data-productsLineage

Lineage

v0.2’s Layer 5 tracked forks of a document — alternative reasoning about the same question. Data products are also derived from other data products (a subset, an overlap, a lookalike, a model trained on a corpus), and each derivative must carry its ancestors’ usage constraints. v0.3 turns Layer 5 into a derivation graph that records both.

lineage:
  forks: [ ... ]                 # v0.2 semantics, unchanged: alternative reasoning
  derivations:
    - parent_id: "https://..."
      parent_hash: "sha256:..."
      transformation: subset | overlap | lookalike | aggregation | enrichment | training | other
      inherits: full | relaxed
      relaxation_basis: "<ancestor permission_id>"   # REQUIRED if inherits == relaxed
  this_document: { forking_permitted: true, derivation_permitted: none | internal | licensed }

Inheritance rule

A derivative’s effective policy is the intersection of its own usage_policy and every ancestor’s — an activation is allowed only if every policy in the ancestry allows it.

  • The walk descends every derivations edge (a true graph, not just one parent).
  • It stops at the root, or at the nearest ancestor whose attestation tier is verified+ (that ancestor vouches for everything above it), whichever comes first.
  • An unresolved ancestor fails closed (deny).

Licensed relaxation. If an ancestor granted derivation: licensed via the permission named in a relaxed edge’s relaxation_basis, that ancestor is enforced via its derivative_policy instead of its usage_policy. If the relaxation isn’t legitimately backed, the walk falls back to the full policy — the safe direction.

Reference walk

from orp_validator.lineage import resolve_effective_decision
 
# fetch(parent_id) -> the ancestor document, or None if unavailable
res = resolve_effective_decision(doc, {"purpose": "analytics"}, fetch=my_fetch)
res.decision    # "allow" | "deny"
res.denied_by   # which node in the chain denied, if any

Next steps

Usage PolicyUsage Records