tables that need extraction and analysis.
• The user wants to compare PDF table contents against other structured files.
• A narrative summary is not enough because the task needs row-level or grouped analysis.

Do not use when

• The PDF is mostly prose.
• The tables are obviously image-only and unreliable for automated extraction unless the user accepts OCR risk.
• The task can be solved by quoting or summarizing the PDF without structuring the tables.

Core rule

Extract first, trust second, query third.

Do not treat a PDF-extracted table as source of truth until you assess extraction quality.

Workflow

1. Inspect table extractability

Check:

• whether the PDF has selectable text
• whether tables have stable rows and columns
• whether merged headers, footnotes, or page breaks distort structure
• whether OCR is required

2. Normalize cautiously

After extraction:

• clean column names
• fix split rows or repeated headers
• preserve units and periods
• note any cells that may be truncated or misread

3. Assess reliability

Classify the extracted result as:

• reliable enough for direct analysis
• usable with caveats
• too weak for confident analysis

If reliability is weak, say so before running deeper queries.

4. Query with DuckDB-style analysis

Once structured data is credible enough:

• inspect schema
• bound the result size
• aggregate to answer the business question
• compare against any reference spreadsheet or memo if needed

5. Report findings with caveats

The final answer must include:

• what was extracted
• how reliable it seems
• what the query suggests
• which parts still need manual confirmation

Output format

Return:

1. Extraction assessment

• PDF type
• extraction risks
• reliability level

2. Structured table view

• normalized table or schema summary
• important cleanup assumptions

3. Query findings

• bounded results
• interpretation

4. Caveats and next step

• what still needs manual review
• best follow-up query or validation

Use together with

• pdf
• office-document-operations
• office-duckdb-read-file
• office-duckdb-query