For most of the year, we use #copr to repeatedly build thousands of packages.
The spikes mid-June are the moment we switch to the actual Fedora build system - meaning more failures, because there will always be inconsistencies in the state of copr and koji. As you can see, the spikes don't hold us back for too long.

Python 3.15 has been the least disruptive from the three so far (we'll see about that when beta1 is out), looking at the number of packages failing to build. 3.13 was a release of many removals and significant internal changes, which is reflected in our data.

This year, we have added a virtual dependency to packages with extension modules that requires to rebuild them with every new alpha Python version. I stopped seeing segmentation faults in the build logs, except for the genuine issues. This makes me happy, segfaults are scary :)