To fix:

• Make the evaluation happen on the next server tick as a kind of TCO
• Add some kind of depth argument to CastingHarness.update() to avoid calling too many times