ZIndi was not allowing me to post that much text and images here so here is google docs on whatever issues I found.
I agree with your arguments. It would have been ideal to have timeseries data, even for a few dates within the season.
Great job going deeper into analyzing the. However, your conclusion that data is (to paraphraise) useless is directly conflicting with some top solutions that got 1.1 scores. In other words, there are features / tricks that other teams did successfully find to get accuracy that high. The beauty of these challenges is that there are so many "out of the box" approaches, with some more fruitful than others (pun intended)