Looks at each field three ways: what it is called (its name, label or question text, matched on whole words), what it contains (email addresses, phone-like numbers, IP addresses, dates of birth, coordinates, person names) and how it behaves (free text, or text that is distinct on nearly every row). Each finding says which kind of evidence it rests on.
Usage
pii_scan(x, show_values = FALSE, sample = NULL, config = pii_config())Arguments
- x
Files or directories, a data frame, or fields from
pii_read()oras_pii_fields().- show_values
If
FALSE, the default, example values in the result are masked so the report does not itself leak.- sample
The most distinct values examined in one table column; 10,000 unless the configuration says otherwise. A column with more is judged from an evenly spaced subset, and
nthen counts matches within that subset. UseInfto examine every value. Running text is always examined in full.- config
The project's configuration, from
pii_config(): files to ignore, custom patterns and extra terms.
Value
A tibble with one row per finding: source, locator, name,
label, type, evidence (name, value or structure), detector,
n and share of matching values, examples and confidence. type
is the kind of personal data: person_name, phone, email,
address, date_of_birth, location, coordinates, government_id,
financial_account, ip_address, online_id, device_id, uuid,
free_text, unique_text, or a custom pattern's type. detector says
which rule matched, such as idcheck:us_ssn.
Details
In a table column a value pattern must match at least 5 values and 5% of them; in free text and documents a single match is reported.
The scan works on the distinct values of each column and uses the column's type, so numeric, date and logical columns skip the text patterns and a coded column costs little however many rows it has.
The result is a list of things to review. An empty result does not show that the data are safe to share.
Examples
survey <- data.frame(
q1 = paste0("person", 1:20, "@example.org"),
reason = rep(c("work", "family"), 10)
)
pii_scan(survey)
#> # A tibble: 1 × 11
#> source locator name label type evidence detector n share examples
#> <chr> <chr> <chr> <chr> <chr> <chr> <chr> <int> <dbl> <list>
#> 1 data q1 q1 NA email value email 20 1 <chr [3]>
#> # ℹ 1 more variable: confidence <chr>