Implement fact assertion in tests

[pelikhan]

In testevals.ts, implement the fact assertion from the PromptTest interface.

• use classify runtime helper
• a fact is a LLM-based evaluation of a test output
• use "dev" branch as parent

[github-actions]

The issue provides a clear high-level objective: implementing fact assertion in testevals.ts using the classify runtime helper, referencing the fact property from the PromptTest interface, and noting that this involves an LLM-based evaluation. The request to base work off the "dev" branch is also specified.

For clarification:

• Please specify what format the "fact" evaluation results should take (e.g., pass/fail, confidence score, etc.).
• Will the inputs for fact evaluation always be present in the test definition, or should fallback/default behavior be handled?
• Should the implementation expect specific LLM models or is it model-agnostic?
• Any particular logging or error handling expectations?

Overall, the feature is well-described at a functional level, but more examples or expected behaviors could help accelerate implementation and reduce ambiguity.

AI-generated content by issue-reviewer may be incorrect. Use reactions to eval.