Ambient temperature was sampled every 3 minutes throughout the entire benchmark suite, using a 0.1C accurate bluetooth temp sensor; I tried to stabilise the temperature around 24.5-25.5C for the duration of the entire battery of tests (which take 4-7 hours).
Battery temperature was sampled every 10 seconds. The MBA has multiple battery sensors placed at different points. HID reports 6 of them. I used the max value (I modified this code).
For each test, battery temperature was first ramped up to +3.5C above ambient (whatever that is), by running a single-core CPU stress command until the battery temp rises ~5C above ambient then letting it cool down to exactly 3.5C above ambient, then immediately start the test.
After each test apart from the Youtube and Idle tests, the battery was allowed to cool down to exactly +3.5C above ambient.
The battery temperature graphs show the adjusted battery temperature
relative to ambient, to remove the ambient factor from the results.
Specifically, the formula plotted in all graphs is
adjustedBatteryTemp = actualBatteryTemp - actualAmbientTemp + 25
(emulates a constant 25C ambient temperature)
The 2 labeled points on each temperature curve are at the end of the test, and the end of the cooldown. You can notice how each test starts at 28.5C and then cools down to 28.5C after the benchmark finishes. The Youtube and Idle tests don't have any cool down.