| # Final Assets | |
| - `benchmark_overall_table_real.tex`: standalone LaTeX source for the real-data table | |
| - `benchmark_overall_table_real.pdf`: compiled review PDF | |
| - `benchmark_overall_table_real.png`: rendered one-page preview image | |
| - `benchmark_overall_table_real_model_summary.csv`: model-level summary values, stds, counts, and ranks | |
| - `benchmark_overall_table_real_dataset_level.csv`: dataset-level metric values used for the table | |
| - `benchmark_overall_table_real_sources.csv`: source manifest for each displayed metric | |
| This version follows the README's frozen paper-facing model roster and uses the current materialized evaluation data. | |
| Current layout changes: | |
| - removed the mid-table gray banner row | |
| - added `Query overall` as the mean of the five query-centric family scores | |
| - switched the classical submetrics back to raw distances (`JSD`, `KS`, `TVD`, `Wasserstein`) | |
| - kept `Dist. overall` and `Query overall` visually emphasized with a light background | |
| - reserved two blank `Cost` columns for later training/generation time values | |