As of Tot last week until yesterday a small change to remove unused headers (on top of an inflight CL) could yield flaky layout test results.

Debugging rr recordings of failed and worked runs essentially no Javascript was being run before the layout tree was output, and the testrunner harness was not invoked.  Nor was any significant changed code.

Adding the unused header files back reduced the flakiness to at most 6/1000 (repeat exercise) in the worst case.  Practically this level is not hit with the trybots.  We'll update after landing if it means we have to disable the one affected test.

In the worst case this indicates a race in the preconditions of the test framework.

Reproduction seems like:
remove include platform/heap/{Handle,SelfKeepAlive}.h from WebViewFrameWidget.h


run-webkit-tests --verbose --iterations=1000 virtual/threaded/fast/animationworklet/animation-worklet-visual-update.html

I observed the failure first on trybot linux_trusty_blink_rel.

also 
https://ci.chromium.org/p/chromium/builders/luci.chromium.try/linux_chromium_rel_ng/37432
https://ci.chromium.org/p/chromium/builders/luci.chromium.try/linux_chromium_rel_ng/37248
https://ci.chromium.org/buildbot/tryserver.blink/linux_trusty_blink_rel/24718
https://ci.chromium.org/buildbot/tryserver.chromium.mac/mac_chromium_rel_ng/662690
https://ci.chromium.org/buildbot/tryserver.chromium.win/win7_chromium_rel_ng/113793


I'm going to own this issue to do more investigation and repetition of results, I don't necessarily have the scope to find and fix it.