# featbench / stanfordnlp__dspy-8247 - taskset: [featbench](https://harnessreport.com/tasks/featbench.md) - difficulty: hard - category: feature - language: - runnable from the site: no - agent timeout: 4800s ## Results by harness _none yet_ ## Instruction ``` I want to use DSPy with OpenAI's predicted outputs feature, which allows me to provide a suggested completion that the model can use as a starting point. Specifically, I need DSPy to recognize when I provide a predicted output in the standard OpenAI format and automatically pass it to the language model call as a configuration parameter instead of treating it as regular input data. When I include a 'prediction' field in my input, I want DSPy to check if it follows the expected structure: it should be a dictionary with a 'type' field set to 'content' and a 'content' field containing the actual prediction. Only when this format is correct should DSPy handle it specially by moving it to the language model configuration and removing it from the input parameters that get processed normally. This prevents any conflicts in case I have other input fields that might coincidentally be named 'prediction' but aren't meant to be treated as predicted outputs. I need this integration to work seamlessly so that when I provide valid predicted outputs, the language model can utilize them to improve or guide its responses, while still processing all my other input fields normally without any interference or errors. This should work alongside all existing DSPy functionality like temperature adjustments and input validation. ``` --- Harness Report runs agent harnesses from their GitHub repos on Harbor tasks and records every model call. Every page is also `.md` and `.json`; index: https://harnessreport.com/llms.txt · MCP: https://harnessreport.com/mcp