[potato3732842]

>I wonder on what basis Boeing thought that damage to a load-bearing part could be safely ignored?

Usually this is because the design constraints are complex and in satisfying one you wind up having orders of magnitude more overkill than you need on others.

For example, in situations involving hollow shafts with through shafts or perhaps fluid passages often times you wind up with insanely huge for the load bearing supporting the outer most part because it simply needs to be that big in order to fit around the shaft and have space for reasonable sized roller elements for the speed and realistic race thicknesses, etc. Sure you could go custom, but $$, sure you could use needles or balls, but maybe the stuff on either side has reasons it shouldn't be hard like a race and that might add assembly/construction cost. Now say this overkill bearing is held up by a big web in a big honkin cast housing, because the housing needs to be like that for structural reasons (say it's a specialty pump or maybe this housing is load bearing in the overall assembly, like a tractor's gearbox). Now, say this bearing is in some more complex gearbox that has lubrication windage problems. A valid fix might be to go and cut out a chunk of the web that holds this bearing. Sure it's only supported by 300deg now instead of 360, but it was so overkill to begin with that doesn't matter.

Edit: better example: You can roach dozens of automotive cartridge style wheel bearings without hurting the knuckle it presses into because the knuckle has to be so strong to withstand suspension forces you basically can't apply enough force via the wheel failure to break it and the assembly becomes unserviceable faster than you can get to the point of damaging it by wearing through it.

Edit2: You also need to consider the cost of QA and testing. Sometimes it's cheaper to do a simple overkill waste of material design than something than to do speed holes and engineered webs, etc, etc, because all those features add testing cost as well as manufacturing cost and (especially in ye olden days of the slide rule) make it harder to predict stuff like resonance, exact failure mode, etc, etc and every feature has to be QA'd to some extent. And this all needs to be balanced against expected production volume.