What does real-world differential privacy use instead of literal coin flips, and why?

Real DP systems add noise drawn from mathematical distributions like the Laplace distribution, which spread values over a continuous range for stronger, tunable guarantees.

The coin-flip example is a simplification for intuition. Production DP doesn't use a single binary coin; it samples noise from continuous distributions — most famously the Laplace distribution (and the Gaussian for some variants). These spread the perturbation across a larger range and let you calibrate the privacy guarantee precisely to the query's sensitivity.

The amount of noise is governed by a privacy budget (epsilon), where more noise = more privacy but less accuracy — exactly the trade-off the Laplace mechanism makes tunable.

Tip: "Add Laplace noise scaled to the query's sensitivity divided by epsilon" is the textbook recipe behind the friendly coin-flip story.

Go deeper:

• Differential privacy — Mechanisms (Wikipedia) — the Laplace/Gaussian mechanisms and the sensitivity/epsilon recipe.