Two asymptotic results for the shuffle model with a fixed finite-output local randomizer and canonical binary neighboring datasets. (1) A universal expansion JSD = χ²/(8n) − μ₃/(16n²) + 7χ⁴/(64n²) + O(n⁻³). (2) Asymptotic μ-GDP equivalence with μ = √(χ²/n) via Berry–Esseen bounds under both hypotheses, yielding an explicit (ε,δ) curve 6–9× tighter than the FMT amplification bound.
Alex Shvets (2026) studied this question.