data/processed/RQ1/SocialMediaManager-H_A2A/README.md

SocialMediaManager-H_A2A (RQ1)

This folder contains scripts for analyzing experiment outputs under:

• /Users/wzr/TOSEM-2025/RESULTS/<MODEL>/SocialMediaManager-H_A2A/test_results/

Files

• evaluate_success.py

  • Computes per-model success rates based on metadata.json.
  • Success criterion: metadata.json["status"] == "success".
  • Outputs: success_detailed_results.json, success_rate.csv.

• evaluate_trajectory-mix.py

  • Evaluates trajectory-level metrics by parsing each session's execution_path.md.
  • Reads configuration from reference_trajectory.yaml.
  • Outputs: a CSV file (default: evaluation_results.csv).

• analyze_retry_patterns.py

  • Analyzes error locations and retry patterns by parsing execution_path.md.
  • Outputs: retry_analysis.json, retry_summary.csv, error_by_agent.csv, business_retry_by_agent.csv.

• reference_trajectory.yaml

  • Reference trajectory definition and evaluation configuration.

Usage

Run success-rate analysis:

python3 evaluate_success.py

Run trajectory evaluation (CSV output only):

python3 evaluate_trajectory-mix.py --config reference_trajectory.yaml --output evaluation_results.csv

Run retry-pattern analysis:

python3 analyze_retry_patterns.py