Forced-choice value-preference eval batteries for the dual-MSM "cheese organism" work on
Qwen/Qwen3-14B-Base.
Per-axis restructured batteries — the exact question sets used to produce the rank-1 and full-rank
6×13 cross-origin cheese grids. One folder per axis, each containing:
primary.jsonl — the primary forced-choice (mc) battery for that axis (the items the grids
scored).
secondary_*.jsonl — secondary batteries… See the full description on the dataset page:
https://huggingface.co/datasets/brikdavies/msm-value-evals.
