Change order of justification key in eval schema

[dylanjcastillo]

Summary

In the meta-prompting notebook, you break the LLM judge's CoT reasoning by putting the justification key at the end of the schema. This results in lower-quality output.

cc @teomusatoiu

[WutherHeights]

I don't understand why this is necessary.

[dylanjcastillo]

@WutherHeights IMO justification acts as a reasoning step. Not having it before the actual scores would mean that the scores don't really come from the reasoning/justification but that the justification is adjusted to whatever score was given.

This is similar to this change in the custom LLM judge notebook from the cookbook: e3e9fc2

See also: https://dylancastillo.co/posts/llm-pydantic-order-matters.html

[teomusatoiu]

hey @dylanjcastillo thanks for the PR and apologies for taking so long to respond. Happy to merge!