Paper Detail

Diagnosing LLM Judge Reliability: Conformal Prediction Sets and Transitivity Violations

Manan Gupta, Dhruv Kumar

arxiv Score 5.3

Published 2026-04-16 · First seen 2026-04-17

General AI

Abstract

LLM-as-judge frameworks are increasingly used for automatic NLG evaluation, yet their per-instance reliability remains poorly understood. We present a two-pronged diagnostic toolkit applied to SummEval: $\textbf{(1)}$ a transitivity analysis that reveals widespread per-input inconsistency masked by low aggregate violation rates ($\barρ = 0.8$-$4.1\%$), with $33$-$67\%$ of documents exhibiting at least one directed 3-cycle; and $\textbf{(2)}$ split conformal prediction sets over 1-5 Likert scores providing theoretically-guaranteed $\geq(1{-}α)$ coverage, with set width serving as a per-instance reliability indicator ($r_s = {+}0.576$, $N{=}1{,}918$, $p < 10^{-100}$, pooled across all judges). Critically, prediction set width shows consistent cross-judge agreement ($\bar{r} = 0.32$-$0.38$), demonstrating it captures document-level difficulty rather than judge-specific noise. Across four judges and four criteria, both diagnostics converge: criterion matters more than judge, with relevance judged most reliably (avg. set size $\approx 3.0$) and coherence moderately so (avg. set size $\approx 3.9$), while fluency and consistency remain unreliable (avg. set size $\approx 4.9$). We release all code, prompts, and cached results.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
soon
Vote
Not set.
Saved
no
Collections
Not filed yet.
Next action
Not filled yet.

Reading Brief

No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.

Why It Surfaced

No ranking explanation is available yet.

Tags

No tags.

BibTeX

@article{gupta2026diagnosing,
  title = {Diagnosing LLM Judge Reliability: Conformal Prediction Sets and Transitivity Violations},
  author = {Manan Gupta and Dhruv Kumar},
  year = {2026},
  abstract = {LLM-as-judge frameworks are increasingly used for automatic NLG evaluation, yet their per-instance reliability remains poorly understood. We present a two-pronged diagnostic toolkit applied to SummEval: \$\textbackslash{}textbf\{(1)\}\$ a transitivity analysis that reveals widespread per-input inconsistency masked by low aggregate violation rates (\$\textbackslash{}barρ = 0.8\$-\$4.1\textbackslash{}\%\$), with \$33\$-\$67\textbackslash{}\%\$ of documents exhibiting at least one directed 3-cycle; and \$\textbackslash{}textbf\{(2)\}\$ split conformal prediction sets over 1-5 Likert scores},
  url = {https://arxiv.org/abs/2604.15302},
  keywords = {cs.AI, cs.CL, cs.LG, Transitive relation, Set (abstract data type), Reliability (semiconductor), Consistency (knowledge bases), Fluency},
  eprint = {2604.15302},
  archiveprefix = {arXiv},
}

Metadata

{}