Practise the standard verbs for versioning and rolling back LLM prompts.
0 / 5 completed
1 / 5
Fill in: 'We ___ every production prompt in source control so a silent edit can't change model behaviour without a reviewable diff.'
We 'version a prompt' — the standard, established collocation for tracking prompt changes like any other source artefact. The other options aren't the recognised term here.
2 / 5
Fill in: 'Editing a prompt directly in a dashboard without versioning can ___ last week's output impossible to reproduce during an incident.'
We say unversioned edits will 'leave' output irreproducible — the standard, natural collocation for the resulting gap. The other options aren't idiomatic here.
3 / 5
Fill in: 'We ___ each prompt version with the model and parameters it was evaluated against, so a regression is traceable to an exact change.'
We 'tag a version' — the standard, established collocation for attaching identifying metadata to a stored artefact. The other options aren't the recognised term here.
4 / 5
Fill in: 'We ___ a regression eval suite against every new prompt version before it replaces the one currently live.'
We 'run' an eval suite — the standard, simple collocation for executing a set of automated checks. The other options are less idiomatic here.
5 / 5
Fill in: 'We ___ to the previous prompt version immediately when an eval score drops, rather than debugging live in production.'
We 'revert' to a version — the standard, established collocation for restoring a known-good prior state. The other options aren't the recognised term here.