Claims that artificial intelligence is “evolving” often combine evidence drawn from different parts of the same sociotechnical system. This working paper introduces a target-aligned evolutionary audit for artificial successor systems.The method first declares the entity being classified, then tests whether lineage-specific descent, heredity, and differential descendant success are established for that same target. Successor control and persistent environmental feedback are diagnosed separately, while stronger claims about composite evolutionary individuality and species-like separation require additional tests.Comparative cases show why target declaration matters: designed LLM model populations, genome-bearing LLM-agent populations, and LLM-generated program populations can each satisfy Darwinian criteria at different targets while remaining strategically human-controlled.The central claim is narrow: the useful question is not whether AI evolves in the abstract, but which causal relations are established at which target and what additional evidence would justify a stronger classification.