advisor.fairness

Alpha (0.1.0a1), and a mock (Agents): it shows where the FAIR Assessor sits in the process flow, and gives no real assessment.

The FAIR Assessor looks at a dataset that is already published. It serves two steps of the Blueprint's process flow (§8.2):

Step	What the mock does
post.fairness (R12)	Checks the metadata given for the dataset against a short list of FAIR-related presence checks.
post.retrofit (R9)	Suggests what to add to the record for each check not met.

Both steps are optional and advisory. The agent changes no record: the researcher decides what to submit to the repository.

Input and output

PublishedRecord holds the dataset's metadata as the researcher gives it: identifier (a DOI or URL), title, licence, keywords, creators, creator_identifiers (such as ORCID iDs), formats (media types), related_identifiers and access_protocol. With neither an identifier nor a title, the agent abstains.

FairnessReport holds one FairCheck per rule: the FAIR sub-principle (F1, R1.1, …), the check's short identifier, whether it is met, and a recommendation when it is not. It also lists the mock's limitations. There is no score.

What it does

It applies the twelve rules in src/dd_agent_advisor_fairness/checks.yaml to its input: for example, whether the identifier is a DOI (F1), keywords are given (F2), at least one format is open (I1), related identifiers are given (I3), there is a licence (R1.1) and every creator has an identifier (R1.2). Its grounding mode is none, and it cites only its input.

What it does not do

• It is not a FAIRness evaluation, and the Blueprint does not ask the tool to replicate one (R4).
• It retrieves nothing: it does not resolve the identifier or read the repository's record, files or landing page, so it cannot tell whether what was given is true.
• It does not discover related datasets or outputs.
• It changes no record.

The real assessor, drawing on community tools such as F-UJI and discovering related datasets (for example through DataCite's search), is TODO(#361).