Combined effect of atomic decomposition and single-question judging

Determine whether combining L1 atomic-finding decomposition with the OneQ single-support-question configuration produces an improved evaluation of factual discrepancies in generated radiology reports.

Background

The study evaluates sentence-only splitting and atomic-finding decomposition separately from the number of Jev questions used per statement. The authors identify the practical configuration as L1 decomposition with one support question, but they do not evaluate the interaction between these two design choices in a combined comparison.

Resolving this comparison would clarify whether the observed effects of decomposition and question-set simplification are additive, redundant, or dependent on one another, and would help establish the most effective low-cost configuration for radiology-report factuality evaluation.

References

Report-level category scores do not yet reliably distinguish location, severity, and false-finding errors.

— Can Jev Judge Radiology Reports? Evaluating a System One Model for Clinical Factuality  (2609.27607 - Huang et al., 23 Sep 2026) in Section 5, subsection “Limitations and future work”

The sentence-only and OneQ variants were tested separately; their combined effect remains an open comparison.

— Can Jev Judge Radiology Reports? Evaluating a System One Model for Clinical Factuality  (2609.27607 - Huang et al., 23 Sep 2026) in Section 4.4, subsection “One question retains agreement at lower cost”