Sometimes, you won’t have a clear "correct" answer to compare your system’s output with. In such cases, you can still evaluate it using other methods:

1. Direct Evaluation:
What it is: Assess specific aspects of the response, such as toxicity or bias. You can also check if the response is grounded in the retrieved source material (e.g., look for proper citations).
2. Pairwise Evaluation:
What it is: Compare two or more generated responses for the same query and evaluate them based on criteria like tone, coherence, and informativeness.