...or maybe it's on the zuul level? I'm not quite sure. In any case,
https://integration.wikimedia.org/ci/job/pywikibot-core-tox-nose/3002/parameters/
https://integration.wikimedia.org/ci/job/pywikibot-core-tox-nose/3003/parameters/
are for the same PS, but one is a merge while the other is the commit from the PS. They give different test results (but this seems to be an independent issue)
When a patchset is received, Zuul attempts to merge it against the tip of the branch and craft a reference for that merge commit which is later reused by Jenkins.
Even if the tip of the branch has not changed, Zuul crafts a new merge commit each time the patchset is triggered. I verified in the Zuul git repo for pywikibot/core and both patches are the exact same code, merges of 5d49845 - Add Live RC Pages generator on top of the current master at the time f3b4cc5 [IMPROV] Site: Warn more detailed if site to old.
The test failure must be due to tox / some dependency issue: T85662.
Yes, you're right. I would have sworn f77098 was the original commit, but it's indeed another merge commit.