What is the idea?

#4700 (comment)

Then, instead of scrolling up dozens of pages to find a test failure, it is immediately highlighted on the report.

We generate allure data (available when a test run has completed, not debuggable in the middle like the github native scrollbacks). Currently someone has to manually run the "update reports" action on conda-build allure. But, once that is done, it is easy to click on the highlighted test failure instead of trying to scroll back up to the second-to-last FAILED in an extremely huge pytest capture.