Two kinds of re-run
They are not interchangeable, and the difference is what gets destroyed.
Fetch new transcript options. Gets new transcripts for the video and offers them as new candidates on step 2. Your current transcript, its timings, its word splits and everything generated from them are left alone. Safe on any video. This is the one you almost always want.
Re-run setup from scratch. Starts from scratch: it replaces the transcript and the selected source, which means hand-edited lines are gone and cannot be recovered. A published video is republished as a new version, so what learners see now is replaced. It also re-pays the cost of fetching and transcribing the video.
When a re-run is the right answer
- Fetch new transcript options when the transcript you have is wrong or thin and you want better candidates to choose from — a video transcribed before an improvement shipped, say. If instead the transcript is fine and what was built from it is not — missing translations, no tags — you do not want a re-run at all: re-confirm the transcript on step 2, which is what enqueues that work.
- Re-run setup from scratch only when the transcript itself is what is wrong and you would rather have a fresh one than fix the lines.
If you are unsure, fetch new transcript options first. It is cheaper, it is reversible in practice, and it fixes most complaints.
When not to re-run at all
- You already fixed the lines by hand. A full re-run throws that away. Fix what is actually wrong on the edit page instead.
- The video is live and correct. Republishing a good video to pick up a marginal improvement means learners briefly see something you have not checked.
- One field is wrong. Ask Roxxy to regenerate that field, or edit it. Re-running the whole video to fix a title is expensive and risky.
How to do it
Both re-runs live on step 1 of the setup page, one video at a time — there is no way to re-run several videos at once, by design. Fetch new transcript options is the safe one: it gets new transcripts for the video, adding them to step 2's candidate list, and leaves your current transcript, its timings and everything generated from it alone. Re-run setup from scratch is the destructive one, and it asks first: it gets new transcripts and rebuilds the whole video from them, replacing the transcript, its timings, its word splits and everything generated from them. The confirmation names what will be lost — read it.
To rebuild what was generated — translations, tags, metadata — there is no button on step 3, on purpose. That work is enqueued when you confirm a transcript on step 2, so go back to step 2 and re-confirm the transcript you want.
Ask Roxxy. She can re-run a video for you. She only has the safe mode — she cannot start a full re-run at all, and that is intentional.
What a re-run does not fix
A re-run adds a transcript candidate; it does not remove the bad one, which stays selectable. If a particular source is consistently wrong for a video, change the selection rather than re-running and hoping.
While it is running
The setup page follows the run live — the step advances on its own, and you do not need to refresh. If the run dies mid-way, the step says so and offers a retry. If it stops needing a decision, you get an email naming the step.