Cooperative Bug Isolation for Fedora 12

The tools capture information about all runs, so that they can compare "good" and "bad" (crashing) runs. The interesting tech is the instrumentation -- their trick is sparse stochastic tracing of whole-program execution, where any given trace is small, cheap, and doesn't reveal much (good for privacy), but in aggregate you can automatically isolate the necessary and sufficient conditions for triggering a crash, even if the root cause is some event that occurred in the past and isn't visible in the core dump/back-traces that most tools collect.

It's an academic project, so there's tons of write-up available: http://www.cs.wisc.edu/cbi/learn-more/