Privacy-Preserving Data Flow Tour

Privacy-Preserving Data Flow Tour

animation
privacy

privacy preserving narrated tour

Play silently with captions on provisional storyboard timing; narration audio is pending. Production pacing is not protocol time. Choose Reduced motion for a complete still at each beat.

Play the tour to open the workbench. Pause, replay a beat, or read the full transcript below. Unsupported processes keep the existing frame and provide captions only.

Full transcript — privacy preserving (8 beats)

B1 — Wearable identity
You are the engineer preparing fitness trends for a wearable service. The wristband collects heart measurements and precise location alongside a person's name. You want useful trends without making that person's behavior easy to trace. How much identity can survive a privacy transformation?

Model caption: A wearable fitness service starts with a name, heart measurement, and precise location.

B2 — Minimize first
Begin with minimization: keep only the detail the purpose needs. For a fitness trend, that means reducing unnecessary personal detail before applying another technique. Pseudonymization means replacing a direct identifier with a substitute label, called a token, while keeping a controlled way to reconnect it.

Model caption: Reduce unnecessary detail first; a pseudonym substitutes a token with a controlled reconnect path.

B3 — Supplied sample comparison
Take the lab's wearable example. The raw sample names a person and gives a heart rate of one hundred eighteen beats per minute. The supplied protected sample has no name. It contains an average of one hundred sixteen beats per minute and a broad location zone.

Model caption: Raw sample: 118 beats per minute. Supplied protected sample: average 116 beats per minute, a location zone, and no name. These are supplied values, not a calculated average from the single raw reading.

B4 — Stable keys and release
The protected sample also retains the band identifier and adds a subject token. At privacy strength three, the token stays stable. The released sample uses a stable joining key and offers a group heart trend with restricted access. Useful analysis remains possible, alongside a way to connect behavior.

Model caption: The band identifier remains. At strength 3, the subject token and release joining key are stable; the restricted output is a group heart trend.

B5 — Utility calculation
For a worked calculation, use wearable data, pseudonymization, and an internal audience, with privacy strength three. The lab starts utility at eighty-eight, adds two for internal use to reach ninety, then subtracts ten for pseudonymization, leaving eighty.

Model caption: Teaching calculation: 88 + 2 = 90; 90 − 10 = 80. The asset displays the final score, not these intermediate steps.

B6 — Final teaching utility
Next, subtract one from strength three to get two. Multiply two by five to get ten, then subtract ten from eighty to get seventy. The selected technique adds zero bonus. Take the smaller of ninety-six and seventy, then the larger of twelve and seventy. The result is seventy percent utility in the lab's teaching model.

Model caption: 3 − 1 = 2; 2 × 5 = 10; 80 − 10 + 0 = 70. Minimum of 96 and 70: 70. Maximum of 12 and 70: 70. Final teaching utility: 70%.

B7 — Token is not anonymity
The misconception is that replacing a name with a token makes the data anonymous. It fails because someone with the reversing key can reconnect the person, and a stable token can connect behavior across datasets. Anonymity aims to make identifying someone again infeasible. Your service has reduced direct identification, while the data remains personal.

Model caption: A token is not anonymity. Someone with the reversing key can reconnect identity; a stable token can link behavior across datasets. This reconnection is explained, not simulated.

B8 — A label still identifies
A token changes the label; it does not erase the person.

Model caption: A token changes the label; it does not erase the person.

Open Privacy-Preserving Data Flow

Back to the chapter