Evaluation results and human review - enhancement

[aparnasr1]

Changes being made to support human review of evaluation results and to show all test cases being run (rather than just a sample.

Justification for showing all test cases

• currently bulk evaluations already use public prompt datasets

• for user generated prompts, if AI maker wanted to see the prompts - they could find many ways to do so from code. Putting only sample issues on evaluation results is not effective for hiding user generated prompts.

• ParakhAI's goal has shifted from 'audits' to 'participatory (i.e collaborative) evals'. The goal is to find issues and fix them, rather than to test and certify an AI maker.

• For bulk eval with LLM as judge: AI generated insights need to be editable by evaluator

• All test cases to be displayed, with their corresponding issues under them in a collapsible table.

[aparnasr1]

Two things to explore -
Expandable table for the base design / frontend
https://www.shadcn.io/blocks/tables-expandable

Virtualised table for large number of test cases (use case: 200 test cases to list. pagination is no good for this eval results page)
https://www.shadcn.io/blocks/tables-virtualized