So I suggest you revise your ideas and percentages and realize that products break for a multitude of reasons beyond crappy designs.

I never stated that "crappy designs" were the ONLY cause of failure. I wanted to say that many more failures were due to "crappy designs" than other factors.

You should be able to manufacture something with a MTBF of 300 years i.e. every year one board of 300 will fail.

Think about the lightning that came into my house last June and took out ...
Again, this just proves my point. There is no need to make a "repair manual" well designed units will be maintained by board replacement, not board repair.

Latent stresses will crack things at any point in a product life cycle
Is it really feasible to repair a board with latent stresses, will that board not fail soon again?

Maybe, my verbage has been a little unclear in getting my point across, but here it is agin: A properly designed unit does not need a repair manual, if the unit is properly desined and manufactured any problem should be solved by scrapping the board and replacing it.

The basis for the opinion:
One of the boards I have inherited when it comes back defective is examined for 3 failures that are due to the designers "oversight" (no buffers on output lines, missing decoupling cap on the supervisor chip and a coil with too large resistance). If none of these problems are the cause, the board is scrapped. This gives a scrap rate of about 3% of the failed boards coming in. Unfortunately, with the cost involved, we have to repair the "no buffers on output lines" by replacing the part driving the line the other two problems are permanently taken care of by adding a capacitor and replacing a coil.

Erik