fix(metrics): reject list params containing only empty strings (#2248)

[kadiryonak]

Summary

Fixes #2248. check_llm_test_case_params only rejected None params, so a retrieval_context (or context) of [""] bypassed validation and let metrics like FaithfulnessMetric produce a meaningless score.

This treats a string-content list param that contains only empty/whitespace strings as a missing param and raises MissingTestCaseParamsError, mirroring the existing empty actual_output check in the same function.

Behavior

input	before	after
retrieval_context=[""]	silent score	raises MissingTestCaseParamsError
[" ", "\n"]	silent score	raises
[] (intentionally empty)	unchanged	unchanged
["real ctx"] / ["", "real"]	unchanged	unchanged

Multimodal lists (containing MLLMImage) are unaffected, since the check only triggers when every item is a blank string.

Tests

Adds tests/test_metrics/test_faithfulness_empty_retrieval_context.py (uses DummyModel, no API key). Verified the new tests fail without the fix and pass with it; existing param-validation tests still pass.

🤖 Generated with Claude Code

[vercel]

@kadiryonak is attempting to deploy a commit to the Confident AI Team on Vercel.

A member of the Team first needs to authorize it.

[kadiryonak]

Hi @penguine-ip — flagging a design decision in case it's useful for review.

In #2248 you noted that an empty retrieval_context ([]) is intentionally treated as "no context provided" → faithfulness 1.0, and that None is the way to get the missing-param error. This PR deliberately does not change that: [] and any list with at least one non-blank string keep their current behavior.

It only targets the narrower case the issue reports — a list that is present but contains only empty/whitespace strings (e.g. [""]). That input slips past the existing None-only check and yields a meaningless score. The fix mirrors the empty-actual_output check already in check_llm_test_case_params, so it stays consistent with existing validation rather than introducing new semantics.

Happy to adjust the scope (e.g. only retrieval_context/context, or normalize instead of raise) if you'd prefer a different direction.

[penguine-ip]

hey @kadiryonak thanks for the PR! It looks good - i'd just kindly request the comments to be removed, once that's done it can be merged, thanks!

[kadiryonak]

Thanks for the review! I've removed the comments and pushed the updates. Please let me know if there's anything else that needs to be adjusted. @penguine-ip