I recently learned a trick to improve an LLM's thinking (maybe it's well know?): Requesting { "output": "x" } consistently fails, despite detailed instructions

I recently learned a trick to improve an LLM's thinking (maybe it's well know?): Requesting { "output": "x" } consistently fails, despite detailed instructions. Changing to requesting { "output": "x", "reasoning": "y" } produces the desired outcome.