Yes. When AmendCheck detects no digital (Adobe Acrobat) annotations in the original document, it automatically switches to handwriting mode — scanning each page visually using AI to identify and interpret handwritten instructions.

This works best with clearly written instructions in the page margins. It has been tested successfully on pen-on-paper annotations scanned to PDF, including shorthand instructions like "add full stop" or arrows indicating positioning changes.

Handwriting recognition relies on AI vision, and some handwriting is genuinely ambiguous — especially numerals (e.g. 0 vs 6, 1 vs 7) or heavily abbreviated instructions. When AmendCheck is not confident in what it has read, it flags the item for human review rather than risk producing a wrong result silently.

This is by design. A result you can trust is more valuable than one that looks confident but may be wrong. On flagged items, AmendCheck shows you what it extracted from the handwriting so you can quickly verify it yourself.

When typed Adobe Acrobat annotations are present, AmendCheck processes those directly — they are more structured and faster to check. Handwriting scanning is not run in parallel by default, as it adds significant processing time.

If your document includes both, we recommend separating them into two checks: one for the digitally annotated version, and one where the handwritten page is the primary markup. If this is a common workflow for you, let us know — we're evaluating support for combined documents in a future update.

• Addressed — The amendment was found in the revised document and matches the instruction.
• Not addressed — The instruction appears to have been missed. Review this item manually.
• Uncertain — AmendCheck could not determine with confidence whether the change was made. Human review required.
• Needs verification — The instruction asks for confirmation of a specific value or fact that AmendCheck cannot independently verify.
• Approved — no change — A reviewer has confirmed in the comment thread that no change is required.
• Informational note — The annotation is a note or context comment, not an instruction to change anything.

In testing on real annual report documents with hundreds of annotations, AmendCheck correctly classifies the large majority of items. The tool is designed to err on the side of flagging items for human review rather than silently passing them — so when in doubt, it will return Uncertain or Needs Verification rather than Addressed.

AmendCheck is a verification-assistance tool. Its output should always be reviewed by a qualified professional before being relied upon. It is not a substitute for human judgement.

Yes. AmendCheck uses visual crop comparison for positional instructions — it looks at the relevant area in both documents rather than just matching text. This works well for instructions like "move to below the table" or "align with column header."

However, very precise typographic instructions — such as "adjust letter spacing by 0.5pt" or "increase leading between these two lines" — are difficult to verify visually. These will typically be flagged as Uncertain, which is the appropriate response: a human with the design file is better placed to confirm these.

AmendCheck reads Adobe Acrobat reply threads and looks for responses to annotations — for example, a reviewer approving a change or a second party confirming a figure. When a reply thread confirms that an item has been resolved or approved, this is taken into account in the classification.

Reply thread handling depends on how your version of Acrobat stores annotations internally. In most cases it works correctly, but if you notice annotations being classified as unresolved when they clearly have a reply, please contact us — this is usually a document-specific threading format issue that can be diagnosed.

There is no fixed page limit. AmendCheck has been successfully tested on documents over 200 pages with more than 500 annotations. Processing time scales with the number of annotations rather than page count — a long document with few annotations can process faster than a short one with many.

For very large documents, plan for the check to run for an extended period — 30 minutes or more for a 500-annotation document is normal. Keep the browser tab open until the results appear.

Processing time depends primarily on the number of annotations, not the number of pages. Rough estimates:

• Under 50 annotations — typically 1–3 minutes
• 50–150 annotations — typically 3–8 minutes
• 150–300 annotations — typically 8–20 minutes
• 300–500 annotations — typically 20–35 minutes
• 500+ annotations — allow 35–60 minutes

Documents with many handwritten annotations take longer, as each page requires visual scanning. You can leave the tab running in the background and return when it completes.

Your results and the shareable link are stored for 30 days from the date the check was run. After that, they are automatically and permanently deleted from our servers.

Your uploaded documents are not stored at all — they are processed in memory and discarded as soon as the check completes.

AmendCheck has been tested successfully on:

• Annual reports and investor relations documents with placeholder fills (e.g. [•] replaced with financial figures)
• Sustainability and ESG reports with mixed comment types
• Contracts and legal documents with word-level changes
• Documents with handwritten pen-on-paper annotations, scanned to PDF
• Documents with layout and positioning instructions

It works with any PDF that has Adobe Acrobat-style highlight annotations, or scanned documents with handwritten margin notes.

Yes. When AmendCheck encounters a scanned page where text cannot be extracted directly, it falls back to AI vision (OCR) to read the page content. This handles most scanned documents reliably, though results on very low-resolution or heavily degraded scans may be less accurate.

No. AmendCheck does not rely on page numbers to match content between documents. It uses text content overlap to align pages, which means it handles front matter offsets, section inserts, and other structural differences between the two files correctly.

This was a deliberate design decision based on real-world annual report documents where the commented version and the amended version often have different page counts due to cover pages, tables of contents, and other additions.

This is on the roadmap. The planned feature would allow you to upload a Word document (typically client-supplied copy) alongside a typeset PDF draft, and verify that all content has been correctly and completely laid out — catching missed paragraphs, transposed text, or copy that was never placed.

This is a natural extension of what AmendCheck already does, adapted for a different input format. We'll be building this after the core amendment check has been fully validated through peak-season use.

Yes — a standalone proofreading feature is planned. You would upload a single PDF and receive a report of spelling and grammatical errors throughout the document, separate from the amendment check flow.

This is designed for the final pre-publication stage, where a clean proofread of the typeset PDF can catch errors that were introduced during layout.