Capture comparable builds
Record the build, graphics settings, resolution, starting position, and route. Run the same sequence after a change to lighting, geometry, or effects. Compare the slowest stretches of the trace and inspect the frames where a sudden load or effect coincides with a hitch.
Turn a hitch into a reproduction
Tag the time and location of a spike, then shorten the test to the scene transition, first encounter, or effect that triggers it. Keep both a cold run and a repeat run. A difference between them can help a team separate initial loading from sustained rendering pressure.