Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

4D-ARE: Bridging the Attribution Gap in LLM Agent Requirements Engineering

Bo Yu, Lei Zhao

TL;DR

4D-ARE addresses a practical gap in LLM agent design: decision-makers demand attribution, not only answers. The authors propose a four-dimensional attribution model (Results, Process, Support, Long-term) and a five-layer specification stack that translates domain knowledge into deployable system prompts, enabling attribution-complete responses when paired with runtime reasoning like ReAct. The work provides a formal framework, concrete artifacts (YAML configurations), and an industrial pilot in financial services demonstrating improved attribution completeness and boundary adherence. While preliminary and domain-specific, the methodology offers a structured path to front-load domain knowledge, reduce prompt-ware fragility, and guide responsible deployment, with clear directions for future empirical validation and automation.

Abstract

We deployed an LLM agent with ReAct reasoning and full data access. It executed flawlessly, yet when asked "Why is completion rate 80%?", it returned metrics instead of causal explanation. The agent knew how to reason but we had not specified what to reason about. This reflects a gap: runtime reasoning frameworks (ReAct, Chain-of-Thought) have transformed LLM agents, but design-time specification--determining what domain knowledge agents need--remains under-explored. We propose 4D-ARE (4-Dimensional Attribution-Driven Agent Requirements Engineering), a preliminary methodology for specifying attribution-driven agents. The core insight: decision-makers seek attribution, not answers. Attribution concerns organize into four dimensions (Results -> Process -> Support -> Long-term), motivated by Pearl's causal hierarchy. The framework operationalizes through five layers producing artifacts that compile directly to system prompts. We demonstrate the methodology through an industrial pilot deployment in financial services. 4D-ARE addresses what agents should reason about, complementing runtime frameworks that address how. We hypothesize systematic specification amplifies the power of these foundational advances. This paper presents a methodological proposal with preliminary industrial validation; rigorous empirical evaluation is planned for future work.

4D-ARE: Bridging the Attribution Gap in LLM Agent Requirements Engineering

TL;DR

4D-ARE addresses a practical gap in LLM agent design: decision-makers demand attribution, not only answers. The authors propose a four-dimensional attribution model (Results, Process, Support, Long-term) and a five-layer specification stack that translates domain knowledge into deployable system prompts, enabling attribution-complete responses when paired with runtime reasoning like ReAct. The work provides a formal framework, concrete artifacts (YAML configurations), and an industrial pilot in financial services demonstrating improved attribution completeness and boundary adherence. While preliminary and domain-specific, the methodology offers a structured path to front-load domain knowledge, reduce prompt-ware fragility, and guide responsible deployment, with clear directions for future empirical validation and automation.

Abstract

We deployed an LLM agent with ReAct reasoning and full data access. It executed flawlessly, yet when asked "Why is completion rate 80%?", it returned metrics instead of causal explanation. The agent knew how to reason but we had not specified what to reason about. This reflects a gap: runtime reasoning frameworks (ReAct, Chain-of-Thought) have transformed LLM agents, but design-time specification--determining what domain knowledge agents need--remains under-explored. We propose 4D-ARE (4-Dimensional Attribution-Driven Agent Requirements Engineering), a preliminary methodology for specifying attribution-driven agents. The core insight: decision-makers seek attribution, not answers. Attribution concerns organize into four dimensions (Results -> Process -> Support -> Long-term), motivated by Pearl's causal hierarchy. The framework operationalizes through five layers producing artifacts that compile directly to system prompts. We demonstrate the methodology through an industrial pilot deployment in financial services. 4D-ARE addresses what agents should reason about, complementing runtime frameworks that address how. We hypothesize systematic specification amplifies the power of these foundational advances. This paper presents a methodological proposal with preliminary industrial validation; rigorous empirical evaluation is planned for future work.
Paper Structure (88 sections, 2 theorems, 6 equations, 3 figures, 11 tables)

This paper contains 88 sections, 2 theorems, 6 equations, 3 figures, 11 tables.

Table of Contents

  1. Introduction
  2. A Failure That Worked Perfectly
  3. The Attribution Gap
  4. The Complementary Challenge: Design-Time Specification
  5. 4D-ARE: Attribution-Driven Agent Specification
  6. What We Contribute
  7. Contributions and Epistemic Status
  8. Scope and Organization
  9. Related Work
  10. Runtime Reasoning: Foundational Advances
  11. Requirements Engineering: Right Idea, Wrong Era
  12. Knowledge Engineering: The Tacit Problem
  13. Prompt Engineering: Technique Without Methodology
  14. Specification-Driven Development: Process Without Content
  15. Causal Attribution in AI Systems
  16. ...and 73 more sections

Key Result

Proposition 3.2

In organizational decision-making, outcomes $Y$ are generated through a four-stage causal process: where each stage causally influences the next, and earlier stages have slower rates of change.

Figures (3)

  • Figure 1: The Four Dimensions. When Results show a gap, attribution traces backward through Process, Support, and Long-term to identify causes.
  • Figure 2: The Attribution Chain. Each "WHY?" traces backward through the causal hierarchy, from observable outcomes to environmental context.
  • Figure 3: The Five-Layer Specification Architecture. Each layer produces artifacts that compile to system prompt components.

Theorems & Definitions (6)

  • Definition 3.1: Attribution Gap
  • Proposition 3.2: Organizational Causal Chain
  • Proposition 3.3: Dimensional Organization---Hypothesis
  • Definition 3.4: 4D Attribution Model
  • Definition 3.5: Attribution Dependency
  • Definition 3.6: Attribution Trace