and provides no information about the spread of the virus or more importantly how it is impacting hospitalizations (the real thing that matters).

When testing first started, because they were hard to come by, only those that had significant symptoms or were famous were tested. By definition, the % positive would be on the high end as the symptoms were fairly well understood and the decision tree on whether to test in many ways drove a high rate of positive.

As testing became more available, more people with lesser symptoms and many who had been in contact with those with symptoms were tested. By the nature of this approach, the % positive should have gone done because your pool of those being tested weren't being screened as much as to be likely to be positive. Thus, the % positive going down has nothing to do with the virus spread and everything to do with the makeup of the denominator in the formula.

Now we are in a situation where one of the major sources of negative tests (people that had symptoms but had the flu) is pretty much gone in most of the country. That means the likelihood that someone with the defined symptoms is positive is up again, but again because of the makeup of the denominator, not anything to do with the virus itself.

Simply put, your pool of those being tested is not controlled and has bias in it that has changed over time, thus your metric is not effective.