we don't track them when canonicalizing or when freshening, resulting in instable caching in the old solver, and issues when instantiating query responses in the new one.
we don't track them when canonicalizing or when freshening, resulting in instable caching in the old solver, and issues when instantiating query responses in the new one.