hiciefte commented on issue #6719:
URL: https://github.com/apache/jmeter/issues/6719#issuecomment-4892555304

   That workflow is exactly the shape I'd want, and native review staying as 
the final gate is non-negotiable for me too — the pipeline never auto-merges, 
it just opens a PR.
   
   The glossary and style-guide pieces are already how localize-pipeline works: 
it takes a per-language glossary of required term mappings plus terms that must 
never be translated, and enforces them in the prompt, so your `glossary_ru.md` 
would drop in almost directly. It runs against a local model and handles the 
ISO-8859-1/`native2ascii` encoding these bundles use.
   
   On the proofreading point — that's exactly how we run it on Bisq. The AI 
drafts ship as a starting point, but they're not treated as final: they flow 
through Transifex, where native speakers correct and improve them over time 
instead of facing an empty file. So the human review isn't a one-time gate, 
it's continuous, and the machine draft just removes the cold-start problem.
   
   If it's useful, I could prototype it end-to-end on a single language against 
the core `messages` bundle — reusing your Russian glossary — and open a draft 
PR so you can see the output quality before committing to anything.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to