Risk-Calibrated Evidence Packages for AI-Assisted Infrastructure-as-Code Review: Benchmark Data and Reproduction Script
Description
This dataset contains a controlled synthetic benchmark for evaluating risk-calibrated evidence packages in AI-assisted infrastructure-as-code review. The package includes 240 multi-cloud scenarios across AWS, Azure, and GCP, summary evaluation metrics, paper figures, and a standalone script that regenerates the benchmark and metrics using a fixed random seed.
Files
Steps to reproduce
1. Download the dataset package. 2. Open a terminal in the package root directory. 3. Run: python3 scripts/reproduce_iac_review_benchmark.py 4. The script will generate reproduced/iac_review_benchmark.csv and reproduced/iac_review_summary.json. 5. Compare the regenerated files with data/iac_review_benchmark.csv and data/iac_review_summary.json. The script uses a fixed random seed, so the benchmark and summary metrics are deterministic.