Each test is classified once per day, over the runs it had that day:
Flaky: it failed at least once — a failure, a timeout or a crash.
Skipped: it did not fail, and was skipped on at least one configuration.
Stable: it did not fail and was not skipped anywhere — 100% pass.
In the charts above, failing beats being skipped, so a test that failed on Linux
and is disabled on Windows counts once, as flaky. The table counts it in
both columns — “how much is skipped here” is a question
about skipping, and answering it with only the tests that never failed hides most
of them. So the table’s Flaky and Skipped columns overlap and do not sum to
Tests. Tests annotated run-if are explicitly scoped to
another platform, so not running elsewhere is not a skip. A test that did not run
at all on a given day is left out of that day entirely. Opening a folder lists
only its flaky and skipped tests — the ones that passed everywhere are
counted in the percentages but not listed, since there is nothing to do about
them.