Use Cases

Use cases translate PET choices into domain constraints. The same PET can be wise in one domain and irresponsible in another.

Use these pages when you know the domain problem but do not yet know which PET stack is defensible.

Playbook Map

Domain	Common decision	Strong starting point	Watch first
Healthcare	Multi-site research, quality metrics, cohort matching	FL, federated analytics, PSI, DP synthetic data	Rare cohorts, clinical utility, local validation
Finance	Fraud, AML, risk, and compliance collaboration	PSI, MPC, federated analytics, FL	Adversarial participants, entity resolution, output misuse
Advertising	Measurement, attribution, audience overlap	Clean rooms, PSI, DP, attribution APIs	Small segments, repeated queries, platform incentives
Public Sector	Official statistics, cross-agency analytics, public-interest releases	DP, MPC, clean rooms, synthetic data	Legitimacy, accessibility, utility disputes
AI	Private RAG, inference, fine-tuning, evaluation	TEEs, HE, DP, FL, access controls	Memorization, prompt/log leakage, output leakage

How To Use A Playbook

1. Pick the scenario closest to your real decision.
2. Check the recommended PET stack and the "when not to use it" section.
3. Write down what the output is allowed to reveal.
4. Measure the listed privacy, utility, cost, and operational metrics.
5. Move to PET Compass or PET Patterns for deeper design guidance.

Cross-Domain Warning Signs

• The PET protects inputs but the output reveals the sensitive fact.
• The domain problem is governance, incentives, or legal authority rather than computation.
• The benchmark uses clean data but the deployment has missingness, drift, or inconsistent definitions.
• The privacy claim ignores logs, prompts, model weights, embeddings, or repeated queries.
• The proposal says "privacy-preserving" without naming an adversary.