I started a task of climateprediction.net, and after it ran for about 50 hours it started a second task on its own. After a while, BOINC stopped running the non-climateprediction.net task. It ran each of those tasks for another 100 or 200 hours, and eventually, one of the tasks just disappeared off of the task manager (it was nowhere near completion). Now the second task is stuck at 26.61% and one timestep, but claims to still be running. So basically I started one task, another started itself, and now one is gone and the other seems broken. What should I do?

Your computer has 2 processors. You're attached to 2 projects.
BOINC is designed to get work from all projects, such that the 2 processors are kept busy.

It's most likely that one of the processors was crunching work for ABC@home, and when that finished, BOINC had to get another work unit. So it chose a 2nd climate model.
All perfectly normal behaviour.

The task that disappeared has crashed, as you can see by looking at your list of models on the project's web site.
If you select the Projects tab of the BOINC manager, and click on climateprediction, you'll get a series of buttons on the left.

Click the one labelled Your account, and that's where you'll end up. Then, if you click on computers, you get a list of your computers.
Click on one, and you'll get a choice of looking at the details of the computer, or getting a list of your models.
This list contains a summary of them, and clicking on the number of one in the first column, will give you details of that model.
Further details on how the model faired during it's run can be found by clicking on the + along side stderr.

As for the stuck model, that particular type of work doesn't like being interrupted around the point at which it creates the zip files.
This occurs each 25% of the way through.
So the best thing to do is to Abort the one that's stuck, and start a new model. Those currently available aren't as touchy.

---

Backups: Here