Same data, same ε. Per-row LDP (raw / clipped) vs single-shot central DP on the aggregate, plus the integer-bucket pipeline.