My understanding is it is for when you are sure that the observation cannot be identified any finer than it’s current level based on the information included in the observation.
Eg. 1, there are some flies that can only be ID’d to species visually by their genitalia (and then, sometimes only on the males), and that genitalia cannot be viewed without a microscope. In the absence of those close shots, a genus-level might be the best the ID could get, so the answer to the question “Based on the evidence, can the Community Taxon still be confirmed or improved?” could be no.
Eg 2., I’ve heard there are some animals (cicadas, if memory serves, but it could have been grasshoppers/frogs/birds too) that can only be distinguished to species by their call. If the observation does not include audio, then when it is still at a higher level taxon the “Based on the evidence, can the Community Taxon still be confirmed or improved?” could be no.
Eg 3. I think there are some organisms that I’ve heard can only be positivly ID’d to species by DNA (some fungi, I think). Minus that evidence, it may never get to species, so the answer to the question “Based on the evidence, can the Community Taxon still be confirmed or improved?” could be no.
As Tony Robledo has opined (and I kind of agree), it’s not best used when you think the pic or audio is “too poor” to get an ID, because you never know what an expert may be able to discern from even the blurriest/noisiest photo or most garbled audio recording. I’ve seen iNat Identifiers give some amazing IDs (with explanations!) for features they noticed in some very low quality evidence.