This is not a duplicate of T89213. The lemma should always be displayed. It's a bug that it depends on successful language fallback, because it shouldn't depend on language fallback in the first place.

For the record, the decision to only show the sense ID when no gloss is available was made in T200983#4518320 and recently reaffirmed in T207392#8492626.

As a total guess, I think the reason for it might be that showing the lemma(s) alone (idea #1 in the task description) would make the sense look like a lexeme and cause confusion? (But if that’s true, I think ideas #2 or #3 would alleviate that.)

I was about to report a bug about links to senses only showing the ID on property pages and then I realised it's this issue yet again. I am seriously tempted to start adding placeholder glosses for English as a workaround for this annoying, confusing design choice.

It would be great if the links simply showed gloss(es) in the language of the lemma if available. As a matter of practicality, I am not in the habit of glossing senses in languages other than the language of the lexeme—each gloss takes a certain amount of time to add, and if I want to minimize the amount of time I am spending writing glosses this is optimal. I don't need to see every gloss in my interface language and don't expect to.

I am here just to second the aforementioned concerns and would support a format that merely displays any gloss, rather than give up, if neither my interface language nor any in its fallback chain are present on a sense.