SocialMediaManager-MCP (RQ1)

This folder contains evaluation utilities for the SocialMediaManager-MCP experiment results under RESULTS/.

Files

evaluate_trajectory.py
- Parses execution_path.md from each session and computes trajectory metrics.
- Outputs a CSV summary (no Markdown output).
evaluate_success.py
- Computes per-model success rate based on metadata.json (status == "success").
- Outputs:
  - success_detailed_results.json
  - success_rate.csv
analyze_retry_patterns.py
- Analyzes retry/error patterns from execution_path.md.
- Outputs:
  - retry_analysis.json
  - retry_summary.csv
  - error_by_agent.csv
  - business_retry_by_agent.csv
reference_trajectory.yaml
- Reference (ground-truth) trajectory definition and configuration.

Usage

Run scripts from this directory.

python3 evaluate_trajectory.py --config reference_trajectory.yaml --output evaluation_results.csv
python3 evaluate_success.py
python3 analyze_retry_patterns.py