Real cases
A real case · you decide
Differ and Laplace and the school survey
Differ & Laplace — a 17-year-old publishing results from a school wellbeing survey
The club will publish how many students in each class answered yes to a sensitive question. Someone could compare this year's count with a version published after one student left.
Differ is the attacker: count-with-Alice minus count-without-Alice reveals Alice's answer.
Laplace is the noise-maker: a random amount, usually small and sometimes larger, of scale 1/ε.
How should the club publish the counts?
Privacy is not about hiding names. It is about what can be worked out.
A real case · you decide
Budget and Epsilon count the spending
Budget & Epsilon — a 16-year-old answering questions about the survey all term
The club has set a privacy budget of ε = 1 for the term. Each published count uses ε = 0.1. By November they have answered 10 queries.
Budget is the accountant: 10 × 0.1 = 1.0.
Epsilon asks the trade-off question each time: how private, how accurate, and who could be re-identified?
A teacher asks for one more count. What should the club say?
Every answer leaks a little. The budget decides how little.