From a post to DAVIDWEBER.NET forums on 6/27/2011

Default failure mode of a hyper generator

    A default return to normal space in the event of total hyper generator failure is a death sentence only if (1) you are too far from a habitable world at the moment to return to it within the limits of your life support and (2) you are unable to make repairs to the generator after returning to normal-space. In the case of Our Heroes (TM), neither of those factors held true. They were still close enough to Mesa when they dropped out of hyper to have returned there through n-space, and the computers responsible for the go/no-go decision (which had to be made in a very tight time window) knew it. Of course, the crew knew that Mesa was not, in fact, a safe haven to which they could return due to certain, ah, political events there, but the computers didn't. IRT the second point --- repairability --- absent the total destruction of the hyper generator, repair is usually possible. Part of the reason this particular situation was structured as it was was that Eric the Machinist wanted a problem which his characters could fix and came up with a component failure which required a replacement part to be frabricated in the shop. I, ah, did a little redesign of the compensator's internal parts to let it work this way for a couple of reasons (one of which no longer applies, but we won't go into that <G>), but the odds of repairing a compensator are actually usually pretty high in cases of failure, and n-space is a much safer place to be (in most ways) while you work on the repair in question.

    [Had] the ship not been within easy return range of Mesa when the compensator failed, the system would have requested operator input before defaulting to normal-space operation. Under those circumstances, the notion of staying in hyper until you can get close enough to an inhabited system (whether or not it was your original destination) to try to attract the attention of someone else approaching it in hyper would be by far the better choice unless there was some other hugely compelling reason to drop into n-space in the middle of nowhere. In either case, the decision would have to be made in the window between failure onset and complete failure. Failure is usually a cascade process, which is why there is a window at all, but the decision has to be made very rapidly which is why the auto-default is n-space if the ship is in n-space cruising range of an inhabited --- or even just plain habitable --- star system.

    Does that help?