File types and Markdown

These apply to both file routes: uploads and POST /parse.

File types

File	Extensions	`type`
PDF	`.pdf`	`pdf`
Word	`.docx`, `.docm`, `.dotx`, `.doc` (97-2003), `.rtf`, `.odt`	`docx`, `doc`, `rtf`, `odt`
PowerPoint	`.pptx`, `.pptm`, `.potx`, `.ppsx`, `.ppt` (97-2003), `.odp`	`pptx`, `ppt`, `odp`
Excel	`.xlsx`, `.xlsm`, `.xls` (97-2003 and 5.0/95), `.ods`; macros never run	`xlsx`, `ods` (`xls`, `xlsm` work too)
CSV, TSV	`.csv`, `.tsv`	`csv`, `tsv`
Text, Markdown	`.txt`, `.md`	`txt`, `md`
Images	`.png`, `.jpg`, `.webp`, `.tif` (every page), `.gif` (first frame), `.bmp`, `.heic`, `.heif`	`png`, `jpeg`, `webp`, `tiff`, `gif`, `bmp`, `heic` (`jpg`, `tif`, `heif` work too)

How it’s read
PDF pages	Every page with something on it is read by OCR, born-digital ones too, so tables, headings and reading order are kept
A page OCR skips	Keeps its own text layer
`.doc`, `.rtf`, `.odt`	Read as Word (`docx`); `.ppt`, `.odp` as PowerPoint (`pptx`); `.xls`, `.ods` as Excel (`xlsx`)
`.ppt` tables	Read as their cells’ text, one per line
Charts	Their title, axis titles and a table of their values

Refused with 415 unsupported_file:

File	Save it as
Word 6.0/95	`.docx`
AVIF images	JPEG or PNG

Element	Written as
Page, slide, sheet	Starts with `<!-- page 3 -->`, `<!-- slide 4 -->` or `<!-- sheet "Q3" -->`
Headings, lists	`#` and `##`; `-` and `1.`
Tables	Markdown with a header row; an HTML `<table>` with `colspan` and `rowspan` when cells merge or it’s over 30 columns or 300 rows
Numbers	Exactly as printed
Running header and footer	`<!-- header: ... -->` and `<!-- footer: ... -->`; page numbers are left out
Footnotes	At the end of their page
Text boxes, sidebars	In reading order
Checkboxes, form fields	☑ and ☐; `key: value` lines

With layout=true, each part OCR read carries blocks, the regions of its page.

Field	Meaning
`type`	`title`, `text`, `list`, `table`, `image`, `header`, `footer`, `page_number`, …
`bbox`	`[x0, y0, x1, y1]` as shares of the page from its top left
`order`	Reading order
`content`	The block’s text
`conf`	OCR’s certainty

Every picture is read by OCR, whatever its size on the page; on upload each is a part of the file’s document (locations).

Skipped as icons or decorations: under 64 pixels on both sides, under 16 pixels on the shorter side, or shown under half an inch on both sides.
A picture used again on a later slide is read once, where it first appears.

Each picture that gave no text is in without_text.images with its reason:

`reason`	Meaning
`too_small`	An icon or decoration, with its `width` and `height` in pixels
`not_raster`	A vector picture (EMF, WMF, SVG), which isn’t read
`unreadable`	The picture is damaged
`not_read`	OCR couldn’t read it in time
`no_text`	OCR found no text