Same data, same ε. Two ways to estimate the histogram: perturb the scalar value then correct, or release the one-hot vector + noise and sum directly.