OUTPUT-PROVENANCE
When the result is generated, not fetched. The architecture for declaring rights provenance on AI-generated outputs — text synthesis, multimedia generation, federated/aggregated outputs — where the unit is produced at query time and layers are synthesised rather than authored.
The three output classes
Class A — text-based outputs (API responses, chat completions, generated articles). Class B — multimedia outputs (generated images, audio, video). Class C — federated/aggregated outputs (RAG responses incorporating multiple sources, multi-model pipeline outputs). Each class has distinct provenance-block structures and emission mechanisms.
The unit of declaration — the Output-Provenance Block
An OPB is a structured metadata record attached to a generated output declaring: what model produced it, what inputs were used, what rights conditions attach to the inputs, what composition type the output represents, and what provenance chain the output carries. OPBs complement the inline cip.md AI Generation field set (v3.24) which handles the per-asset case.
v3.24 reconciliation — inline cip.md precedence over OPB
Where an asset carries both inline cip.md AI Generation fields and an OPB, the inline fields take precedence per the v3.24 reconciliation rule. This prevents conflicts between operator-declared metadata and runtime-emitted metadata.
Integration with framework features
Integrates with Mixed-Rights Architecture (v3.4) for per-layer rights in generated content, Rights Registry for provenance-chain resolution, Terms Enforcement API for automated compliance checking, and the NILP Architecture for identity-rights in generated outputs.
CIP Output-Provenance Architecture v3.17, https://creativeip.org/output-provenance