I've found the near term forecast a bit flaky in New York as well, though I've always imagined the city to be particularly unpredictable. I still use it regularly, but I hope they'll devote some of their new resources to improvements to the core forecast algorithm.
It would be useful to have some insight into the underlying data, it's freshness, and an easy way to report inaccuracy (I've sent support emails but they've gone unanswered -- I assume the support load is enormous for a small team).