VIGÍA 4+1: Mitigating LLM Sycophancy and Normative Bias in Requirements Engineering via Deterministic Agentic Orchestration

Read full text Summarise with VIRAT
Authors: Jordan Cruz-Tarazona
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This paper presents the VIGÍA 4+1 Protocol, a deterministic agentic orchestration framework designed to mitigate Large Language Model (LLM) sycophancy and normative bias in requirements engineering. By decoupling the LLM from the application state through a Finite State Machine (FSM) implemented in .NET 10, we enforce Domain-Driven Design (DDD) constraints. The framework introduces a novel "Cognitive Try/Catch" layer utilizing a Negative Acknowledgment (NACK) mechanism and a Structured State Vector (SSV) for context persistence. Empirical simulations demonstrate that VIGÍA 4+1 reduces the Sycophancy Rate from 84.0% to a residual 2.0%, achieving a 73.8% reduction in token consumption compared to standard Chain-of-Thought (CoT) implementations. The results validate the feasibility of deploying high-integrity autonomous software architectures on consumer-grade high-end hardware.

Fields of Study & Subjects

Software Engineering Techniques and Practices Advanced Software Engineering Methodologies Ethics and Social Impacts of AI

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
24 March 2026
Language
EN
Document type
Preprint