Here's the failure mode:

0. A new test foo.html is being created/imported. This test has a virtual variant virtual/v/foo.html.
1. Trigger try jobs. foo.html fails, but virtual/v/foo.html passes.
2. Run rebaseline-cl. The tool only sees foo.html is failing, so it only rebaselines foo.html, but not virtual/v/foo.html.
3. A failing generic baseline is created for foo.html.
4. Now run the tests again. virtual/v/foo.html fails because of the failing generic baseline.