[jp-sync] Replay failed Angular + xplat JP syncs for EN PRs #522, #438, #535, #491

[ChronosSF]

Original authors: Marin Popov (desig9stein@users.noreply.github.com) — #522 · Stamen Stoychev (chronos.stz@gmail.com) — #438 · Dobromir Tsvetkov (dobromirts@users.noreply.github.com) — #535 · Hristo Hristov (Hristo313@users.noreply.github.com) — #491

Summary

Manual replay of the eight failed JP sync runs for the four vnext pushes #522, #438, #535 and #491 — both the Angular workflow (33877005832, 34238718782, 34250163709, 34329511530) and the xplat workflow (33877005838, 34238718784, 34250163990, 34329511576). Every run died in the Execute GitHub Copilot CLI step with transient_bad_request on all retries before doing any work, so nothing partial exists; the workflows have since been recompiled with gh-aw v0.88.7 in #555.

Each change set is the union of EN changes since the respective last successful sync (34497daa95 for Angular, d34687170d for xplat), filtered to TOC-covered files exactly as the workflow specs prescribe. Two commits: Angular first, xplat second.

Angular JP (docs/angular/src/content/jp/)

English file	Japanese counterpart	Action
components/chat.mdx	jp/components/chat.mdx	Updated — messages row + new "双方向のメッセージ コレクション" section (#438)
components/ai/theming-mcp.mdx	jp/components/ai/theming-mcp.mdx	Updated — WCAG contrast DocsAside before the tools table (#522)
components/themes/palettes.mdx	jp/components/themes/palettes.mdx	Updated — hsl() → hsla(… / 1) contrast variables in the CSS excerpt (#522)
components/toc.json	jp/components/toc.json	Updated — new テーマ › アクセシビリティ entry (new: true); Button → inputs/button.mdx (#522, #491)
components/dialog.mdx, time-picker.mdx, tooltip.mdx	same paths under jp/	Updated — IgxButton link → ./inputs/button.mdx (#491)
components/themes/misc/angular-material-theming.mdx, bootstrap-theming.mdx	same paths under jp/	Updated — Button link → ../../inputs/button.mdx (#491)
components/themes/roundness.mdx	jp/components/themes/roundness.mdx	Updated — Button link → ../inputs/button.mdx (#491, #535)
grids_templates/row-editing.mdx, search.mdx	same paths under jp/	Updated — button links → inputs/button / inputs/button#スタイル設定 (#491)
components/list.mdx	jp/components/list.mdx	No change needed — JP already linked ./inputs/button-group.mdx (#535)
components/button.mdx (deleted in EN)	jp/components/button.mdx	Deleted — superseded by the xplat-generated inputs/button.mdx (#491)
components/layouts/avatar.mdx (deleted in EN)	jp/components/layouts/avatar.mdx	Deleted — superseded by the xplat-generated page (#522)
images/button/*.png, images/anatomy-content-light/button-lt-a.png	same paths under jp/images/	Added — the Angular build resolves @xplat-images per language, so the generated JP button page needs them

xplat JP (docs/xplat/src/content/jp/)

English file	Action
components/inputs/button.mdx	Re-mirrored — EN was rewritten (482 → 1096 lines); JP now follows the new structure, reusing existing JP prose wherever EN kept it (#491)
components/inputs/button-group.mdx	Updated — Angular-specific PlatformBlocks, {PackageCommon} install block, per-platform Properties/API References/Dependencies, new Related Components and FAQ (#491)
components/inputs/badge.mdx	Updated — {PackageCommon} install block, Angular import sentence, CSS variables table, troubleshooting intro, FAQ (#491)
components/interactivity/accessibility-compliance.mdx	Re-mirrored — Section 508 / WCAG matrices replaced by the new standards, criteria, per-component documentation index, configuration, VPAT and FAQ sections; cross-page anchors use the JP heading slugs (#キーボード-ナビゲーション, #アクセシビリティ, #アクセシビリティと-aria-サポート), and link to the page only where the JP target lacks the section yet (Angular carousel/tooltip, xplat carousel) (#522, #491)
components/interactivity/chat.mdx	Updated — Blazor MessageCreated guidance, per-platform property names in the table, new bidirectional messages section with per-platform snippets (#438)
components/themes/overview.mdx	Updated — configureTheme variant argument, new Sass によるカスタマイズ section (#522)
components/themes/palettes.mdx	Updated — runtime relative-color contrast note, new CSS excerpt with seed / --ig-wcag-* variables, seed and contrast-level paragraphs (#522)
components/themes/accessibility.mdx	Added — full translation of the new page (#522); target of the new toc entries and the theming-mcp note
components/ai/theming-mcp.mdx	Updated — WCAG contrast note (#522)
toc.json	Updated — lifted "exclude": ["Angular"] on ボタン, アイコン, アイコン ボタン (EN #491) and アバター (EN had already lifted it; the stale JP exclude is why the hand-authored JP avatar page was still being maintained), "updated": true where EN has it, new テーマ › アクセシビリティ entry (#522)

