Recover an interrupted localization run
Find what the submitted run actually completed, retrieve recoverable output, and preserve local corrections while resolving the rest.
A terminal closing during lingo push --wait does not stop the server by itself. Translation may still be running, waiting for review, or already complete. Find that state before submitting the same content again.
This guide begins with submitted work. For a new correction, use a scoped update. For accepted targets that must survive ordinary changes, use preservation guidance.
Keep the evidence and local state #
Record the printed run ID, source revision, selected files and locales, and last known status. Preserve the checkout's run state, lockfile, and any local target edits. Deleting those records first can remove the information needed to recover the work.
For a run submitted to two target locales, investigate each target rather than assuming that losing the terminal interrupted both at the same stage.
Inspect retrieval before starting again #
From the submitting checkout, use its recorded run:
lingo pull --dry-runThis checks prospective local writes without changing files. It can poll unfinished work and report a review wait, failure, or available output. Pull writes each document once all of its locales are done and leaves documents waiting for human review alone; run it again after the reviewers finish. Resolve required review in its own workflow instead of repeatedly resubmitting source.
When output is ready, inspect conflicts, then retrieve:
lingo pullPull uses local recorded state; it does not search for arbitrary runs from another machine. Receive translations committed elsewhere through version control. The run-state reference and Pull explain the installed CLI's exact files and conflict rules.
Retry failed work with push #
After correcting the reported cause, repeat the original push, with the same arguments, from the submitting checkout. For an ordinary source update:
lingo push --waitIf the earlier push was never collected and the request is unchanged, push reuses it instead of submitting the same work twice. A job that is still running is re-attached, a completed one is collected, and a failed one is replaced by a new job with a new ID. Documents collected before a failure are already recorded in the lockfile, so the retry asks only for work that is still missing or changed.
After the retry, inspect target outcomes and the retrieved files, then identify what remains. Before backfilling missing target files in an existing project, complete, retrieve, and inspect ordinary source updates for existing targets. Keep that source revision fixed while filling missing files with lingo push --backfill-missing --wait; Add a language explains the sequence. Changed keys inside existing files need the appropriate ordinary or scoped update instead.
Match the repair to the failure #
A missing local record differs from a failed server run. Invalid content needs correction, refused authority needs access, and a review wait needs a decision. An uncertain network response needs inspection of whether the server accepted work before another submission.
Do not change source merely to obtain another attempt at a failed run. Retain the old run ID and confirm that the newly printed job ID differs. The CLI recovery procedure covers output verification and the support route if the same failed job is returned.
For partial results, preserve successful reviewed targets and list failed targets separately. If a new run is needed, record why it follows or replaces the earlier one and which source revision it uses. That prevents a late result from being mistaken for current output.
Reconcile local divergence #
Pull can refuse to replace a target changed since its recorded version. Compare the local edit and incoming output against their source revisions, then choose accepted wording through review.
lingo pull --force bypasses that protection and overwrites divergent targets. Save local work and deliberately choose replacement first. With shared catalogs, protect the entire file, including source and untouched locales.
Finish with usable output #
Compare retrieved paths, locales, variables, and the diff. Load the result in its application or renderer and examine meaning. A file can be retrievable without being approved for use.
Keep the recovered run ID, accepted revision, lockfile, and targets as the next baseline. Recovery is complete when remaining failures are accounted for and the accepted output matches the intended source, not merely when one command returns successfully.