Looks like a flaky failure in Firefox:

 1:07.29 INFO |                               Test                              | Subtest |          Results           | Messages |
 1:07.29 INFO |-----------------------------------------------------------------|---------|----------------------------|----------|
 1:07.29 INFO | `/css/css-scroll-anchoring/nested-overflow-subtree-layout.html` |         | **FAIL: 2/10, PASS: 8/10** |          |

 1:05.68 TEST_END: FAIL, expected PASS - Testing http://web-platform.test:8000/css/css-scroll-anchoring/nested-overflow-subtree-layout.html == http://web-platform.test:8000/css/css-scroll-anchoring/nested-overflow-subtree-layout-ref.html
Found 7290 pixels different, maximum difference per channel 255

I'm a little surprised not to see the screenshot dumped into the logs (do we not do that for the stability checks?), so I'm going to run this locally on Firefox and see if I can reproduce before pinging the Chromium author.

Stability checks eat most of the logs because the log size was occasionally massive (when we were on Travis it would blow through the 50Mb limit). We could make screenshots come through, but some work would be required.

Ah, it looks like the class includes reftest-wait.js, but is missing any use of the reftest-wait class. That's probably the problem; I'll add that and see if it passes stability.

That appeared to have worked (stability checks now pass, CL was merged automatically).

FYI @chrishtr - I had to make a slight modification to your patch, see 904b1cf. This should require no action from you; the next WPT import will just import the diff and update Chromium accordingly. Please just take a quick look to make sure my change is sane, thanks.

Looks good.