I'm using 0, 0.1, 0.2, 0.3, 0.6, 0.95, because at the time I set mine up that made my results kindof in line with Daniels' point scale (see 2nd edition for that), though I'm not sure that is necessarily better.
TRIMP is summed as minutes_in_zone * zone_load_factor, over each zone.
You can find some background on the performance modeling
here and
here.