Read parsed content
Read each element’stype, text, and metadata.page_number. The captured example below contains nine elements from one page.
Choose a representation
The OpenAPI contract defaultsoutput to markdown. Request elements for individual document elements; the captured response then returns markdown: null. See Parse parameters.
Check page failures and incomplete content
A completed result can omit unreadable pages or include pages with missing content. Checkfailed_pages and degraded_pages before using the result.
These arrays are optional. Transform omits each array when there are no entries. Page reports apply to both Markdown and elements output.
The
warnings array can include pages_failed or pages_degraded. Use the page reports to locate the affected pages. Compare those pages with the source document before accepting the result. See warning handling.
Read element coordinates
When coordinates are present,metadata.coordinates contains one box for an element on one page. An element that spans pages has an array of boxes. Each box in that array includes its page_number.
Coordinates can be null or absent. Check the value’s shape before treating coordinates as a single object.
Read Parse confidence
Withoutput=elements, an element’s metadata.parse_confidence can contain a number from 0 to 1, rounded to four decimal places. The score measures how closely the element text agrees with optical character recognition (OCR) text. It is not a probability that the text is correct.
A null or missing value means the score is unavailable, not zero. Receiving a score does not require an include option. Extract does not return this score for extracted fields.
Handle an accepted request
HTTP 202 means the request is still running. Keep the completeLocation URL and follow request progress. The endpoint reference includes schemas for pending responses.
See asynchronous processing for progress and Extract to request structured fields.

