Tagged PDF (PDF/UA) Publish Conformance Gate

Audit the PDF files that you publish against the tagged-document rules of PDF/UA-1, PDF/UA-2 or WCAG 2.2, before the file goes on the website.

Give the Actor a bounded list of PDF URLs. The Actor downloads each file once, reads its structure tree, and writes one dataset row for each file plus one gate summary row. It does not crawl, it does not log in, it sends no credential, and it uses no browser and no proxy.

A failed gate is a result, not a failed run. The run always ends with success, and the verdict is in the dataset and in the run status message.

Input

Field	Type	Default	Function
pdf_urls	array of strings	three public example files	1 to 500 absolute HTTPS URLs of PDF files.
profile	string	PDF/UA-1	PDF/UA-1, PDF/UA-2 or WCAG-2.2-document.
max_pages_per_file	integer	200	How many pages of one file the Actor reads.
fail_on	array of strings	["error"]	The severities that give a file the verdict fail.
timeout_seconds_per_file	integer	60	Download timeout of one file.

Every field has a default, so a run with empty input audits the public example fixture in .actor/default_input.json.

Output

One row for each file, with record_type file:

file_url, http_status, bytes, page_count, pages_audited, is_encrypted, is_tagged, tag_tree_depth, document_language_set, title_set, display_doc_title_flag, images_total, images_without_alt_text, tables_without_header_scope, headings_out_of_order, untagged_content_items, reading_order_conflicts, form_fields_without_label, findings[] and verdict (pass, fail or unreadable).

Each entry of findings[] holds rule_id, severity, page, object_path and message.

One summary row, with record_type summary: files_requested, files_audited, files_pass, files_fail, files_unreadable, findings_by_rule, findings_by_severity and gate_result.

Rules

The profile sets the severity of each rule. A rule that a profile does not state is not applied.

Rule	PDF/UA-1	PDF/UA-2	WCAG 2.2	Function
PDFUA-TAGGED	error	error	error	The document has /MarkInfo /Marked true and a /StructTreeRoot.
PDFUA-LANG	error	error	error	The document states a default language.
PDFUA-TITLE	error	error	error	The document states a title in XMP dc:title or in /Info /Title.
PDFUA-DISPLAY-TITLE	error	error	warning	/ViewerPreferences /DisplayDocTitle is true, so the reader shows the title and not the file name.
PDFUA-XMP-IDENT	error	error	—	The XMP metadata states the pdfuaid:part of the profile.
PDFUA-FIGURE-ALT	error	error	error	Each figure holds alternative text.
PDFUA-UNTAGGED-CONTENT	error	error	error	No content is outside the structure tree and unmarked as an artifact.
PDFUA-TABLE-HEADERS	error	error	error	Each table holds header cells, and each header cell states its scope.
PDFUA-HEADING-ORDER	error	error	warning	The heading levels go down one step at a time.
PDFUA-FORM-LABEL	error	error	error	Each form field holds a label or a tooltip.
PDFUA-READING-ORDER	warning	warning	warning	The structure order agrees with the content order of the page.
PDFUA-TAG-DEPTH	warning	warning	warning	The structure tree is more than one level deep.
PDFUA-OUTLINE	—	—	warning	A document of 10 pages or more holds a bookmark list.
PDFUA-PAGES-TRUNCATED	info	info	info	The page cap of the run stopped the audit before the last page.
PDFUA-UNREADABLE	error	error	error	The Actor could not download the file or could not read its structure.

Pricing

The Actor uses the pay-per-event model. It charges these three events:

Event	Unit	Function
run_start	one for each run	The container start and the input read. Charged once, before the Actor reads the input.
pdf_audited	one for each file that the Actor reads and audits	The download and the structure parse of one file. A file that the Actor could not download is not charged.
page_batch_over_50	one for each 100 pages above the first 50 pages of a file	The extra parse time of a long document. A file of 50 pages or fewer charges none.

Each event is charged once for the whole run, with the total count. The Actor never charges more than the run's maximum total charge allows.

Limits

• 1 to 500 URLs for each run, HTTPS only, no credentials in the URL.
• Loopback, private and reserved addresses are refused, and every redirect hop is checked again.
• 3 redirect hops, 32 MB for each file, 5 parallel downloads.
• 1 to 5000 pages for each file, 5 to 300 seconds for each download.

Development

uv sync
uv run pytest
uv run ruff check .