Supposing that the analysis tool works correctly from a technical perspective, I keep wondering where these thresholds actually come from. What scientific foundations justify them?
Even if you analyze a large collection of codebases and compute a mean, median, or some other derived value, why should code that fails to meet such an arbitrary number automatically be classified as poor quality?
In my subjective opinion, the idea that an automated analysis can simply output an OK or NACK verdict is… well, I cannot write that word here.
I think they're just randomly chosen as "we need to indicate at some point that our analysis is failing", and 50% sounded good enough. Perhaps more investigation would reveal better thresholds, but that would cost money and time for dubious benefit.
Also, to my reading, the warning is that the analysis itself is low quality, not that the code is. The only reason it suggests remediation is because otherwise there may be issues that fly under the radar, due to the incomplete analysis.