I don’t think size should matter at all. Preventing ambiguity is the driving factor for splitting up domains. You’d want to split appart where one an the same string might need to be translated differently.

I agree with the above. As an example, we’re building an application with two separate front-ends, hosted on two different subdomains for different users. But they share the same core (even go to the same context files and use the same ecto schema’s). The same error might be translated differently, as the message is presented to two different populations.

One “mechanical” separation we do is keeping enum translations in a separate domain. These can’t be extracted, and are translated with Gettext.dgettext/4.

But I wouldn’t overthink it. Being prepared by avoiding literal strings in the front-end is 99% of the work. And since it’s so easy to do when starting fresh, and such a pain to add later, I have the tendency to always go through gettext for messages in the front-end, even if only one language is used. Maybe I’m biased living in a country that has three official languages for 10M people, where language is always an issue