Walk through the four steps that turn a raw table into a k-anonymous dataset.

Start with the full table → remove identifiers → generalize quasi-identifiers → suppress what's left until every QID combination is shared by at least k records.

* The k-anonymity pipeline: cheap moves first, suppression last, until classes of size k form. *

Using a beer-rating dataset (Name, Gender, Age, Beer Rating) as the example:

1. Original dataset — all attributes visible; classify them (Name = identifying, Gender/Age = QID, Beer Rating = sensitive).
2. Remove names — the identifier is gone, but the data is still identifiable: a 32-year-old male with rating 8 may be unique.
3. Generalize age — replace exact ages with ranges ([30–35[, [25–30[…). Better, but Gender + Age Range can still single out small groups.
4. Suppress gender to * — now each age range holds at least 3 indistinguishable records, forming equivalence classes. 3-anonymity achieved.

The progression shows the philosophy: you keep coarsening quasi-identifiers (generalize, then suppress) until everyone hides in a crowd of size ≥ k.

Tip: Notice the order — cheap moves first (remove, generalize), heavy hammer (suppress) last, because each step costs utility.

Go deeper:

• k-Anonymity (Sweeney, 2002) — generalization + suppression to equivalence classes.
• k-anonymity (Wikipedia) — worked walkthrough of the same process.