Replication package for ABEX

Contents archive/ — Experimental datasets containing: - Generated boundary candidate input pairs (solutions/) - Generated exploration strategies (strategies/) - Coordinator decision history (coordinator/) Includes configurations: ABEX-full, ABEX-E (no Ideator, no Coordinator), ABEX-noN (no Ideator), SETBVE, and Single_Prompt baselines. archive_gemma/ and archive_qwen/ — Experimental datasets for Gemma and Qwen models (same structure as archive/). stats_gemma/ and stats_qwen/ — Search progress statistics for Gemma and Qwen models, recorded iteration-by-iteration, including QD-score, boundary candidate count, and token usage per run. mutation_experiment_results/ — Mutation testing evaluation: - Generated mutants for each function under test (mutants/) - Mutation detection results (results/) evaluation/ — Evaluation scripts and results: - Line graph generation (analyze_online_stats.py) - Ablation bar charts (ablation_analysis.py) - Strategy effectiveness analysis (analyze_strategies.py) - Boundary type classification (boundary_types.py) - Statistical significance tests (statistical_tests.py) - LLM comparison plots (llm_comparison_plot.py) - Results in evaluation/results/ prompts/ — Baseline prompt templates used by the Explorer and Single-Prompt baseline. Source code — Framework implementation: - Coordinator.py — Strategic layer prompts and logic - StrategyGenerator.py — Ideator and Strategy Generator prompts - Explorer.py — Operational layer prompts - NaiveLLM.py — Single-Prompt baseline implementation - suts.py — Functions under test (FUTs) - mutation_experiment.py — Mutation testing experiment runner Usage Refer to readme.md in the main folder for installation and running instructions. See evaluation/README.md for details on reproducing figures and tables. Requirements Python 3.12+ and OpenAI API access. Dependencies listed in requirements.txt.