The test scene is deliberately boring: a fixed camera, one directional light, one shadow cascade, and enough geometry to keep the GPU busy at 60 fps. We capped the frame rate at 60 and logged frame time, GPU frequency and battery temperature once per second.
On all six phones the first three minutes were stable. Between minute three and minute eight the frame rate fell to a sustained level and stayed there: 60 to 48 on two devices, 60 to 41 on three, and 60 to 34 on the oldest. The skin temperature rise matched the drop within a minute, which is why the throttle curve looks like a step rather than a slope.
The practical conclusion is not “phones are slow”. It is that a 60 fps target measured on a cold phone is a claim about the first three minutes. Benchmarks and reviews should report the sustained number, and games should measure it before choosing a quality tier.