Long before any incidents, Boeing estimated a smoke event as once in 10 million hours. The measured rate of fire events is much higher than Boeing's pre-certification estimate of the rate of smoke events. There is absolutely no implication that Boeing's definition of smoke event doesn't include fires.
Edit: as mentioned below, the real problem seems to be that you're assuming smoke implies fire, when smoke merely suggests fire.
> During the 787 certification process, Boeing studied possible failures that could occur within the battery. Those assessments included the likelihood of particular types of failures occurring, as well as the effects they could have on the battery. In tests to validate these assessments, Boeing found no evidence of cell-to-cell propagation or fire, both of which occurred in the JAL event. The NTSB learned that as part of the risk assessment Boeing conducted during the certification process, it determined that the likelihood of a smoke emission event from a 787 battery would occur less than once in every 10 million flight hours.
So, to summarize, in the certification process, Boeing claimed:
a: no evidence of a fire.
b: a 10-million-hour estimate for a "smoke event".
My original question was merely from a statistical/technical perspective: how is this figure calculated if they have no evidence of the event they are supposed to measure? From my understanding, automobiles undergoing similar certification are tested under abnormally rigorous conditions, and the chances of those conditions happening are taken into account when calculating failure rates. But if Boeing claims it had no evidence of a fire happening at all, what is the reasoning behind the number? Is it just a convenient way to say probability=0?
http://en.wikipedia.org/wiki/Fault_tree_analysis
It sounds like one of their incoming assumptions--a probability, a manufacturing spec, something like that--didn't match up with reality. That happens. They're probably going over the fault tree, madly trying to figure out which one and by how much.
Boeing counted a malfunction as a "smoke event" which is a release of emissions from a battery overheating, not necessarily a fire. The no evidence of a fire is a claim that regardless of a battery failure, there is no plausible failure chain that leads to a fire beginning in the battery.
>>But if Boeing claims it had no evidence of a fire happening at all, what is the reasoning behind the number? Is it just a convenient way to say probability=0?
In the case of a fire, yes. But this just means that the battery fault alone won't cause a fire. If other systems fail all bets are off. I suspect you are conflating smoke event with a fire.
>>My original question was merely from a statistical/technical perspective: how is this figure calculated if they have no evidence of the event they are supposed to measure? From my understanding, automobiles undergoing similar certification are tested under abnormally rigorous conditions, and the chances of those conditions happening are taken into account when calculating failure rates.
As I explained above, Boeing claimed that it was possible that a battery could malfunction(one in ten million hour), but they had no plausible scenario that could lead to a fire. In this case, battery malfunctioning does not necessarily equal a fire. The smoke event is calculated from a number of probabilistic factors such as the confluence of the possibility of electrical overcharging and arcing and simple faults with the battery. It is a made up number that doesn't take into account that this is a new design that introduces other possibilities of failure.
If smoke doesn't imply fire then P(smoke) > 0 doesn't imply P(fire) > 0.