Format Details¶

Detailed breakdown of how OCC extracts metrics from each format.

Word (.docx)¶

Parser: mammoth

Metrics extracted:

• Words — raw text extracted via mammoth.extractRawText(), then split on whitespace
• Pages — estimated at 250 words per page (Math.max(1, Math.ceil(words / 250)))
• Paragraphs — text split on double newlines, filtered for non-empty segments

Structure extraction: mammoth converts DOCX to HTML (mapping Heading 1–Heading 6 styles to <h1>–<h6>), then turndown converts to markdown with #–###### headers. This gives accurate heading hierarchy without parsing DOCX XML directly.

Table extraction: occ table inspect parses <table>/<tr>/<td> elements from mammoth's HTML output, preserving colspan and rowspan attributes for merged cells. Headers are auto-detected from <th> tags or unique first-row values.

PDF (.pdf)¶

Parser: pdf-parse

Metrics extracted:

• Words — text extracted by pdf-parse, then split on whitespace
• Pages — actual page count from the PDF metadata (data.numpages)

Structure extraction: pdf-parse is invoked with a custom pagerender callback that prepends [Page N] markers before each page's text. These markers enable section-to-page mapping in the structure tree. Headers in the extracted text are identified by markdown heading syntax.

PDF is the only format that provides a true page count rather than an estimate.

Excel (.xlsx)¶

Parser: SheetJS (xlsx)

Metrics extracted:

• Sheets — workbook.SheetNames.length
• Rows — derived from each sheet's !ref range
• Cells — rows × columns derived from each sheet's !ref range

Word and page counts are not extracted from spreadsheets.

Spreadsheet inspection: occ sheet inspect <file> uses the same SheetJS workbook model to expose workbook properties, hidden sheet state, defined names, formula/comment/hyperlink signals, inferred schema, and lightweight row samples for agent-oriented preflight.

Table extraction: occ table inspect treats each sheet as a table, using the same getCell() and renderCell() utilities as sheet inspection. Merged cells are handled via the !merges array, with the top-left cell carrying colspan/rowspan values.

PowerPoint (.pptx)¶

Parser: JSZip + officeparser

Metrics extracted:

• Words — text extracted via officeparser, then split on whitespace
• Slides — counted by inspecting the ZIP structure for ppt/slides/slideN.xml entries

Structure extraction: Slides are enumerated from the ZIP in order and # Slide N headers are inserted, creating a flat one-level structure.

Table extraction: occ table inspect finds <a:tbl> elements in slide XML, extracting text from <a:t> tags within <a:tc> cells. gridSpan and rowSpan attributes are read for merged cells. Each table's location is reported as Slide N.

ODT (OpenDocument Text)¶

Parser: officeparser

Metrics extracted:

• Words — text extracted via officeparser, then split on whitespace
• Pages — estimated at 250 words per page (same as Word)
• Paragraphs — text split on newlines, filtered for non-empty segments

Structure extraction: Text is extracted via officeparser. Heading detection is best-effort since ODT formatting may not always be preserved in the plain text output.

Table extraction: occ table inspect parses <table:table> elements from content.xml, extracting text from <table:table-cell> elements. table:number-columns-spanned and table:number-rows-spanned attributes are read for merged cells.

ODS (OpenDocument Spreadsheet)¶

Parser: JSZip + officeparser

Metrics extracted:

• Sheets — counted by matching <table:table elements in content.xml
• Rows — counted by matching <table:table-row elements in content.xml
• Cells — counted from officeparser text output (non-empty lines)

ODP (OpenDocument Presentation)¶

Parser: JSZip + officeparser

Metrics extracted:

• Words — text extracted via officeparser, then split on whitespace
• Slides — counted by matching <draw:page elements in content.xml

Structure extraction: Similar to PPTX — slides are counted from content.xml and # Slide N headers are inserted.

Table extraction: occ table inspect parses <table:table> elements from content.xml within each <draw:page>, providing per-slide context for each table's location.