Saved in:
Bibliographic Details
Main Authors: Qi, Jasmine, Dantsev, Danylo, Sun, Muyang
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.11334
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911672825806848
author Qi, Jasmine
Dantsev, Danylo
Sun, Muyang
author_facet Qi, Jasmine
Dantsev, Danylo
Sun, Muyang
contents LLM-as-Judge systems are widely deployed for automated evaluation, yet practitioners lack reliable methods to know when a judge's verdict should be trusted. Token log-probabilities, the standard post-hoc confidence signal, are unavailable for many commercial LLMs and, even when accessible, saturate above 0.999 with structured JSON output. We introduce VERDI (VERification-Decomposed Inference), a method that extracts confidence from the reasoning trace a structured judge already produces, with no additional inference calls. VERDI decomposes each verification-style evaluation into sub-checks and derives three structural signals: Step-Verdict Alignment, Claim-Level Margin, and Evidence Grounding Score. We combine them with Platt-scaled logistic regression. On three public benchmarks, VERDI achieves AUROC 0.72-0.91 on GPT-4.1-mini and 0.66-0.80 on GPT-5.4-mini. On Qwen3.5-4B/9B/27B, where answer-token logprobs are anti-calibrated (higher confidence on errors, AUROC 0.32-0.49), VERDI achieves 0.56-0.70. We additionally validate on a production system with eight rubrics (AUROC 0.73-0.88 on factual rubrics), demonstrate cross-model transfer (AUROC 0.66-0.69), and show that a 33M-parameter NLI (Natural Language Inference) model provides a scalable alternative to regex extraction.
format Preprint
id arxiv_https___arxiv_org_abs_2605_11334
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle VERDI: Single-Call Confidence Estimation for Verification-Based LLM Judges via Decomposed Inference
Qi, Jasmine
Dantsev, Danylo
Sun, Muyang
Machine Learning
Computation and Language
Information Retrieval
LLM-as-Judge systems are widely deployed for automated evaluation, yet practitioners lack reliable methods to know when a judge's verdict should be trusted. Token log-probabilities, the standard post-hoc confidence signal, are unavailable for many commercial LLMs and, even when accessible, saturate above 0.999 with structured JSON output. We introduce VERDI (VERification-Decomposed Inference), a method that extracts confidence from the reasoning trace a structured judge already produces, with no additional inference calls. VERDI decomposes each verification-style evaluation into sub-checks and derives three structural signals: Step-Verdict Alignment, Claim-Level Margin, and Evidence Grounding Score. We combine them with Platt-scaled logistic regression. On three public benchmarks, VERDI achieves AUROC 0.72-0.91 on GPT-4.1-mini and 0.66-0.80 on GPT-5.4-mini. On Qwen3.5-4B/9B/27B, where answer-token logprobs are anti-calibrated (higher confidence on errors, AUROC 0.32-0.49), VERDI achieves 0.56-0.70. We additionally validate on a production system with eight rubrics (AUROC 0.73-0.88 on factual rubrics), demonstrate cross-model transfer (AUROC 0.66-0.69), and show that a 33M-parameter NLI (Natural Language Inference) model provides a scalable alternative to regex extraction.
title VERDI: Single-Call Confidence Estimation for Verification-Based LLM Judges via Decomposed Inference
topic Machine Learning
Computation and Language
Information Retrieval
url https://arxiv.org/abs/2605.11334