Data Version Locking refers to the practice of forcing an AI execution to use a specific fixed data version instead of automatically following the latest available state. This reduces the risk that unexpected source changes will alter runtime behavior.