Then according to the graph, 59/85 were awoken from sleep, which is about 71%.

59 awoken = 74 (awake at quake) - 15 (awake before the quake)

85 = 100 - 15 (awake before quake)

Maybe they are adjusting for other things that are not shown on the graph.

Good point. I wonder if they are smoothing the data, which would make the peaks less peaky (which brings up another question, what is the time resolution for UP's ability to measure wakefulness?).