#Chunk Overrides

The "Chunk Override" panel in the CAT Editor re-translates a single chunk with settings that differ from the rest of the job — most importantly, a different translation engine. Combined with direct editing, it gives you two ways to fix a chunk without re-running the whole document.

#Two Ways To Override A Chunk

Goal	Tool
You know the exact wording you want	Edit the Target pane directly and Save — the chunk becomes "Edited (Human)" and your text is what exports
The engine produced a bad or failed result and you want a fresh machine attempt	Chunk Override → "Re-run chunk" with a different engine

A failed chunk shows exactly this choice: "This chunk failed. Type your translation below and Save, or use Re-run chunk with a different engine."

#What Happens When You Re-run

Click "Re-run chunk". The button switches to "Re-translating..." and an overlay covers the chunk: "Re-translating chunk..." / "Please wait. Your changes will be overwritten."
Only this chunk is sent to the selected engine — the rest of the job is untouched.
The new machine output replaces the current target text, and the chunk's editor status resets to "Draft (Machine)" — a re-run is machine output again, so it needs review and approval like any other draft.
The previous text is preserved in Version History; use "Revert" if the re-run made things worse.
QA flags and the quality score are recomputed, and the run is recorded in the "Translation Layers Inspector" under "Chunk Override".

A re-run polls for up to 5 minutes before timing out; local engines that are still starting up can make the first re-run slow.

#When To Use A Re-run

A chunk failed during the job (network hiccup, engine restart) and you want to retry it with the same or another engine. For many failed chunks at once, prefer "Retry Failed" on the Translations screen instead.
One engine consistently mangles a specific passage — idioms, poetry, dense technical text — and you want to compare another engine's attempt.
You translated the job with a fast engine and want higher-quality output for a handful of important chunks (for example re-running key passages with NLLB).
The chunk contains a language-pair edge case your default engine handles poorly.

#When Not To Use It

Terminology drift across the project — that is a Glossaries problem. A re-run without a glossary fix will drift again.
Awkward but correct text — Improve With AI refines existing output instead of replacing it with a fresh machine draft.
Small wording tweaks — just edit and Save; it is faster and keeps your status as Edited (Human).

#Glossary Vs Override Vs Edit

Need	Use
A character name should always use the same spelling	Glossary entry
A product term has one required translation everywhere	Glossary entry
A repeated legal sentence must match approved wording	Edit the chunk(s) directly and Save; use Find and Replace on the Translations screen for many occurrences
One chunk came out wrong and a different engine might do better	Chunk Override re-run
A subtitle line needs a shorter version to fit its timing	Edit directly in the Subtitle editor — see Translating Subtitles

#Keeping Overridden Chunks Under Control

Approve after overriding. A re-run resets the chunk to Draft (Machine); review it and press Ctrl+Enter so it is counted in review progress and feeds translation memory.
Re-runs overwrite edits. The overlay warns you: any manual changes to the chunk are replaced by the new machine output. Save valuable manual work first — Version History keeps it recoverable, but only if it was saved.
Use the Layers inspector for audits. When a chunk has been translated, overridden, improved, and edited, the "Modification Timeline" shows exactly what happened in what order.
Watch the score. After a re-run, the quality badge and QA flags tell you quickly whether the new engine actually did better.