Memory Provenance — References

The standard vocabulary for recording derivation, and the obligations that attach to being able to explain and correct what a system holds.

Standard representations of derivation

• W3C PROV-DM: The PROV Data Model — entities, activities, and agents, and the relations between them: the vocabulary for stating that one item was derived from another by a particular process attributed to a particular actor.
• W3C PROV-O: The PROV Ontology — the OWL encoding of the same model, for systems that need to exchange lineage rather than keep it internal.
• W3C PROV-Overview — the entry point explaining which parts of the family are relevant to which problems, which is worth reading before adopting any of it.
• Provenance: An Introduction to PROV — Luc Moreau and Paul Groth, Morgan & Claypool, 2013. A practical treatment of modelling and querying provenance, including the reverse-traversal questions that retraction depends on. Cited by name.

Why the origin of a claim is a regulated concern

• Regulation (EU) 2016/679 (GDPR) — Article 16 on rectification, Article 15 on the right to know what is held and where it came from, and Article 22 on decisions based on automated processing. Each requires being able to say how a held claim was produced.
• NIST AI Risk Management Framework (AI RMF 1.0) — the documentation and traceability expectations that make an automated system's outputs explainable after the fact.

Why derived claims drift from their sources

• Remembering: A Study in Experimental and Social Psychology — Frederic C. Bartlett, Cambridge University Press, 1932. Successive retellings become shorter, more coherent, and more confident than the original, which is why a derivation chain needs a depth bound and a path back to primary material. Cited by name.

No verified reference is recorded for producer-version stamping on memory writes, or for provenance retention after a source has been erased. Both are engineering practice without published study.