An argument becomes a decision only when both sides carry a number. On one side, the published operating range for diffusion policies is ten to one hundred serial network evaluations per action. On the other, an embedded field-programmable gate array running simulated bifurcation over two thousand spins has a measured end-to-end latency of about one hundred milliseconds, and the important word is that this figure is independent of problem size, achieved with thirty-two-fold lossless compression of the coupling matrix and an estimator replacing per-problem tuning. One cost grows with the number of steps. The other does not, because there are no steps. So they cross. The only question worth asking is whether they cross inside the range anybody actually operates in, and this lab answers it.