Runs

Locate a translation submission, inspect each document and locale, and recover the part that did not finish.

A translation run contains the documents and target languages submitted together. Open it to find which targets completed, which are still running, and which need attention. A successful target can coexist with another target's failure.

Run investigation is in Studio. For HTTP submissions, open Localization API. For file translation, open the CLI channel or follow the run link printed by the command. Repository and content integrations also link to their submitted work from their own histories.

Find the submission #

Use the available engine and API-key filters to narrow the history. Compare the run ID with the one saved by your caller, and keep its source revision alongside the investigation. Load older submissions continues through the history; the first page is not the whole account's activity.

The row shows when the submission started, its file count, target progress, and status. Expand it to see document keys and target locales. Open a stage under the affected locale to inspect its progress, warnings, failures, and model attempts.

An API job begins as Pending, displayed as Not started. Creating it saves the submission; uploading its missing artifacts and starting it begins processing. An unstarted job is not evidence of a provider outage.

Read the target and stage together #

The job's states are Pending, Running, Completed, Failed, and Cancelled. Individual targets become Completed or Failed; while a target is pending, its stages explain what is happening.

A stage can be planning batches, running, waiting for preceding work, completed, completed with warnings, skipped, cancelled, or failed. Read its accompanying message. A skipped stage can mean it was disabled or an existing translation was reused. A failed optional refinement can leave useful earlier text and a warning.

The interface can show Queued, Finalizing, or Stage failed while a target still awaits its final result. A completed model attempt or stage alone does not establish that the target artifact is ready. See Stage progress for the API fields and Requests for individual attempts.

Inspect the result at its destination #

For the API, retrieve the job and download each target whose status is Completed. A failed job can still contain completed targets with usable download URLs. Save those results before deciding what to retry.

For the CLI, use Pull to collect into the original project. For an integration, inspect its destination record or repository change. Translation completion and successful writeback are separate milestones; a generated artifact does not prove that a repository branch or published page received it.

Check the reconstructed format, target locale, variables, and accepted wording. Keep incomplete languages visible in the release decision even when other targets are ready.

Recover the affected work #

FindingNext action
Pending API jobUpload any missing artifacts, then start the saved job.
Parsing or alignment failureCorrect the selected format's source or previous translation.
Permission, entitlement, or billing failureResolve the caller's access or account condition before retrying.
Missing model coverageInspect the selected engine and its locale-pair configuration.
Human-review waitOpen the assigned work in Inbox.
Provider or stage failureRead the error and identify which targets need a new attempt.
Destination write failureRecover delivery of the completed result.

An API start retry resumes the same saved submission when appropriate; it does not restart a terminal failed job. Submit the remaining work as a new job after fixing its cause. The job reference explains request idempotency and recovery.

For CLI work, repeat the original push after correcting the cause, keeping the original state and completed files. Recover an interrupted localization run explains how to retrieve results and preserve local corrections.