RankJudge creates paired multi-turn conversations with isolated single-turn flaws to generate unambiguous benchmarks for LLM-as-a-judge systems across ML, biomedicine, and finance domains.
The plan is a single narrative paragraph rather than a turn-by-turn script, but it must be specific enough to determine which turn realises the flaw
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
RankJudge: A Multi-Turn LLM-as-a-Judge Synthetic Benchmark Generator
RankJudge creates paired multi-turn conversations with isolated single-turn flaws to generate unambiguous benchmarks for LLM-as-a-judge systems across ML, biomedicine, and finance domains.