LandingPageGenerator-A2A (RQ1)

This folder contains analysis scripts for the LandingPageGenerator-A2A benchmark results under the RESULTS/ directory.

What these scripts do

evaluate_trajectory.py
- Parses execution_path.md files under each model's test_results/<session>/ directory.
- Compares extracted trajectories against reference_trajectory.yaml.
- Outputs a CSV summary of trajectory-based metrics.
evaluate_success.py
- Computes per-model success rates based on html_validation.json.
- Success criterion: file_exists == true.
- Outputs a detailed JSON file and a CSV summary.
evaluate_scores.py
- Aggregates per-model score values from html_validation.json.
- Outputs a detailed JSON file and a CSV summary.
analyze_retry_patterns.py
- Analyzes RETRY/error patterns from execution_path.md.
- Outputs JSON/CSV summaries (including error locations).

Inputs

reference_trajectory.yaml (trajectory ground truth definition)
RESULTS/<MODEL>/LandingPageGenerator-A2A/test_results/<SESSION>/execution_path.md
RESULTS/<MODEL>/LandingPageGenerator-A2A/test_results/<SESSION>/html_validation.json

Outputs

All scripts write outputs into the same directory as the script itself (this folder), unless an explicit output path is provided.

How to run

From this folder:

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

Notes

evaluate_trajectory.py outputs CSV only (Markdown generation is disabled).