Good question. One idea might be going through the EA funds grants and, where possible, make a preciction on a testable outcome of the project. For example whether a workshop will happen and have more than X attendees; or whether a paper will be published by the researcher on that topic within 2 years. Unfortunately the grant reports might be too little information to do meaningful predictions.

I don't have much to add, other than letting you know you're not alone in looking for this. I started doing a similar thing a few weeks ago. A couple more "sandboxes" you could add are GlobalGiving and Kickstarter. It's a bit difficult to find the projects that an EA might be looking for on Kickstarter, but the "evaluation exercise" that you're trying to do could be good practice even there (e.g. trying to determine the potential positive impact of this app that aims to build a habit of not touching your face)

Unfortunately I can't think of a service that would help with what you're looking for! A potential alternative way to improve your skills could be to identify promising options, post them on the Forum, and see what feedback you get?

DonorsChoose (recommended by an Experiment friend). I haven’t seen yet whether these projects do report back, but the infrastructure is there for following up about whether the project went as expected.