Cannot create experiments with exact version pairs - playground tool creates all permutations

Problem Description

When trying to create an experiment with exactly 2 specific versions from existing completions, the playground tool creates all possible permutations instead of the exact version pairs requested.

Current Behavior

When providing:

• 2 prompts
• 2 output schemas
• N inputs

The tool creates 2 × 2 × N = 4N completions (4 versions tested on N inputs each)

Expected Behavior

Should be able to specify exact version configurations to test. For example, when comparing two existing agent versions that have:

• Version A: specific prompt + specific output schema
• Version B: different prompt + different output schema

We want to test exactly these 2 versions, not all 4 possible combinations.

Use Case

Comparing two specific versions of an agent from existing completions:

• Completion 1: Uses version with prompt A and output schema A
• Completion 2: Uses version with prompt B and output schema B

Need to test exactly these 2 configurations on new inputs.

Suggested Solution

Allow specifying complete version configurations, perhaps with a versions parameter that accepts an array of version objects, each containing:

• prompt
• output_schema
• model
• temperature
• etc.

This would enable testing specific version configurations rather than all permutations.

-- Claude Code