If you run the tests multiple times, the "partial restore failures" metric didn't get reset in between.