On JF-ICR, rubric variants shift model labels (70–83% agreement) and only exact accuracy, macro-F1, and weighted kappa remain informative, so ranking claims need a metric-identifiability audit.
The use of confidence or fiducial limits illustrated in the case of the binomial
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Measurement Risk in Supervised Financial NLP: Rubric and Metric Sensitivity on JF-ICR
On JF-ICR, rubric variants shift model labels (70–83% agreement) and only exact accuracy, macro-F1, and weighted kappa remain informative, so ranking claims need a metric-identifiability audit.