How identifiable is this table?
Load a CSV or spreadsheet, mark the columns an outsider could know about a person, and see k-anonymity, l-diversity, t-closeness and re-identification risk. Then try generalisations and watch k change before you share anything.
Your file is not uploaded. It is read and analysed inside this browser tab and is never sent anywhere. The page only reports, without any data, that the tool was used.
1Data
Loading the synthetic sample…
2Columns
A quasi-identifier is something another dataset, or someone who knows the person, could also hold: age, birth date, postcode, sex, nationality. The sensitive column is what must not be learned. Identifiers such as names are dropped from the export. Suggestions come from column names and types; check them.
| Column | Role |
|---|
3Results
Class sizes
Records grouped by identical quasi-identifier values. A record in a class of one is unique: anyone who knows those values knows which row is theirs.
Re-identification risk
Following ARX. Threshold: a record counts as at risk when its chance of being picked out exceeds
What makes records unique
Counts only. Removing a column, or keeping just a pair, shows which quasi-identifiers do the identifying.
Smallest classes
Sizes and the number of different sensitive values only. The combinations themselves stay hidden unless you ask.
4What if…
Generalise each quasi-identifier and the results update at once. Generalised values are what the export contains.
More options
If the file is the full population and only some rows will be published, choose a column holding 1/yes for published rows. Journalist and marketer risk are then measured against the whole file.
5Export
The file is written in this tab, too. It is a starting point: read what k-anonymity does not guarantee before you call it anonymous.