My first batch was 33 Astra strings. That was enough to show me that software translation is slower than translating an ordinary paragraph.

Some entries were straightforward buttons or settings. Others contained variables, HTML, or almost no context. I had to check what could be translated and what had to remain untouched. I skipped the worst examples and concentrated on user-facing strings I could understand.

The first batch also exposed consistency problems. I had to pay more attention to formal “usted,” sentence case, and the Costa Rica glossary before increasing the count.