FAIR notes
How this repository implements each of the four FAIR principles, and where F(AI²)R adds the second AI pass.
Findable
- A persistent identifier will be minted via Zenodo on first archival
release (
.zenodo.jsonis the metadata source). - Repo-internal identifiers: every
fair2r:Claimgets an IRI under thehttps://noheton.org/f-ai-r/claim/namespace. - Machine-readable surface:
CITATION.cff(humans + GitHub),codemeta.json(CodeMeta tooling),.zenodo.json(DataCite via Zenodo).
Accessible
- Source under MIT (
LICENSE); prose under CC-BY-4.0 (LICENSE-paper). - Public Git repository —
git cloneis the access protocol. - Sources cited that are themselves open-licensed are vendored under
doc/sources/. Closed sources are referenced by DOI only and flagged in the FAIR audit.
Interoperable
- Provenance in PROV-O plus DCTERMS plus
FOAF, serialised as Turtle (
doc/provenance.ttl). - Manuscript in LaTeX (
paper/); a portable PDF is the build output. - Bibliography in BibTeX with DOIs whenever available.
Reusable
- All artefacts versioned in Git with a clean commit history.
- Each agent prompt is a versioned file in
agents/, so the process is reusable, not just the output. - The Makefile is the canonical build entry point; CI (when added) calls the same targets.
The F(AI²)R extension
| Principle | Author pass (AI₁) | Auditor pass (AI₂) |
|---|---|---|
| Findable | Drafts metadata, suggests keywords. | Verifies metadata is consistent across CFF / CodeMeta / Zenodo. |
| Accessible | Vendors sources, generates README. | Verifies licences, broken links, missing files. |
| Interoperable | Emits PROV-O triples for new claims. | Validates Turtle, runs SHACL shapes if defined. |
| Reusable | Documents prompts, decisions. | Replays the pipeline on a clean clone and reports drift. |