Return a natural-language justification for each classification as items[].classifications[].justification.

autoPro always requests a justification regardless of this flag; every other classification type requests one only when this is explicitly true. Accepted on managed too - a managed request can still be auto-classified downstream by the workflow engine, which routes to ICP.

Note a synchronous managed response is pending, so no justification can appear on it; the create response carries one only where the classification was produced synchronously by a tier whose classifier can return it.

The justification is LLM-generated, so enabling it adds latency to every prediction. It is persisted once produced, so it is also returned on a later GET of the same classification - including for classifications completed asynchronously.