Removing asterisks before unattested terms
Reconstructions cannot be in a “wrong script”. Your script should not put terms starting with * into Category:Terms using script detection fallback.
Of course they can. Writing reconstructed Armenian in Ogham would make no sense. Also, not all reconstructed terms start with *, because that * can be embedded into a link, like: {{l|la|[[egō]] [[*cominitiō]]}} > egō *cominitiō
But writing reconstructed Middle/Old Iranian in Latin script does make sense. In fact, it's the only way it is done.
I suppose we could add a rule that reconstructions always have Latin as an additional possible script. But whether something is a reconstruction is not always that clear, like I showed. What should be done in mixed cases like the above? We can probably rule out mixed script, like if someone had written the "ego" in a native script but "cominitio" in Latin script. We don't support script mixing like that for any term, attested or not, so that is not a problem. But should we allow Latin as a possible script even when one of the terms in the link is attested but the other is reconstructed? Personally, I don't think it should. So that would keep the detection relatively simple: if the whole term begins with * then it's all reconstructed, and then you can add Latin, otherwise not. How is that?
That sounds good. I don't think we should complicate detection to accommodate rare outlying cases.
Ok. But we should still at least reason why those outlying cases "don't work", not just because it's too complicated. These modules are used in too many places to leave any undefined "gaps". My own reasoning is that if a link contains attested terms, then those terms can't be in Latin script unless the language allows that anyway, because such entries would only be allowed to exist if they're in one of the language's native scripts, and of course a link only makes sense if the entry may exist at some point. So the exception shouldn't apply to mixed links.