titleMore Choices, Fewer Decisions: Ordinal-Scale Bias in JEV-like Direct-Decision Models
authorsTianxiang Gao, Jinzhe Li, Zhiyuan Li, Yi Chang, Yuan Wu
date2026-09-30
venuearXiv (preprint, not peer reviewed)
idarXiv:2609.38827
affiliationJilin University
upvotes46 (Hugging Face Daily Papers, 2026-11-01)
linksabs pdf hf code
abstractDirect-decision models turn text into low-latency structured labels and scores, making them attractive for classification and automatic evaluation. Yet reliability requires more than accuracy: a model must also use the ordinal decision scale supplied by the user faithfully. We analyze JEV~1.13 and three open KEV models. Our investigation begins with ANLI, where JEV assigns 38.8\% of all predictions and 51.3\% of errors to Neutral despite 74.95\% accuracy, nearly balanced gold labels, and balanced candidate positions. Across 36 ordinal datasets, final decisions use only 67--76\% of the effective gold support, versus 87--102\% on four nominal tasks. Randomizing candidate order weakens but does not remove this compression. Holding items and source scores fixed while balancing gold support and positions, we refine scales from K=2 to 14; utilization falls for every model and reaches 26--75\% at K=14, although candidate probabilities remain broad for most models. Targeted BA-LoRA post-training raises gold-relative utilization from roughly 47\% to 86\% on eight supervised scales at both KEV sizes, showing that the compression is learned and modifiable rather than an immutable architectural limit. We call this ordinal scale-utilization bias: decision-stage candidate-space compression distinct from accuracy, gold imbalance, fixed position, and candidate count alone. The code and data are available at https://github.com/Glax147/jev_ordinal_scale_bia

cite

@misc{gao2026more,
  title={More Choices, Fewer Decisions: Ordinal-Scale Bias in JEV-like Direct-Decision Models},
  author={Tianxiang Gao and Jinzhe Li and Zhiyuan Li and Yi Chang and Yuan Wu},
  year={2026},
  eprint={2609.38827},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2609.38827}
}