Finally caught it while running with GST_DEBUG=5. This time, I had paused the track somewhat close to its end. Unfortunately, the logging is on the insane level, so I couldn't copy it all, and I don't even know if it has anything useful. Here is the last part of it, just in case: