TL;DR: Textract's PDF parser chokes on the signature structures inside a digitally signed PDF. Make an unsigned working copy (render the pages to a fresh PDF), run Textract on the copy, and keep the signed original as the legal record. Do not try to "fix" the signed file itself.

```text
InvalidParameterException
```

1. Confirm the signature is the trigger: run Textract on any unsigned PDF from the same source.
   Expected: unsigned files process fine; signed ones throw InvalidParameterException.
2. Check the file structure:
   `qpdf --check signed.pdf`
   Expected: warnings about incremental updates or signature dictionaries - consistent with a signed file.
3. Render an unsigned working copy:
   `gs -dBATCH -dNOPAUSE -sDEVICE=pdfwrite -sOutputFile=unsigned-working-copy.pdf signed.pdf`
   Expected: a new PDF with the same visual pages and no signature structures.
4. Submit the working copy to Textract.
   Expected: the job starts and completes normally.
5. Cite page numbers against the signed original in your review output, and note the OCR came from a working copy.

## Use this when
- Textract throws InvalidParameterException only on digitally signed PDFs
- the PDF opens fine in Acrobat or Preview but Textract rejects it
- you need OCR text, not signature validation

## Not for this skill when
- the PDF is password-protected (different cause - decrypt first with the owner password)
- Textract rejects every PDF you send (check format, size, and page-count limits instead)
- you need to validate the signature itself - Textract is the wrong tool for that

## Variant phrasings
### textract InvalidParameterException signed pdf
### aws textract fails on digitally signed document
### textract signed pdf not supported workaround

## Why it happens
Digital signatures append incremental updates and signature dictionaries that Textract's parser does not handle; it treats the structure as malformed input and returns InvalidParameterException. The visual content is fine, which is why viewers open it without complaint.

## Edge cases
- Rendering to a fresh PDF drops the signature - never present the working copy as the signed original.
- Some signers lock against modification; the print-to-PDF path through a viewer still works for an OCR copy in most cases.
- If the signed PDF is also encrypted, decrypt first, then render the unsigned copy.
- Keep both files linked in your pipeline metadata so audits can trace OCR text back to the signed source.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_5CSZYqNvV5hJilaZYaO6Ug