But right now we know neither (Bob or Steve being bad), and we hide that behind some objective metric that gets diluted at every level. We hide it in Steve's re

But right now we know neither (Bob or Steve being bad), and we hide that behind some objective metric that gets diluted at every level. We hide it in Steve's reviews, in Bob's reviews, in the post-weld reports, in the weld-quality audit report, the inspection. Heck even NASA oversight inspectors are probably downplaying some of the severity when stating it in their findings report (saying it's partially due to, or the welds "contributed" to the quality, etc.) instead of just saying "these guys got an amateur to weld the exterior". So this "Steve is Bad" property gets spread out across all those various things, when we could just as easily have fired Steve because Bob evaluated him. Or maybe Bob didn't state it, maybe the NASA inspector told him "Bob, whoever you got to weld this thing is an amateur. Fire him before you get someone killed." And sure, does that mean Bob might get it wrong? Of course, he might even be a petty little tyrant with an ego or he just might not like Steve because he made a funny comment about Bob's tie on his first day. I don't think some things can scale and be optimized away like we're all a bunch of cogs in a machine with our own individual little tasks and performance metrics that get aggregated into the larger whole.