After 102 entries, we stopped managing translation backlog and made bilingual content the default

Situation

The portfolio Worklog had grown to 102 Hungarian entries while the English interface still inherited the original Hungarian content. Once we decided to publish real English versions, several problems surfaced at the same time: the engineering meaning had to remain intact, translations had to follow the exact chronological display order, the search index had to be localized as well, and an older broad text-replacement approach could even create mixed HU/EN artifacts inside entry text. At that point the task was no longer simple translation; it had become a content-architecture and quality-assurance problem.

Approach

We introduced a dedicated English content source keyed by the same slugs, and the generator now uses it for English titles, summaries, full entry bodies and searchable text. Translation moved through reviewed chronological batches of ten, while a dedicated QA guard checked that there were no gaps, no out-of-sequence entries, no required fields left identical to Hungarian, no translated text lost during generation, and no stale Hungarian content left in the English search index. Once the Worklog reached 102/102, the more important decision was to stop treating this as a one-off migration: every new entry now lands with HU and EN source content together.

Outcome

The Worklog went to production with 102/102 genuine English entries, bilingual search behaviour and a dedicated localization quality gate. More importantly, the future translation-backlog model disappeared. The next entry does not create a temporary 103/102 state that somebody has to clean up later; it is added as 103/103 from the start. Language quality is now part of the same feature definition as the new content itself, rather than a separate maintenance project scheduled for later.