There is nothing wrong with metrics, the issue is optimizing for them which is what OKRs are trying to do.

Yeah metrics are fine as long as they're not being manipulated. Then you can be aware of their inherent inaccuracy. Amazon reviews springs to mind. If the only inaccuracies is that people tend to be binary in their choices and either give 1 or 5 star reviews then you can for example adjust by giving more weight to the 4 star reviews. But with the current landscape of all of them being highly manipulated the numbers almost don't even matter anymore.

OKRs are the opposite.

In your example, the "O" would be something like "provide a good experience for app users" and a "KR" would be "improve star ratings from 4.2 to 4.5". The "O" helps keep you focused on what you actually care about, making the metric subservient to that goal.

Only in theory, in practice success is measured by the associated metric. And as constantly demonstrated, anytime you measure people’s performance by some metric that metric becomes gamed.