DOIs vs Blockchain Ledgers for LLM Training Ingestion

This field note documents the functional differences between Digital Object Identifiers (DOIs) and blockchain ledger entries as provenance and attribution mechanisms, evaluated through the lens of large language model training ingestion and corpus inclusion probability. The comparison examines structural and systemic properties including resolver infrastructure, metadata schema requirements, governance models, trust environment filtering, and longevity guarantees. External references include the Semantic Scholar Open Research Corpus (S2ORC), major open training datasets (Common Corpus, Dolma, RedPajama), the DataCite Metadata Schema 4.6, and published research on Ethereum timestamp accuracy limitations. The analysis concludes that DOIs are structurally aligned with how training corpora are assembled and filtered, while blockchain entries provide complementary cryptographic proof but lack the metadata, resolver, and governance properties that influence corpus inclusion. The optimal provenance strategy combines DOI publication with optional blockchain hash anchoring. This work extends the AI Visibility Canonical Definition (https://doi.org/10.5281/zenodo.18395772) and the AI Visibility Upstream Ingestion Conditions Theorem (https://doi.org/10.5281/zenodo.18475454). Classification: Structural Comparison Canonical page: https://josephmas.com/ai-visibility-field-notes/dois-vs-blockchain-ledgers-for-llm-training-ingestion/