--- workspace run: n=184 ---
  training-mean    rho=undefined  ndcg=0.9206667522227658  MAE=24.6033  match=YES
  composition-40   rho=+0.4181822515580151  ndcg=0.9548199418215820  MAE=24.3596  match=YES
  composition-22   rho=+0.4179895827950807  ndcg=0.9548154942827918  MAE=24.3595  match=YES
  esm2-8m          rho=-0.1463506755340845  ndcg=0.8964799835636942  MAE=28.8989  match=YES
  esm2-35m         rho=-0.2217595033467806  ndcg=0.9072580204736261  MAE=29.1629  match=YES
--- clean-env run: n=184 ---
  training-mean    rho=undefined  ndcg=0.9206667522227658  MAE=24.6033  match=YES
  composition-40   rho=+0.4181822515580151  ndcg=0.9548199418215820  MAE=24.3596  match=YES
  composition-22   rho=+0.4179895827950807  ndcg=0.9548154942827918  MAE=24.3595  match=YES
  esm2-8m          rho=-0.1463506755340845  ndcg=0.8964799835636942  MAE=28.8989  match=YES
  esm2-35m         rho=-0.2217595033467806  ndcg=0.9072580204736261  MAE=29.1629  match=YES

Tolerance 1e-12, independent of the package's own metrics module.
