i18n(zh): give the tablet demo its own Chinese subtitles - #297
Conversation
The zh landing page and README were borrowing the `-en` clip, declared as a Fallback in verify-landing-versions.ps1 so the borrow stayed visible. This replaces it with a real zh clip and removes the scaffolding. No re-recording was involved, and none was needed. The six existing demos are one English screen recording with a per-language subtitle band, and the band sits entirely below the app screen — measured, not assumed: the screen occupies y=13..933 and the subtitles y=946..1049, with pure black between. Cropping the top 934 rows off any of the six yields the original recording, unmodified, which is why the zh clip is pixel-identical to its siblings rather than a fresh take that would have drifted in timing, clock and app version. That mattered, because the emulator no longer starts on this machine. Seven attempts — three launch mechanisms including a fully detached WMI process, three GPU backends, and a second AVD — all exit ~7s in with EXITCODE=1 and no error in the log; netsimd records the guest disconnecting. Disk, WHPX and AVD locks were ruled out. Filed nothing for it here since it is a workstation problem, not a repo one. Assets follow the existing spec exactly: 960x540, 10fps, 238 frames (setpts=PTS/1.6), GIF at 128 colours. Sizes land next to their siblings — mp4 102KB (en 118KB), gif 993KB (en 951KB), still 18.4KB (en 18.7KB). Subtitle widths were checked against the 1000px box limit before rendering; widest is 679px. Font is Microsoft YaHei Bold, the Simplified Chinese counterpart to the malgunbd/YuGothB choices. Master and SRT archived alongside the other six in D:\Output\markleaf-tablet-demo-2026-07-24\. Verified: verify-landing-versions.ps1 clean — all 14 demo surfaces now point at their own language, no Fallback rows left; landing overflow gate 14 pages x 6 widths. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a631a32e1a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| @{ File = "docs/index.ko.html"; Lang = "ko"; Kind = "video" } | ||
| @{ File = "docs/index.ja.html"; Lang = "ja"; Kind = "video" } | ||
| @{ File = "docs/index.zh.html"; Lang = "en"; Kind = "video"; Fallback = "zh 데모 미촬영" } | ||
| @{ File = "docs/index.zh.html"; Lang = "zh"; Kind = "video" } |
There was a problem hiding this comment.
Adding the dedicated zh rows makes this the seventh localized demo set, but docs/assets/README.md:12-14 still inventories each tablet format as (6). Because that document exists to record what every public asset is for and when it was captured, the new files now make its counts inaccurate and appear untracked; update all three counts to seven and document the Chinese derivative alongside the existing recording details.
AGENTS.md reference: AGENTS.md:L261-L263
Useful? React with 👍 / 👎.
1 participant
Closes the last borrowed asset from the Chinese rollout. The zh landing page and README were pointing at the
-entablet clip, declared as aFallbackinverify-landing-versions.ps1so the borrow stayed visible in the output. This replaces it with a real zh clip and deletes the scaffolding.No re-recording, and none was needed
The six existing demos are one English screen recording with a per-language subtitle band — the app UI is English in all six, including ko and ja. The band sits entirely below the app screen. Measured rather than assumed, on the ko clip:
So cropping the top 934 rows off any of the six returns the original recording. The zh clip is therefore pixel-identical to its siblings in the screen area — a fresh recording would have drifted in timing, clock, and app version for no gain.
Which is fortunate, because the emulator is dead on this machine
Seven attempts before finding the above, all failing identically — exit ~7s in,
EXITCODE=1, no error line in 158 log lines, with netsimd recordingShutDown { reason: "last device disconnected" }:Win32_Process.Create)auto,swiftshader_indirect,angle_indirect)markleaf-phone-api36)Ruled out: disk (79GB free), WHPX (
-accel-checkclean), stale AVD locks. Both physical devices were unreachable over wireless ADB. This is a workstation problem rather than a repo one, so nothing is filed for it here — but it will block the next demo that genuinely needs a device.Assets match the existing spec exactly
960x540, 10fps, 238 frames (
setpts=PTS/1.6), GIF at 128 colours. Sizes land next to their siblings:.mp4.gif-still.webpSubtitle widths were checked against the pipeline's 1000px box limit before rendering — widest is 679px, so nothing clips. Font is Microsoft YaHei Bold (
msyhbd.ttc), the Simplified Chinese counterpart to themalgunbd/YuGothBchoices already in use.Master and SRT archived beside the other six in
D:\Output\markleaf-tablet-demo-2026-07-24\.Scaffolding removed
The
Fallbackfield, its comment block, and the yellow-output branch are gone fromverify-landing-versions.ps1. It existed to make one temporary borrow visible; keeping it now would describe a state that no longer exists.Verified
verify-landing-versions.ps1— all 14 demo surfaces point at their own language, no Fallback rowscheck-landing-overflow.mjs— 14 pages × 6 widths, no horizontal overflowNot in this PR
No
markleaf-tablet-demo-zh-youtube.md. The other six have one, but uploading is a manual step and nothing is queued for zh.🤖 Generated with Claude Code