Privacy Lab
DCR/NNDR analysis and EDPB Three-Criteria Test results
⚠ Proxy Metric Disclaimer
DCR and NNDR are distance-based proxy metrics. Peer-reviewed research (Yao et al., ESORICS 2025) shows records passing these tests can still be vulnerable to membership-inference attacks. Treat as heuristics alongside the (ε,δ)-DP guarantee and the EDPB three-criteria test — never as a standalone “anonymous” verdict.
DCR Distribution
Distance to Closest Record — red zone (0–0.05) = clone risk, green (>0.10) = safe.
No synthesis jobs yet. Generate a cohort.
NNDR Distribution
Nearest Neighbour Distance Ratio — bins below 0.5 flagged for DPO review. Acceptable: <1% flagged.
EDPB Three-Criteria Test
Under EDPB Guidelines 02/2026, data is only considered legally anonymous if it passes three rigorous tests: No Record Isolation (singling-out), No Linkage (combining with external datasets), and No Inference (deducing sensitive attributes). This panel executes simulated adversarial attacks to empirically prove your synthetic dataset passes all three criteria.
EDPB Three-Criteria Test
No Singling-OutMitigated
No LinkageMitigated
No InferenceMitigated
EDPB Opinion 28/2024, para 40
Three-criteria are referenced in EDPB Op. 28/2024 para 40 (citing WP29 Opinion 05/2014, p.24). They form a two-prong alternative test, not an absolute bar — so the UI never presents “one criterion failed = not anonymous”.