when you're forecasting past a week out, using ensembles and focusing on the broader pattern to point out timeframes of interest is generally the way to go. this will be taught in every forecasting classroom on the planet. now that we're a few days later on, we can see that there are two distinct waves rather than a single trough (which, by the way, some OP runs had)
this is the way long range forecasting is... the data change sometimes. it doesn't make the process any less valid