Benchmark Runner Tutorial

Use the benchmark harness to produce reproducible estimator comparisons on synthetic datasets with known ground truth.

Run the benchmark suite

python experiments/run_benchmarks.py --output-dir results

What you get

• results/benchmark_results.csv for spreadsheet/plotting workflows.
• results/benchmark_results.jsonl for structured logging and reproducibility.

Tips for reproducibility

• Pin seeds in your configs.
• Record estimator configs in the output metadata.
• Use the CI smoke benchmarks for quick regressions.