Markdown/MDX translation test files (BookTranslator, 2026-10-25). Free to reuse.

sample-source.mdx               73-line English MDX page written for this test (Shelfmark is a
                                fictional app): front matter, an import, a JSX component, an MDX
                                comment, two code blocks, inline code, a table, a task list, a hard
                                line break, an admonition, an image, a reference link, a footnote,
                                and an in-page link to a heading
run_llm_markdown_test.py        runs the three prompt conditions against a local OpenAI-compatible
                                server (whole-file, instructed, protected)
outputs/*.mdx                   the nine Simplified Chinese outputs, unedited (moved into outputs/
                                after the run; the raw model replies are not published)
check_markdown_translation.py   structural checker (Python standard library)
check-results.txt               checker output for all nine, run with --keep Shelfmark
render_check.mjs                counts rendered bold, italics, strikethrough and literal asterisks
                                and tries an MDX compile (needs @mdx-js/mdx 3, unified 11,
                                remark-parse 11, remark-gfm 4)
render-cases.md                 seven one-line cases for CJK bold boundaries and tilde ranges
results.csv                     one row per run, including details the checker does not report

Model: Qwen 3.6 27B (qwen/qwen3.6-27b, Q4_K_M) in LM Studio, reasoning_effort "none",
temperature 0.3, one request per condition per run, Apple M4 Max with 64 GB of memory.
Rendering checked with @mdx-js/mdx 3.1.1, remark-parse 11.0.0, remark-gfm 4.0.1.
The protected condition leaves the front matter's --- fences and the closing quote of the
title and description values outside the placeholders; that design choice is part of the
result, not a bug we fixed afterwards.