Public Sector
Public-sector PET decisions must balance public benefit, confidentiality, legitimacy, accessibility, procurement, and the need for understandable tradeoffs.
Scenario Playbook
| Scenario | Primary PET | Supporting PETs | Why | What can go wrong | What to measure |
|---|---|---|---|---|---|
| Official statistics release | Differential privacy | Post-processing, utility evaluation, public documentation | Formal privacy can support broad public release | Utility disputes, misunderstood epsilon, small-area harm | Error by geography/group, privacy budget, stakeholder impact |
| Agencies compute joint eligibility or outcomes | MPC or clean room | Audit logs, purpose controls, thresholds | Agencies can collaborate without broad raw-data pooling | Governance is unclear, outputs affect rights | Accuracy, appeals process, allowed use, auditability |
| Public-health dashboard | Federated analytics | DP, suppression, grouping | Local agencies can keep source data local | Rare conditions and small geographies leak | Small-cell risk, timeliness, interpretability |
| Open-data exploration | DP synthetic data | Residual-risk labels, utility tests | Public users get a data-like artifact | Synthetic data is treated as ground truth | Utility for intended tasks, memorization, misuse risk |
Use This When
- The public benefit is specific and documented.
- The output can be explained to non-specialists.
- The privacy/utility tradeoff is reviewable.
- The deployment includes public communication and recourse where decisions affect people.
Avoid This When
- The PET hides a policy decision that should be debated publicly.
- The output affects benefits, enforcement, or rights without appeal.
- Utility loss will fall hardest on small communities.
- Accessibility and documentation are treated as optional.
Recommended Starting Stack
For public statistics, start with DP + transparent utility reporting. For interagency computation, start with MPC or a governed clean room, then add thresholds and audit controls.
Failure Modes
- Formal privacy parameters become opaque policy choices.
- Small-area statistics become unreliable without clear communication.
- Synthetic data is reused outside intended purposes.
- Procurement selects a PET tool without a threat model.
- Public trust is damaged because tradeoffs were hidden.
Evaluation Checklist
- Who benefits, who bears utility loss, and who can challenge decisions?
- Are privacy parameters and release rules publicly documented?
- Are small populations analyzed separately?
- Does the system meet accessibility and transparency expectations?
- Are deployment claims independently reviewable?