The two most common causes for breaking tests in official only are probably:

1. Enabling some feature depending on finch flags, assuming in a test that the flags are a certain way, but they are different in official. (e.g.  issue 836261  issue 850917 issue 844384;  issue 838543  also somewhat similar)

2. EXPECT_DEATH checking for a death string, while CHECK << string strings get stripped in official builds ( issue 771998   issue 644768   issue 612503  etc)

There are a couple of related issues here.

First, I actually want sheriffs to be monitoring the internal desktop (official) builds, though they don't do so today and we don't yet have an actual plan for changing that. I think that if we were doing that, we wouldn't actually need the chromium.chrome waterfall anymore, we'd just use official.desktop.continuous.

Second, I agree that we should be running tests on some config that *is* being monitored by sheriffs to catch as many of the official build breakages as possible.

A complication to both of these points is that the results of the chromium.chrome builds (which builds are green) are currently used by the release scripts to pick revisions for the branches. We intentionally (at least at this point) do *not* want to gate branching criteria on whether all of the tests passed or not, because often the test failures are not severe enough to delay the branch. 

So, if we treated chromium.chrome as any other builder and just started adding tests, that might have side effects we didn't want. I *think* that if we did a builder/tester split for the configs, then things would probably be better. Alternatively, we could look at modifying the release scripts to know about test failures, or we could decide that we really *did* want the branches to not use revisions that had failing tests.

+mmoss, jbudorick because we've discussed this off and on as part of the luci migration.

Running the tests on a separate tester makes sense to me.

This issue has an owner, a component and a priority, but is still listed as untriaged or unconfirmed. By definition, this bug is triaged. Changing status to "assigned". Please reach out to me if you disagree with how I've done this.