Schema validation
Checks whether the authored case follows the PII contract before its observations are evaluated.
Evaluation
Evaluation methods
Choose a domain to understand how its test inputs are authored, varied and evaluated. A method listed here is not a claim that it ran or that a product passed.
Credentials
Methods for secrets, harmless lookalikes and changes to credential-bearing text. Open a card for its procedure and recorded evidence.
Twin
Does the scanner tell a secret from its harmless twin?
View method →Benign
Does the scanner leave harmless look-alikes alone?
View method →Metamorphic
Does a scanner find the same thing when only the surrounding text changes?
View method →Mutation
Does a scanner still find a value that was altered but is still valid?
View method →Differential
Where does another scanner report something different from redact-secret?
View method →Holdout
Did the frozen candidate run its holdout cases?
View method →PII
Methods for personal-data type, location, sensitivity and context. Coverage and measurement are recorded separately for each population.
Type validation
Uses a reference validator to check the authored personal-data type and compares the scanner’s observation with that expectation.
Context discrimination
Checks whether the same kind of personal data is treated according to its authored sensitivity and surrounding context.
Benign controls
Checks authored harmless values that resemble personal data, keeping type and sensitivity expectations separate.
Jurisdiction collisions
Checks values that can match competing identifier formats, against the authored family and jurisdiction expectations.
Mutation
Applies recorded changes to an authored input and evaluates each generated variant against its expected result.
Reference differential
Compares scanner observations with authored reference evidence and validator observations.