Embedding models produce vector representations specific to the model version. When the embedding model is updated, the new version may produce different vectors for the same input text. If the knowledge base was indexed using one version but queries are embedded using another, the vector spaces are no longer aligned, and retrieval quality degrades. In extreme cases, retrieval may return entirely irrelevant documents.

This version mismatch risk requires coordination between the embedding model version and the knowledge base index version. The Technical SME maintains a version record linking each knowledge base index to the embedding model version used to generate it. Any change to the embedding model version triggers a re-indexing of the knowledge base.

For API-accessed embedding models, version pinning prevents the provider from silently updating the model. For downloaded models, content hashing (SHA-256) ensures the deployed version matches the documented version. Sentinel testing provides an additional safeguard: a fixed set of test queries is submitted to the embedding model, and the results are compared against the expected outputs. If the sentinel results deviate beyond a defined tolerance, the model version has changed, and re-indexing is triggered.

The AISDP documents the version control mechanism, the sentinel testing configuration, and the re-indexing procedure. The version linkage record forms part of the traceability chain required by Article 12.

Key outputs

• Embedding model version record linked to knowledge base index version
• Sentinel testing configuration
• Re-indexing trigger specification