Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-Context Review: Improving LLM Output Quality by Separating Production and Review Sessions

Tae-Eun Song

Abstract

Large language models struggle to catch errors in their own outputs when the review happens in the same session that produced them. This paper introduces Cross-Context Review (CCR), a straightforward method where the review is conducted in a fresh session with no access to the production conversation history. We ran a controlled experiment: 30 artifacts (code, technical documents, presentation scripts) with 150 injected errors, tested under four review conditions -- same-session Self-Review (SR), repeated Self-Review (SR2), context-aware Subagent Review (SA), and Cross-Context Review (CCR). Over 360 reviews, CCR reached an F1 of 28.6%, outperforming SR (24.6%, p=0.008, d=0.52), SR2 (21.7%, p<0.001, d=0.72), and SA (23.8%, p=0.004, d=0.57). The SR2 result matters most for interpretation: reviewing twice in the same session did not beat reviewing once (p=0.11), which rules out repetition as an explanation for CCR's advantage. The benefit comes from context separation itself. CCR works with any model, needs no infrastructure, and costs only one extra session.

Cross-Context Review: Improving LLM Output Quality by Separating Production and Review Sessions

Abstract

Large language models struggle to catch errors in their own outputs when the review happens in the same session that produced them. This paper introduces Cross-Context Review (CCR), a straightforward method where the review is conducted in a fresh session with no access to the production conversation history. We ran a controlled experiment: 30 artifacts (code, technical documents, presentation scripts) with 150 injected errors, tested under four review conditions -- same-session Self-Review (SR), repeated Self-Review (SR2), context-aware Subagent Review (SA), and Cross-Context Review (CCR). Over 360 reviews, CCR reached an F1 of 28.6%, outperforming SR (24.6%, p=0.008, d=0.52), SR2 (21.7%, p<0.001, d=0.72), and SA (23.8%, p=0.004, d=0.57). The SR2 result matters most for interpretation: reviewing twice in the same session did not beat reviewing once (p=0.11), which rules out repetition as an explanation for CCR's advantage. The benefit comes from context separation itself. CCR works with any model, needs no infrastructure, and costs only one extra session.
Paper Structure (39 sections, 2 figures, 8 tables)

This paper contains 39 sections, 2 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Self-Correction in LLMs
  4. Anchoring Bias and Sycophancy
  5. Context Degradation
  6. Multi-Agent and Role-Based Approaches
  7. Method: Cross-Context Review
  8. Definition
  9. Protocol
  10. Theoretical Motivation
  11. Distinction from Incubation Effects
  12. Experimental Setup
  13. Artifacts and Error Injection
  14. Review Conditions
  15. Execution
  16. ...and 24 more sections

Figures (2)

  • Figure 1: F1 scores by condition and artifact category (3-run average). CCR consistently outperforms SR, SA, and SR2 across code, document, and script categories.
  • Figure 2: Detection rate by error severity. CCR's advantage over baselines grows with severity: +11pp for Critical errors, narrowing to roughly zero for Minor errors, indicating that context separation is most valuable for high-impact errors.