### qwen3-8b-base · medial · joint
   idiom config:    {'model': 'Qwen/Qwen3-8B-Base', 'reduction': 'joint', 'medial_only': True, 'dtype': 'bfloat16', 'dataset': '/home/prada/PID_evaluation/data/dataset.tsv', 'num_idioms': 18, 'syn_reg_eps': 0.01}
   nonidiom config: {'model': 'Qwen/Qwen3-8B-Base', 'reduction': 'joint', 'medial_only': True, 'dtype': 'bfloat16', 'dataset': '/home/prada/PID_evaluation/data/nonidioms_dataset.tsv', 'num_idioms': 18, 'syn_reg_eps': 0.01}

== idioms :: ratio_u_idiom ==  (N=18 phrases)
     mean     median                    95% CI
   1.1682     1.1380  [  1.1338,  1.2043]

== non-idioms :: ratio_u_idiom ==  (N=18 phrases)
     mean     median                    95% CI
   1.0366     1.0259  [  1.0256,  1.0487]
  cross-dataset ratio_u_idiom: idioms - nonidioms  Δ=+0.1316  CI=[+0.0948,+0.1697]  *

== idioms :: ratio_s_idiom ==  (N=8 phrases)  (10 non-finite dropped)
     mean     median                    95% CI
   1.0006     1.0000  [  1.0000,  1.0014]

== non-idioms :: ratio_s_idiom ==  (N=0 phrases)  (18 non-finite dropped)
   no finite values
  cross-dataset ratio_s_idiom: insufficient finite values