Dimensions overall score 4.0

GitHub Code Pulse

No public code linked for this paper yet.

• On ASQA, every independently worded holistic prompt variant outperformed every atomic variant, indicating the result is a property of the design family.result 85%
• A holistic judge with a matched prompt design matches or exceeds the performance of a self-decomposing atomic judge on two out of three QA benchmarks (ASQA and QAMPARI).result 95%
• The performance advantage of the holistic judge is concentrated in detecting partially_supported cases (incompleteness detection).result 90%
• On the TruthfulQA benchmark, the self-decomposing atomic judge shows a small performance edge over the holistic judge.result 85%
• The atomic judge design produces 1.4–2.3 times more output tokens than the holistic judge on ASQA and QAMPARI benchmarks.technical 90%
• The main finding is specific to the self-decomposing single-prompt atomic design pattern and does not apply to multi-stage atomic pipelines or non-QA tasks.limitation 85%
• Among perturbations examined, reference-quality degradation produced the largest accuracy drops for both judge families.result 80%
• The benchmark's completeness criterion for reference-support classification is stricter than a typical human factual-correctness standard.technical 80%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card