https://chromium.googlesource.com/chromium/src/+/master/docs/testing/json_test_results_format.md says that "is_unexpected"

> If present and true, the failure was unexpected (a regression). If false (or if the key is not present at all), the failure was expected and will be ignored.

It doesn't cover a bunch of edge cases.

1. Unexpected passes currently also have is_unexpected set to true, which I think is sensible and should be documented.

2. Things are more complicated when we retry. For example, if a test is expected to pass but has a 'FAIL PASS' actual result, should is_unexpected be true? In  issue 822078 , I looked at recipe_modules' util.py, especially the logics related to "unexpected flake", and thought is_unexpected would be true in this case, but apparently that's not the case in blinkpy:
https://cs.chromium.org/chromium/src/third_party/blink/tools/blinkpy/web_tests/models/test_run_results.py?l=279&rcl=cb6928930fc1fb8e29aebc6dd24af31367719bfb

In layout tests, is_unexpected is set to true when *none* of the results is expected. The document is unclear about this edge case, but I think we should set is_unexpected to true when *any* result is unexpected, so that we can conveniently identify unexpected flakes, which will be of great help for FindIt.

Based on my understanding of the infra, my proposed change will make the webkit_layout_tests step to be orange more often on the bots because unexpected flakes are displayed as warnings, but won't fail the build. This will also make it easier for people to spot bad (flaky) layout tests.