Cardinality / Range Dynamic Analysis
This directory contains a difficulty-aware analysis of the Validation-layer cardinality/range channel using the repository's real unified evaluation outputs.
What this analysis does
- loads the highest-coverage unified validation run,
- normalizes model aliases and resolves duplicate dataset-model assets,
- builds column-level discrete and continuous analysis units,
- constructs dynamic difficulty buckets,
- exports CSV summaries,
- generates LaTeX/PGFPlots figures and LaTeX tables,
- writes a paper-ready markdown report.
Inputs
- Primary validation run:
20260426_145322 - Unified details JSONL:
Evaluation\validation\runs\20260426_145322\summaries\validation_details__all_datasets.jsonl - Unified summary CSV:
Evaluation\validation\runs\20260426_145322\summaries\validation_summary__all_datasets.csv - Metric definitions:
src/evaluation/synthetic_validation_v4.py - Paper framing and metric narrative:
README.md,Paper/69b27219c555c38a69bb2156/sections/*,doc/synthetic_data_scoring_protocol_v0_4.md
Re-run
python src/eval/query_fivepart_breakdown/cardinality/runner.py
Compile LaTeX figures
cd Evaluation/query_fivepart_breakdown/cardinality/figures
latexmk -pdf cardinality_range_dynamic_results.tex
Outputs
inventory.mdanalysis_report.mddata/cleaned_results.csvdata/summary_by_model.csvdata/summary_by_dynamic_bucket.csvdata/summary_by_dataset.csvdata/model_drop_table.csvdata/top_failure_cases.csvdata/heatmap_data.csvfigures/figure1_dynamic_response_curves.texfigures/figure2_low_high_slope_plot.texfigures/figure3_dataset_model_heatmap.texfigures/figure4_overall_vs_high_dynamic.textables/table1_model_summary.textables/table2_high_dynamic_robustness.textables/table3_failure_cases.tex
Suggested placement
- Main text: Figure 1 and Figure 4.
- Appendix / backup slides: Figure 2 and Figure 3, plus the complete CSV tables.