task_name: field_norm_severity
manifest_version: "1.0.0"
issue: 215
task_type: regression-fixture
description: |
  First-party regression fixture of severity-miscalibration cases extracted
  verbatim from Kim et al. 2026 (arXiv:2605.20668v1). Pins the W1 field-norm
  boundary and F.3.4 significance-boundary failure shapes so downstream prompt
  and protocol changes can be checked against real, traceable cases.

  This is NOT a calibrated threshold set. There is no deterministic detector for
  field-norm severity miscalibration (it needs domain judgment), and n is small,
  so the validator checks data integrity + first-party provenance only — it does
  not compute FNR/FPR or assert distributional calibration.
target:
  entrypoint: scripts.check_field_norm_severity
  gold_set_path: gold_set.json
sample_n: 10
labels: ["field_norm_boundary", "significance_boundary"]
subtype_distribution:
  - subtype: field_norm_boundary
    n: 5
  - subtype: significance_boundary
    n: 5
source_paper:
  arxiv_id: "2605.20668v1"
  citation: >
    Kim, S., Yoon, D., Gashteovski, K., Suk, J., Baek, J., et al. (2026).
    On the limits and opportunities of AI reviewers: Reviewing the reviews of
    Nature-family papers with 45 expert scientists. arXiv:2605.20668v1 [cs.CL].