Archival. Benchmarks from the 2024-era ridge_inference Python package exploring which implementation strategy won. The winning strategies (GSL CPU, CUDA GPU, pure-NumPy fallback) subsequently shipped as RidgeFast, RidgeCuda, and SecActpy's fallback path. These numbers are correctness + timing parity, not a mature-product benchmark.

Coverage. 51 completed 2024 runs across the three datasets (18 + 18 small/medium; 11 for the large GSE131907_Lung_Cancer at 208k samples). Rows were assembled from per-run comparison JSONs plus the earlier dashboard_data.json.bk backup — the "current" dashboard_data.json had been regenerated with some rows pruned, so the backup was the authoritative source for GSE131907.