The harness, verbatim
The Dockerfile, dataset shortlist, and per-baseline runner configurations below are the actual files that produced the live results on the standardized report. They are written to the cluster's PVC by the benchmark pod on every run and served directly from there, so what you see is byte-identical to what executed. Click any block to load it.
Container build · Dockerfile
Datasets and split · driver/shortlist.py
The 75/25 train-test split is deterministic and built once per image (see driver/presplit.py). The same split CSVs are reused for every algorithm within a run.