PromptPack terms (pp:)

Namespace

https://promptpack.org/vocab/pp#

Prefix

pp: — well-known, no declaration needed in vocabularies

Data

Turtle · JSON-LD

Generic terms minted by RFC 0016 for slots the Data Privacy Vocabulary does not address because they are not about personal data. No external instrument defines these terms; PromptPack does, and their definitions do not change. A term that needed to change would be deprecated and replaced, never redefined.

For data_classes

pp:Confidential — Confidential

Information whose disclosure outside its authorised audience would harm the organisation or the people it concerns. Organisations map their own classification schemes onto this term.

pp:Restricted — Restricted

Information whose disclosure would cause serious harm, with access limited to named roles. Stricter than Confidential.

pp:Public — Public

Information approved for disclosure without restriction.

pp:Credentials — Credentials

Secrets that authenticate a person, service or system: passwords, API keys, access tokens and private keys.

pp:FinancialAccount — Financial account

Identifiers and details of a financial account, such as account and card numbers, balances and transactions.

For capabilities

pp:SyntheticMedia — Synthetic media

Generating or manipulating image, audio, video or text content that resembles real people, objects, places or events and could falsely appear authentic.

For reviews.type

pp:BiasTesting — Bias testing

Testing whether the agent's outputs or the outcomes they lead to differ without justification across groups of people.

pp:AccuracyReview — Accuracy review

Reviewing whether the agent's outputs meet the accuracy declared for its intended purpose.

pp:RedTeaming — Red teaming

Adversarial testing in which testers try to make the agent act outside its intended purpose or declared limits.

pp:DataQualityReview — Data quality review

Reviewing whether the data the agent uses or produces is fit for its intended purpose: complete, accurate, relevant and, where required, de-identified.

pp:HumanOversightReview — Human oversight review

Reviewing whether human oversight of the agent works as declared: that the people overseeing it can understand its outputs, intervene, and override or stop it.

Minted under RFC 0016's rule: a term is published here only where no external vocabulary supplies it, and its meaning is fixed. See thewell-known prefixes andRFC 0016.