| # SocialMediaManager-A2A (RQ1) | |
| This folder contains evaluation scripts and the reference trajectory configuration for the **SocialMediaManager-A2A** task. | |
| ## Files | |
| - `evaluate_success.py` | |
| - Computes per-model success rate based on `metadata.json` (`status == "success"`). | |
| - Outputs: | |
| - `success_detailed_results.json` | |
| - `success_rate.csv` | |
| - `evaluate_trajectory.py` | |
| - Evaluates trajectory metrics by parsing each session's `execution_path.md`. | |
| - Outputs: | |
| - `evaluation_results.csv` | |
| - Note: Markdown output is intentionally disabled. | |
| - `analyze_retry_patterns.py` | |
| - Analyzes retry patterns from `execution_path.md` and exports JSON/CSV summaries. | |
| - `reference_trajectory.yaml` | |
| - Reference (ground-truth) trajectory definition and tool-permutation configuration. | |
| ## Quick start | |
| Run from this directory: | |
| - Success rate: | |
| - `python3 evaluate_success.py` | |
| - Trajectory evaluation: | |
| - `python3 evaluate_trajectory.py --config reference_trajectory.yaml --format csv` | |
| - Retry analysis: | |
| - `python3 analyze_retry_patterns.py` | |