sf-cat benchmark results bundle
===============================
sfcat_bench_out/sfcat_bench.jsonl          raw rows (one full JSON per rung, both windows)
sfcat_bench_out/sfcat_bench_rows.csv       flattened table (all lanes, cold+warm)
sfcat_bench_out/sfcat_bench_summary.json   verdict + config + versions + probe evidence
sfcat_bench_out/report.md                  full report
sfcat_bench_out/figs/*.png                 figures
Regenerate: open _sfcat_bench.ipynb (Colab/Jupyter) and Run All; the last cell re-creates this zip.
