Counterexample-Guided Requirements Discovery

Read full text Summarise with VIRAT
Authors: Anonymous, Authors
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

C-GRiD: Counterexample-Guided Requirements Discovery — Replication Package This repository contains the full replication package for the paper "Neuro-Symbolic Requirements Elicitation: Utilizing Formal Verification Counterexamples as Contextual Prompts", submitted to ASE 2026 (41st IEEE/ACM International Conference on Automated Software Engineering). C-GRiD is a neuro-symbolic framework for automated requirements elicitation. It combines large language models with symbolic model checking in an iterative feedback loop: an LLM synthesizes an initial formal specification, a TLA+ model checker (TLC) produces counterexamples that expose missing constraints, and those counterexamples are translated into targeted natural-language questions posed to a stakeholder oracle. The process repeats until the specification converges or no further constraints are recovered. The package includes the complete implementation of C-GRiD and three baselines (B1: one-shot LLM, B1-Strengthened: few-shot + chain-of-thought, B3: CEGIS template synthesis), a benchmark suite of 10 formally specified systems across 6 domains (banking, e-commerce, healthcare, concurrency, distributed systems, embedded control), a specification lobotomy evaluation protocol that systematically removes k constraints to simulate incomplete requirements, and all experiment scripts needed to reproduce the results reported in the paper. Key results: C-GRiD (perfect oracle) achieves 79.2% recall and 90.0% precision, outperforming the best baseline (B1) by 65 percentage points in recall (p < .0001, Cohen's d = 2.64). Under a noisy oracle with 25% error rate, recall degrades by only 8.3 percentage points to 70.8%. Software: Python 3.12 · Groq API (Llama-3.3-70b-versatile) · TLC model checker for TLA+

Fields of Study & Subjects

Formal Methods in Verification Advanced Software Engineering Methodologies Software Engineering Techniques and Practices

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN