The narration and walkthrough tests replace the real voice with fakes that misbehave on cue: a speaker name or a word in the text triggers each failure. Most of those names were German, which reads as noise to English-speaking contributors. They are now English, and each is listed with what it does.
| Name | Was | What it does |
|---|---|---|
dies | (unchanged) | dies at the second clip |
mute | dies while loading | |
stalls | says it loads, then nothing more | |
slow | waits 350 ms before it starts: longer than the 300 ms loading deadline the tests give it, as a process start on a busy machine is | |
| clip text “hangs” | the clip gets no answer |
| Name | Was | What it does |
|---|---|---|
| sentence with the word “mute” | the voice fails that sentence |
The walkthrough test that uses it now speaks English slides (“Mute”, “One.”, “Next.”), matching the English language it runs in.
The wait has to be longer than the 300 ms loading deadline on its own, without counting on the process start to add time; 350 ms leaves 50 ms of margin.
bun test --parallel 2035 pass, 0 fail, 8 skipped (143 files, 53.7 s) bun run typecheck clean