Start with a trusted cut and text track
This workflow is for a translator, reviewer and video editor coordinating one language track in a spreadsheet. It is not a synchronization rescue or bilingual merge. Confirm the exact cut and inspect opening, middle and closing dialogue before exporting. Record which extracted track is authoritative; a duration match alone cannot establish the right edit.
Use existing SRT or VTT when possible. Converting text-based ASS or SSA to SRT trades away styling, positioning and other presentation features, so compare the converted dialogue and timing before adopting it. Image-based captions are not editable text merely because extraction succeeded. If timing or source recovery is unresolved, stop and use the related workflow instead.
Make the sheet a controlled review artifact
Export a baseline and keep it untouched. Index, Start, End and Source identify the source cue; Target and Notes are the working columns. Preserve timecodes as text and keep multiline dialogue inside its cell. CSV delimiter, quoting and UTF-8 handling must be tested in the team's spreadsheet application. In XLSX, keep the approved table on the first worksheet because that is the sheet the importer reads.
The importer reads edited time cells; it does not enforce an immutable timing lock. A spreadsheet numeric time may be interpreted as milliseconds rather than as an Excel day fraction, so do not allow automatic date/time formatting to rewrite the baseline. Optional AI translation can seed a draft, but map it to the source by meaning and cue boundaries. Keep the original Source unchanged and record human approval of terminology, names and accessibility cues.
Reconcile the round trip before celebrating success
Choose the blank-target policy explicitly. Keeping Source produces untranslated dialogue; skipping removes a cue. A sheet with no translations is rejected, but a partly translated sheet can still import. Rows whose times cannot be parsed are skipped, and accepted rows are sorted by start time. Consequently neither row order nor output count proves correct correspondence.
Compare the rebuilt SRT to the baseline cue by cue using source meaning and start/end pairs. Explain every missing, duplicated or changed entry and inspect equal-start cases. Keep reviewer Notes as a sidecar: they do not become subtitle dialogue. If shorter wording solves reading-speed findings, revise Target and reimport; timing changes require separate approval and a new baseline, not an unnoticed edit to the sheet.
Verify readable playback and a traceable package
Audit the translated result, not only the source. Expansion can make a formerly readable cue too dense. Review line lengths, overlap and chosen local delivery-profile findings with the correct frame rate and the receiver's actual requirements. Profiles are implemented checks, not live platform certification or proof of translation quality. Replay dense and revised scenes.
Embed the approved track with intentional language metadata and a deliberate decision about retaining original subtitles. Container conversion can change presentation; inspect resulting streams and labels rather than assuming the intended track is selected. Open the exact output in the target player, toggle captions and verify display, seeking and sampled anchors.
Deliver the subtitle file, tested video where required, baseline, approved sheet and separate review evidence. Record reviewer, revision, player and unresolved exceptions. New timing repair or stacked bilingual display belongs in the sync workflow; a complete accessibility audit remains a separate responsibility.