Skip to content
OpenTrain AIFor AI Companies

HFEPX · Eval paper review

What Pairwise Evaluation Misses: Reward Scores across Prompts

Mateusz Modrzejewski, Tomasz Radzikowski

Published

Oct 22, 2026

Citations

0

Trust level

Moderate

Usefulness score

65/100 (Medium)

Extraction confidence

70% (Moderate)

Derived from extracted protocol signals and abstract evidence.

Rater population

Not reported

Signals refreshed

Oct 22, 2026

Open arXiv Find implementation Shortlist (0) RSS feed

Tags

Automatic Metrics General Pairwise Preference

Should you rely on this paper?

This paper has useful evaluation signal, but protocol completeness is partial; pair it with related papers before deciding implementation strategy.

Use this for comparison and orientation, not as your only source.

Best use

Secondary protocol comparison source

Use if you need

A secondary eval reference to pair with stronger protocol papers.

What to verify

Validate the evaluation procedure and quality controls in the full paper before operational use.

Main weakness

No major weakness surfaced.

Human feedback signal
Detected
From extracted signals
Evaluation signal
Detected
Eval setup described
Usefulness for eval research
65/100
Moderate-confidence candidate

Useful as a secondary reference; validate protocol details against neighboring papers.

If you are doing eval pipeline work, start here

Human Eval Papers LLM-as-Judge Papers Pairwise Preference Papers Tool-Use Evaluation Search

Abstract

Reward models score responses for preference optimization, filtering and ranking, and pooling scores across prompts needs them to be comparable. Yet pairwise evaluation compares two responses to the same prompt, so any score component shared by a prompt's responses cancels and goes untested. We study this component by comparing between-prompt variation in model scores with the same variation in human ratings of the same responses. We propose a human-referenced audit, built from a between-prompt variance share, a sibling-rating regression and prompt substitutions, and apply it to eighteen released scorers and six matched training conditions. We find that scores cluster by prompt more than the ratings do: on HelpSteer2, six reward models have between-prompt shares of 0.34 to 0.64 against a human share near zero. Replacing the original prompt with a fixed instruction keeps substantial clustering but weakens agreement with prompt-level human ratings, so clustering alone does not show agreement with humans. Part of that variation is useful: subtracting prompt means lowers the helpfulness of globally selected responses, and scores stay sensitive to a wrong prompt even when training shuffles prompt means. We propose that reward evaluation check between-prompt variation, sensitivity to a wrong prompt and usefulness for selection separately.

What we could verify

These are the protocol signals we could actually recover from the available paper metadata. Use them to decide whether this paper is worth deeper reading.

Human Feedback Types

strong

Pairwise Preference

Directly usable for protocol triage.

"Reward models score responses for preference optimization, filtering and ranking, and pooling scores across prompts needs them to be comparable."

Evaluation Modes

strong

Automatic Metrics

Includes extracted eval setup.

"Reward models score responses for preference optimization, filtering and ranking, and pooling scores across prompts needs them to be comparable."

Quality Controls

missing

Not reported

No explicit QC controls found.

"Reward models score responses for preference optimization, filtering and ranking, and pooling scores across prompts needs them to be comparable."

Benchmarks / Datasets

missing

Not extracted

No benchmark anchors detected.

"Reward models score responses for preference optimization, filtering and ranking, and pooling scores across prompts needs them to be comparable."

Reported Metrics

strong

Helpfulness

Useful for evaluation criteria comparison.

"Part of that variation is useful: subtracting prompt means lowers the helpfulness of globally selected responses, and scores stay sensitive to a wrong prompt even when training shuffles prompt means."

Benchmarks and datasets

No benchmark or dataset names were extracted from the available abstract.

Reported metrics

helpfulness
Human feedback details
Uses human feedback
Yes
Feedback types
Pairwise Preference
Rater population
Not reported
Unit of annotation
Pairwise
Expertise required
General
Evaluation details
Evaluation modes
Automatic Metrics
Agentic eval
None
Quality controls
Not reported
Evidence quality
Moderate
Use this page as
Secondary protocol comparison source

Research brief

Metadata summary

Reward models score responses for preference optimization, filtering and ranking, and pooling scores across prompts needs them to be comparable.

Based on abstract + metadata only. Check the source paper before making high-confidence protocol decisions.

Key takeaways

  • Reward models score responses for preference optimization, filtering and ranking, and pooling scores across prompts needs them to be comparable.
  • Yet pairwise evaluation compares two responses to the same prompt, so any score component shared by a prompt's responses cancels and goes untested.
  • We study this component by comparing between-prompt variation in model scores with the same variation in human ratings of the same responses.

Researcher actions

  • Compare this paper against nearby papers in the same arXiv category before using it for protocol decisions.
  • Check the full text for explicit evaluation design choices (raters, protocol, and metrics).
  • Use related-paper links to find stronger protocol-specific references.

Caveats

  • Generated from abstract + metadata only; no PDF parsing.
  • Signals below are heuristic and may miss details reported outside the abstract.

Recommended queries

Pairwise preference evaluation Agreement / Kappa

Contribution summary

  • Reward models score responses for preference optimization, filtering and ranking, and pooling scores across prompts needs them to be comparable.
  • We propose a human-referenced audit, built from a between-prompt variance share, a sibling-rating regression and prompt substitutions, and apply it to eighteen released scorers and six matched training conditions.
  • We propose that reward evaluation check between-prompt variation, sensitivity to a wrong prompt and usefulness for selection separately.

Why it matters for eval

  • We propose a human-referenced audit, built from a between-prompt variance share, a sibling-rating regression and prompt substitutions, and apply it to eighteen released scorers and six matched training conditions.
  • We propose that reward evaluation check between-prompt variation, sensitivity to a wrong prompt and usefulness for selection separately.

Researcher checklist

  • Human feedback protocol is explicit

    Detected: Pairwise Preference

  • Evaluation mode is explicit

    Detected: Automatic Metrics

  • Quality control reporting appears

    No calibration/adjudication/IAA control explicitly detected.

  • Benchmark or dataset anchors are present

    No benchmark/dataset anchor extracted from abstract.

  • Metric reporting is present

    Detected: helpfulness