Hmm, I'm a little surprised it is that expensive: for the Southern San Andreas fault there are so many seismometers that I would have though that it would just be software and communications and a plan to make something work. But, given the disruption of false alarms, maybe it has to be done right.

The longer the warning you want to give for Los Angeles for a NW-propagating rupture, the smaller the fault rupture at which you put out the warning; I think if a rupture started near Bombay Beach, and you warned when it had ruptured 50 km of fault, it might stop as a bit more than a M7 and do no damage in L.A. If the software waited to see if the rupture got through the San Gorgonio structural knot before warning, the time between warning and damage would be shorter.

Same for Cascadia: it does not do a M9 every time, and if you tried to give a 3 minute warning to Seattle from a rupture starting down by Cascadia, the even might only rupture part of the way.

I imagine the warning would be issued anyway.

Chris