Not mirrored on purpose: the EN toc href fix grids/theming-grid.mdx → grids/grid/theming-grid.mdx (#535) only corrected EN after a July file move (#355) that was never applied to JP. JP still keeps the page at grids/theming-grid.mdx with ~22 consistent links, so the relocation is a separate follow-up rather than part of this replay.

Verification

• npm run check:llms-metadata — passed (1096 frontmatter blocks, 964 EN+JP topics)
• markdownlint and cspell on all 21 changed .mdx files — 0 issues
• check-relative-links after the full CI-equivalent generation (Angular EN/JP sync + grid templates, React/WC/Blazor EN+JP): Angular 0 broken, xplat 0 broken
• en/jp xplat toc.json exclude comparison: only one pre-existing, unrelated drift remains (general-cli-overview.mdx)

Checklist:

• check topic's TOC/menu and paragraph headings
• Include TOC topic labels in the topic content when it has a valuable update, is new, or is considered preview / beta
• link to other topics using ./page.mdx or ../relative/path.mdx (.mdx extension required)
• at the References section at the end of the topic add links to topics, samples, etc
• reference API documentation instead of adding a section with API

• use valid component names - [Data] Grid, IgxSelectComponent, <igx-combo>
• use spell checker tool (VS Code, Grammarly, Microsoft Editor)
• add inline code blocks for the names of classes / tags / properties
• add language descriptor for the code blocks
• check if links function by running the check-api/mdx npm commands
• check if sample is working and fully visible in the topic
• check if sample is working and fully visible in the StackBlitz
• check if code blocks match the code in StackBlitz demo
• add or update meaningful llms.description metadata and run npm run check:llms-metadata

• follow SEO guidelines to fill topic's metadata (SEO guidelines)

• do not resolve requested changes (leave that to the reviewer)

🤖 Generated with Claude Code

[ChronosSF]

@jsakamotoIGJP , this run does not exactly replay the GH AW workflow but rather uses the same rules but passed on Fable 5.1 Max . I know you guys have issues with the quality of the default GH AW run and I am wondering if Fable 5.1 produced better results translating this. The model quality variable in the output result will also be important when choosing the models for the next translation process.

Let me know what you think about the work Fable 5.1 did with these.

[jsakamotoIGJP]

@ChronosSF I've already started reviewing this PR, but I'd appreciate until tomorrow to share my findings.

[jsakamotoIGJP]

@ChronosSF Thank you for your patience.

In summary

I think the quality is significantly better than before, but I'm not sure Fable is required to achieve that, since our AI review was running on cheaper models.

Details

I do not have any measured numbers, so this is only my personal impression. Compared to the past JP syncs, the translation quality is much better.

The improvement is not only in the translation itself. I didn't find any untranslated English paragraphs, and I didn't find structural problems such as missing or misplaced PlatformBlocks. This is a very good improvement.

That said, a few fixes were still needed.

To be transparent, our "human review" is now heavily assisted by AI agents. For this review, I used Claude Code with Opus 5 (High effort) and a custom prompt. I did not use Fable for the review side.

Claude Code's fixes fell into three main areas.

First, one translation issue. The word "boundaries" was translated correctly at the word level, but the result sounded unnatural in context. A more natural interpretation was needed. This was the only translation correction I made, and it was a difficult case.

Second, consistency issues. Some terms and expressions were different from existing Japanese documentation. My custom prompt checks existing .mdx files and treats them as translation examples. This helps keep wording consistent across pages, even without a formal glossary.

Third, anchor links. When headings are translated, anchor names can change, and links can break. The review process detected such issues, and they were fixed.

Overall, the amount of translation correction was very small compared to previous reviews. In fact, there was only one real translation fix.

Because of that, I am not sure how much of the result comes from the model itself and how much comes from the review prompt and workflow. It may be possible to get similar results with Opus 5, Sonnet 5, or even less expensive models if the prompt is well designed.

One more thought. The remaining translation issue was a difficult wording choice. This suggests that writing English documentation in a simpler, more consistent style may also improve translation quality across all languages.