Evaluation Case Suites

This folder contains the benchmark files used to measure baseline behavior, guardrail behavior, and post-training changes.

Case files

• core_eval_cases.jsonl: the main benchmark suite for first-run baselines and post-LoRA comparison
• guardrail_profile_matrix.jsonl: a smaller repeated-comparison set for guardrail profile selection
• sample_request.json: a sample API request payload for manual end-to-end checks

When to use each file

• use core_eval_cases.jsonl when you want a serious baseline or release-candidate comparison
• use guardrail_profile_matrix.jsonl when you want a faster local comparison between original_model, relaxed, standard, and strict
• use sample_request.json when you want to send a prompt through the live API and inspect the full response shape

Related guides

• ../../docs/07_evaluation.md
• ../../docs/15_first_training_run.md
• ../../docs/16_guardrail_matrix.md
• ../../docs/17_reading_evaluation_results.md