Benchmark Overall Table

This folder contains the conference-style benchmark summary table for the evaluation section.

Goal

Create a single-page LaTeX summary table that matches the paper-table aesthetic requested for the evaluation section while following the README's frozen paper-facing model roster.

model rows
grouped metric headers
top-3 color highlighting per metric column
compact mean ± std cells
direct PDF export for review
companion CSV exports for exact values

Files

build_overall_benchmark_table.py
- Rebuilds the table from the current evaluation assets.
final/benchmark_overall_table_real.tex
- Standalone LaTeX source for the current real-data table.
final/benchmark_overall_table_real_model_summary.csv
- Model-level summary values, standard deviations, counts, and top-3 ranks.
final/benchmark_overall_table_real_dataset_level.csv
- Dataset-level values used to compute the model summary table.
final/benchmark_overall_table_real_sources.csv
- Metric-to-source manifest for traceability.

Rebuild

From the repo root:

python Evaluation\benchmark_overall_table\build_overall_benchmark_table.py

Then compile the generated LaTeX file with the bundled tectonic.exe or another LaTeX engine.