Finite-mode sampling limits how well low-ℓ observables can be measured even with perfect detectors; ΔC_ℓ/C_ℓ = √(2/(2ℓ+1)).