Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 

Repository files navigation

AskRAG-Bench

Anonymous research manuscript for evaluating the pragmatic decision that precedes answer generation in retrieval-augmented generation.

Paper

Beyond Answer or Abstain: Human-Validated RAG Action Selection

Abstract

Most retrieval-augmented generation evaluations begin after a system has already chosen to answer. AskRAG-Bench evaluates the earlier, pragmatic choice among four mutually exclusive actions: ANSWER, CLARIFY, ABSTAIN, or QUALIFY.

The resource separates a source-grouped construction pool from target-blind human validation and a hash-reconstructable, high-consensus evaluation core. It also provides controlled same-record contrasts that preserve source identity while changing an action-relevant question or evidence condition. The benchmark supports reproducible analysis of when a RAG system should answer, ask for clarification, abstain, or provide a qualified response.

Repository Contents

  • askrag-bench-anonymous.pdf: Anonymous manuscript.
  • README.md: Publication overview.

Anonymous Release

This repository is maintained as an anonymous research release. It intentionally contains no identifying author, affiliation, or contact information.

Citation

Citation metadata will be provided after the anonymous review process is complete.

About

Anonymous paper release for AskRAG-Bench, a human-validated benchmark for RAG action selection.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors