The Linux builds should be 0 failures (out of 9000 tests) apart from a few flaky tests that I and others try to monitor and fix the worst of them.
On Windows there are some issues with stability, but there are more issues with which toolchain(s) to test against. Some versions of tools add quotes, some escape, some are posix like (cygwin/msys2), some are Windows like.
I can review the individual failures you are asking about when I am at a computer (writing this on my phone), but you can also just push your change to gerrit and it will run all the tests there (in less than 20 minutes). If your patch isn't ready for review, that's fine, just make a note of that to save a reviewer some time.
As for which tests to run locally, you should run the ones in the same area. I.e., run the dsf tests if you are doing dsf changes, core/ui tests for indexer or editor, etc. (IIRC there is more than one
AutomatedIntegrationSuite.) But if you want to run as the build machine does (I almost never do a full test run locally) then use maven (mvn verify) to build and test.
HTH,
Jonah