SOP04 | Build a comparison between the work and the output

English translation produced by the platform. It has not been reviewed by a native speaker and is not a certified translation. The Chinese version is the source.

Version 0.1 | 2026-09-24 | draft walkthrough. Goal: turn "it looks very similar" into specific, locatable correspondences — while keeping the differences.

Read this first

This is a locator table for a human reviewer, not an automated similarity verdict. Preserve the original work and the output first, then compare on copies.

There is no worldwide standard under which "X% similarity" means infringement. A comparison cannot on its own identify a training source. Subject matter, facts, function, public domain material and specific expression all have to be told apart. Do not ask an AI to fill in a more similar version of your original or of the output.

Time, cost and tools

Plan 10–30 minutes depending on how long the material is. This is an estimate, not a measured figure. Prefer built-in or existing software.

• Your system file manager, a notes app or the writing software you already use: nothing new to pay for; save to the device itself where possible; if cloud sync is already switched on, files may end up in your cloud account.

Before you start

• Preserve anything in front of you that might disappear before doing anything else. Do not sign out or refresh first.
• Do not crop original files or screenshots, do not repair them with AI, and do not fabricate metadata.
• Record who actually did the work if someone helps you. Never give a helper your password or a verification code.

Step by step

Step 1 | Pick one work and one incident

Where to click: In your notes, write W001 and I001 with the original file name of each, and copy both files into your working-copies folder.

What you should see: The originals are untouched, and the comparison can always be traced back to the single source material.

If you can't: If all you have is a screenshot, mark it "screenshot comparison" and do not claim the pixels are identical to the original.

Step 2 | Keep the whole thing first, then choose the corresponding parts

Where to click: For text, record the chapter, page and paragraph. For images, keep the full image and mark regions only on a separate saved copy.

What you should see: A reviewer can see where the fragment sits within the whole work.

If you can't: Page numbers differ between editions, so record the edition, the chapter and the opening words so the passage can be found.

Step 3 | Text: build two columns without altering the original words

Where to click: Open the "Comparison table" template at the foot of this page. For each pair, paste a short fragment of the original first, then the corresponding fragment of the output, and record where each came from. A two-column table in Word or WPS is fine, but plain text is enough.

What you should see: Both sides come from the actual files, not from a retelling you have polished.

If you can't: If you cannot copy the text, write the page number and the screenshot reference instead. Mark anything typed out by hand as "transcribed by hand, not checked", and note the date when it has been checked.

Step 4 | Images: look at the two copies side by side

Where to click: On a computer, open each in the image viewer you already have. On a phone, switch back and forth between the two in your photo library. Describe the specific region in words — for example "three irregular notches in the third panel, lower left". Do not start with an image-recognition website.

What you should see: The correspondence is a specific structure or detail, not just "the style is the same".

If you can't: If your phone cannot show them side by side, you do not need to install anything. Record both file names and a description of the region.

Step 5 | When you annotate, only ever annotate a copy

Where to click: Duplicate the file and add _annotated to the name before using any annotation tool to draw boxes or numbers. Never overwrite the originals folder. Record any cropping, zooming or rotation in your notes.

What you should see: Both the unannotated full image and the annotated copy exist.

If you can't: If you do not know how to annotate, skip it and describe the region in words. Do not remove watermarks, paint in missing areas or sharpen an image with AI to "strengthen the evidence".

Step 6 | Record the differences and the uncertainties too

Where to click: Item by item, write what is the same, what is different, and what might be a common element, a fact or a generic style. Mark your own judgements as "I think". Do not write them as findings a court has made.

What you should see: The comparison is not only the fragments that favour you.

If you can't: Where you cannot tell whether an expression is protected, mark it as needing professional analysis rather than deleting the item.

Step 7 | State whether you gave the original to the model yourself

Where to click: Put the yes/no/unknown from SOP03 at the top of the comparison, and note any retrieval citations, rewrite instructions and attachments.

What you should see: Nobody reading the comparison mistakes a copy produced after direct input for evidence about training.

If you can't: If your records are incomplete, say unknown. Do not choose no because guessing that way suits you.

Step 8 | Save the comparison and a record of what you did

Where to click: Save your comparison table. Write any newly annotated images and their parent file names into the "Derivative and redistribution record". Generate the corresponding fingerprints as in SOP05.

What you should see: Every derived file has a stated origin and a description of what was done to it.

If you can't: Without hashes, still save both files and the written relationship between them. It does not stop you taking part.

Check when done

• There is a full-image or full-text version to locate fragments in.
• Both the similarities and the differences are recorded, along with the input conditions.
• Annotated files cannot be mistaken for the original output.

What you end up keeping

• The comparison table, or one made in the editor you already have
• The complete source material, plus optional annotated copies
• The derivative and redistribution record

If something is missing

• Only an abstract stylistic resemblance: you can still record the impression and samples, but it does not automatically count as established copying.
• The content is very long: do the clearest one to three correspondences first and keep the whole for later. You are not required to finish it all at once.

Where your data goes, and sharing

Files on your device are not uploaded by this process. Your photo library, iCloud, Google Drive, OneDrive or your editor may already be syncing, so check where something is saved before you save it. An account export communicates with the original service. A full account export is not handed to this project by default.

Keep originals somewhere private. Make a separate copy before sending anything to anyone outside, and check it first for unrelated personal material and for contract restrictions. Share only when the recipient, the purpose and the scope are clear. Nothing is sent automatically in this phase.

When to get professional help

If evidence is about to disappear, if you have received court papers or a deadline, if the chain of rights is unclear, if trade secrets or a minor is involved, or if a formal certificate is required, have a qualified professional in the relevant place check it promptly rather than waiting for this research to finish.

Basis and testing status

Source identifiers: S01, S04, S06, S10, S14, S17. Official URLs and locators are listed on the law pages. The actions in this walkthrough are preservation suggestions, not a claim that the law requires every person to complete every step.

Document checks have been done. Real-device testing on Mac, Windows, iPhone and Android, testing with five non-technical users, and lawyer sign-off have not.

Comparison table

A working note for human review, not a finding of infringement.

• Work number, version and file
• Incident number and output file
• Did you supply the original work: yes / no / unknown — details
• Retrieval, attachments and retries
• Basis of comparison: original text / original image / screenshot / transcribed by hand
• — Correspondence 1 —
• Location in the work (page, paragraph, image region) and the fragment
• Location in the output and the fragment
• Specific similarities
• Specific differences
• Parts that may be generic facts, subject matter or style
• Parts that are uncertain and need professional analysis
• Names of annotated or cropped copies and their source files
• — Correspondence 2 (add if needed) —
• Who recorded this, and when

Derivative and redistribution record

Keep the original.

• Parent file name or evidence number
• Parent file SHA-256 (leave blank if not computed)
• New copy file name
• What was done: crop / draw boxes / export PDF / convert format / redact / other
• Software used and who operated it
• When it was done, and where that clock came from
• What specifically changed, and why
• New copy SHA-256 (leave blank if not computed)
• Where the original is kept (private)
• What the new copy is for: comparison / sharing / other