That’s a great point, poVoq.
This is also the reason why I recommend GNU gettext so hard (in favor of a random hand-baked system) because it has very widespread support in translation tools, including Weblate. With Gettext, you gain access to so many crucial tools, while with a random hand-baked format it gets much harder.
I’m not sure how to squeeze in Weblate into the essay tho since it’s about mistakes, mostly. I don’t think it’s really a mistake to not use an online service but I agree it's a massive help.
I looked at Dry localization and it's not great. Documentation is very short (only 1 page). Important features seem to be missing: Support for numerus. Any kind of automation to update the translation files when the source strings changed. User-centric tools, etc. Sigh. Maybe I’m overlooking something?
OK, I get Dry is an engine but I don't see how inventing their own translation system did them any good. Because feature-wise, this doesn't come even close to Gettext. If you absolutely have to roll your own translation system, it better be good. This is why I'm fine with Qt's system; this one is actually feature-complete and there is also a reason because it allows for closer integration with Qt. I've worked with the Qt library, so I know what I’m talking about. I see no such rationale for Dry.
I also found this gem in the Dry documentation:
Text UI elements also support automatic translation to avoid manual work.
Well, guess I have to add another point to my essay then: Blindly trusting auto-translation. 😅 (OK, this is not what they suggest, but still)