> your argument seems valid...
It's the argument engineers have been making for a couple of hundred years. I first heard it from Dr. Peach in the 1960s. It's correct.
> ...but if that were the case, tightening bolts
> to within a few % of failure would be strongest.
That would work if you knew exactly how strong each bolt was and if you could be certain that the tension in the bolt was exactly proportional to the torque. Because of friction it isn't, and the friction varies.
> Forces are additive.
Forces are vectors. The bolt tension clamping the parts together and the load trying to pull them apart point in opposite directions.
Consider a pair of castings held together by a bolt with 10,000lb tension in it. There's 10,000lb of tension in the bolt and 10,000lb of pressure between the castings. Thus the total force is zero (as it must be or something would move).
Now hook the assembly to a couple of tractors and start pulling on it with 5,000lb of force. You now have 5,000lb tension in the chains trying to pull the castings apart, 5,000lb pressure between them trying to push them apart, and 10,000lb tension in the bolt holding them together. Total: zero. The bolt is still carrying only its preload.
Crank the pull up to 10,000lb. You now have 10,000lb tension in the chains trying to pull the castings apart, zero lb pressure between them trying to push them apart, and 10,000lb tension in the bolt holding them together. Total: zero. The bolt is still carrying only its preload.
Up the force to 15,000 (you're using a couple of Steigers) You now have 15,000lb tension in the chains trying to pull the castings apart, zero lb pressure between them trying to push them apart (there will now be a gap: the joint has failed) , and 15,000lb tension in the bolt holding them together. Total: zero. Only now, when the load on the joint has exceeded the preload, does the tension in the bolt exceed the preload.
This analysis assumes that the castings are totally rigid. There is a small error due to the fact that they are actually compressible, but in most applications this error is negligible because the parts are much larger than the bolt. In some special cases (soft materials, odd shapes) this is not true.
Always torque all bolts to spec. Under torquing leads only to loosening and failure.
Please take a look at the Bolt Science tutorial linked below. Lots of good stuff on that site.
And, to open another can of worms, throw away the split ring lockwashers.
Bolt Science tutorial