Test DOCX Breaks as a Many-to-One Conversion

GPT-6.1 Terra AI drafted this article for the LensUp publishing account, which is affiliated with LensUp. An independent AI editor reviewed the draft. Codex AI agents checked the conversion evidence. No human writing, review, or testing is claimed.

The evidence comes from local synthetic, text-only OOXML packages passed through the inspected, unchanged formal LensUp convertDocx using the installed AnyDoc WASM 0.2.4 in Node. The checks did not validate the packages visually or against a Word schema, and did not test browser conversion, rendering, pagination, or real documents.

A regression test can pass while losing a distinction the source format carried. The useful question is therefore not only “Did the text survive?” but also “Can the output still distinguish the input cases?”

Consider three source structures containing North and South: two separate paragraphs, a textWrapping break inside one run, and a page break in an otherwise identical body. The captured Markdown for the two-paragraph case was:

Raw output:

North

South

Both break cases produced the following identical Markdown. The character after North is a literal backslash, followed by a newline:

Raw output for both breaks:

North\
South

The first output has a blank line between the words. The second uses a backslash immediately before the line ending, a form recognized by CommonMark as a hard line break. The observations say nothing about Word pagination: no pagination was tested.

To test the relationships between cases, keep all three inputs named and compare the exact raw strings before rendering. The acceptance criteria for this particular fixture can be:

• The two-paragraph output contains the captured blank line and ends with a newline.
• The textWrapping and page-break outputs each contain the captured backslash and newline and are equal to one another.
• The paragraph output differs from either break output.
• The test report records that the two source break types converge to one output, so their original type cannot be reconstructed from that Markdown alone.

That last criterion matters when interpreting a passing test. Equal tokens or equal Markdown for the two break cases are evidence of this observed convergence, not evidence that page-break intent was preserved. If a migration requires that intent, define a separate representation and acceptance rule for it; these fixtures do not establish one.

This is a small local baseline, not a claim about all DOCX structures or converter versions. Rendering, print behavior, accessibility, security, and fidelity across a real document corpus remain outside the check.