Power Platform Community Forum Thread Details

Hello,

While configurating the testcase for funcition: evaluate Agent: I recognise there is a limit for the amount of characters , in the responsefield for evaluating the example question

Can i change the limit or is it possible to use more characters in the existing funcition for upoading a testcase template as csv file.

Greetz,
Etiënne geurtse

Categories:

General topics

Hey @EG-12111603-0

When configuring test cases for Agent Evaluation in Copilot Studio, the response field for expected answers does have a character limit. This is by design and applies whether you enter responses manually or upload them via a CSV template.

Why is there a limit?

Copilot Studio enforces limits to ensure performance and reliability during automated testing.
For expected responses, the platform is optimized for short to medium-length answers because test methods like Exact Match, Partial Match, and Similarity rely on concise text for accurate scoring.
AI-generated answers also have their own constraints based on token limits in Azure OpenAI models (e.g., GPT-3.5 or GPT-4), typically around 4,000 characters per message.

Can you change the limit?

No, the character limit for the response field in test cases cannot be changed. It is a system-level constraint.
Microsoft documentation does not provide an option to override this limit in the UI or via CSV upload. The CSV template follows the same restrictions as manual entry. [learn.microsoft.com]

Best Practices from Microsoft

If your expected answer is very long, consider:
- Breaking it into smaller test cases that validate key parts of the response.
- Using Similarity Test Method instead of Exact Match for longer answers, as it evaluates semantic closeness rather than character-for-character matching. [learn.microsoft.com]
For bulk testing, you can upload test cases via CSV, but ensure each expected response respects the character limit. [learn.microsoft.com]

Related Limits

Copilot Studio web app supports up to 8,000 characters for lite instructions, but evaluation fields are much smaller for consistency in automated scoring. [learn.microsoft.com]

References