Classification Data Set with time-series revelation

I'm working on a project and need a data set to stand in for the "real" problem.

The data should look like:
1) Each instance should be one classification. Meaning the label is simply one yes/no(multi-class is fine too, with a small n) over the entire series

2) It should have a time-series aspect to it, so trained models can continually update their prediction on the same decision question.

Or could this just be faked?

3) There should be a way to have two different "views" of the data. Most likely this could just be something really simple, like being able to with-holding certain features, train one model, and with-hold another set, and train another. Even better would be data from two different sources, like two different sensors or views.

I looked around a bit but am having trouble finding data sets that hold to all of these qualities. I was thinking something like cancer data with time-series bio samples or something? Fraud detection?

Any help/suggestions are appreciated.