Skip to main content
GET
Get an AI Dubbing QC Run
Returns a run and its stored results. Reading a run does not start another evaluation or charge credits.

Authorization

Send the integration API key in x-api-key. The order and run must belong to the client identified by the key.

Path parameters

string
required
ID of the order.
string
required
ID of the AI Dubbing QC run.

Run status and quality verdict

Run status is one of queued, running, completed, partial, failed, or blocked. Poll while queued/running; stop at a terminal status. Infrastructure outcomes may be partial, failed, or blocked and are refunded at most once; partial indicates incomplete coverage. A quality verdict is separate: report.verdict is pass, pass_with_fixes, or fail. Completed quality failures remain charged. AI Dubbing QC covers completed AI Dubbing orders that were paid with AI credits. For an order with visual translation, the translated on-screen text is checked too, as a separate section of the report. Lip-sync accuracy is scored only when LipSync covers every segment. The API key’s user needs the client_owner, client_view or client_createOrder role; other roles receive 403. The report uses schema 1.0 field names; it is not wrapped in another object. Its top-level fields are schema_version, job_id, category, generated_at, languages, duration_seconds, options, score, verdict, dimensions, checks, findings, rejected_findings, corrective_actions, passes, and limitations. The dubbing score has five dimensions: translation accuracy, synchronization, voice, audio mix and completeness. LipSync on every segment contributes to the sync dimension; there is no separate LipSync score. When LipSync covers only some segments, the order is checked as a standard dub and lip-sync accuracy is not scored. Each entry in passes has a name, a status and its duration in seconds. The following is an illustrative excerpt only, not a complete qc_report.json:

Language checks

Translation accuracy is examined one language check at a time. report.checks has one entry per check, and every translation accuracy finding names its check in check. The checks score inside the translation accuracy dimension; they do not add dimensions or change the score formula. Each entry has id, label, dimension (accuracy), result, findings (the number of standing findings) and an optional note. result is one of: note also says what a result rests on, for example that terminology was checked without a glossary, or that an issue was reported but not upheld when it was re-checked. A finding that was not upheld is listed in rejected_findings, not in findings. A completeness finding for a missing line can also carry check: "omission_addition".
A report made before language checks were introduced has no checks and no check on its findings.

On-screen text

For an order with visual translation, the run also checks the translated text in the picture: titles, burned-in captions, lower thirds, labels, interface text, end cards and disclaimers. Brand and product names, logos and web addresses are expected to stay as they are. Such a run has scope.modules.visual_translation: true, and its report carries extension_version: "av-qc-ost/v1" and a visual_translation section. That section is a complete report of its own, with the same fields as the dubbing report (score, verdict, dimensions, findings, rejected_findings, corrective_actions, passes, limitations) and category: "visual_translation". It does not change the dubbing score, verdict or findings. coverage_complete says whether all on-screen text could be read and compared; when it is false, coverage_limitations lists what was not. A run whose on-screen text was only partly readable is still completed and charged. If the on-screen-text check could not run at all, the run is partial: the dubbing report is returned and the charge is refunded.
A run made before on-screen text was checked has scope.modules.visual_translation: false and no visual_translation section. scope is the scope stored with this run: what was produced and scored. Do not substitute the current pricing response’s scope when displaying a historical report. artifactFingerprint and checkedArtifacts identify the exact files that were checked. Each checkedArtifacts item has id, type, name, createdAt, updatedAt, and urlHash; no media URLs are returned. If outdated is true, the order’s files have changed since this report was made. Use the export endpoint to download an XLSX built from this stored report. Exporting does not run QC or charge again.