CS7 — AI/ML Dataset Passport
Setting. A lab curates datasets and model cards that require provenance and governance controls.
Design
Safety declarations in EP; milestone anchors; replicas for rare data; restricted governance for PII-containing subsets.
Policy Profile
CFS‑Dual; DAVS≥0.8; OKB and PQ‑hybrid signing.
Workflow
- Data ingestion → EP → curation receipts → release EP with anchors and catalog entry.
Evidence & Verification
Verifiers confirm lineage of dataset versions and model cards offline; tombstones for removals.
Cost & Operations
ETH frequent; BTC quarterly archival; cloud‑neutral replicas.
Outcomes
- Improved dataset trust and reuse with clear provenance.
Risks & Mitigations
- Linkability risk → redaction/tombstones; high costs → batching; paraphrase → VCE‑v2.
KPIs
- All releases with EP; ≥3 replicas across ≥2 regions; governance flags respected.